Crossing the divide: a longitudinal study of effective treatments for people with autism and attention deficit hyperactivity disorder across the lifespan

Methods used that were common to both ADHD and ASD

We conducted an observational study based on face-to-face interviews and self-completion questionnaires with young people with ADHD and an ASD and their parents (usually mothers) at yearly intervals. The ADHD study was a 3-year prospective study; the ASD group was followed for 2 years [this difference in follow-up time arose owing to delay from Research Ethics Committees (RECs), resource constraints and time limits for the programme set by the National Institute for Health Research (NIHR); for further details, please see Obstacles and solutions]. The study collected information on health service use, needs and social/demographic and health information relating to participants and their family member at the time of the interview. For the ADHD study, self-completion questionnaires were also administered to adolescents/young adults to obtain information regarding drug and alcohol use and problems with the police. Differences in the study design between the ADHD and ASD groups are highlighted in the sections that follow.

Study site

Data collection largely took place in participants’ homes or at the Institute of Psychiatry (IoP). Most participants were from Greater London; the remainder were spread throughout England, extending from Cornwall to Lincolnshire in the north-east.

Sample

Our study included 183 families consisting of young people aged 14–24 years (n = 82 with ADHD and n = 101 with ASD) and their parents. More detailed information on sample sizes is provided in Tables 1 and 4. Families were recruited through their children’s childhood clinical diagnoses of ADHD or an ASD from Child and Adolescent Mental Health Services (CAMHS) and adult clinics, charities and research databases that form part of our clinical research networks. Clinical diagnosis of autism was confirmed in all cases using the Autism Diagnostic Interview – Revised27 (ADI-R) and ADHD (combined type) was defined using the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV), criteria. As participants in the ADHD group were originally recruited for the International Multi-Centre ADHD Genetics (IMAGE) programme, they were excluded if they had been diagnosed with autism, epilepsy, general learning difficulties, brain disorders and any genetic or medical disorder associated with externalising behaviours that might mimic ADHD based on both history and clinical assessment.

Recruitment for the ADHD sample began on 1 April 2009 and ended on 23 February 2011, for the ASD sample it began on 7 June 2010 and ended on 20 October 2011. Both studies received REC approval from the London – Camberwell St. Giles REC (previously known as the Joint South London and Maudsley and the IoP REC, South East London REC 4). The ASD REC reference is 04/H0807/71 and the ADHD REC reference is 08/H0807/68.

Instruments common to both ADHD and ASD studies

The instruments common to both studies are described below. Those used only in either the ADHD or the ASD elements of the project are described in the relevant disorder-specific section (see Chapters 2 and 3).

The level of need of the young person was captured using the Camberwell Assessment of Need for Adults with Developmental and Intellectual Disabilities (CANDID). 28 CANDID assesses needs across 25 domains, including social, physical, mental health, self-care and practical needs. For each domain, the young person and the parent were asked whether or not there was a significant need (coded as 1). If so, further questions were asked to ascertain if sufficient formal and/or informal support was being received. If adequate support was received, the need was classified as ‘met’ (regardless of whether or not it was met by formal or informal carers); if insufficient support was received the need was classified as ‘unmet’. Both met and unmet need scores were calculated by summing the number of domains where each was recorded. A total needs score (range 0–25) was calculated by summing the met and unmet need scores. 28

We adapted a brief series of questions on drug use from the Office for National Statistics (ONS) survey Mental Health of Children and Young People in Great Britain, 2004 Report. 29 Young people were asked to self-rate the frequency and nature of drug use from a range of drugs including cannabis, cocaine and heroin. For each individual drug, a question was asked regarding whether or not the participant had ever used this drug, even if just once. If the young person answered yes to this first opening question, two more questions were asked: (1) at what age the young person had first used this drug and (2) whether or not the young person had used this drug in the past month.

Alcohol use was assessed using the Alcohol Use Disorders Identification Test Consumption (AUDIT-C), a brief and validated three-question screen that can help identify hazardous and harmful drinking. 30 The AUDIT-C is an abbreviated version of the 10-question Alcohol Use Disorders Identification Test (AUDIT); the only screening instrument of hazardous alcohol use specifically designed for international use that is consistent with International Classification of Diseases, Tenth Edition (ICD-10), definitions of alcohol dependence and harmful alcohol use. Higher scores indicate greater likelihood of hazardous and harmful drinking and may reflect greater severity of alcohol problems and dependence, as well as a greater need for more intensive treatment.

Problems with police were examined through a series of questions based on those in the background information questionnaire used in the adult ADHD service at the Maudsley Hospital (adapted for this study). All participants were asked whether or not they had been in trouble with the police in the past 12 months. Those who answered yes were asked a brief series of questions regarding the nature and frequency of these problems (e.g. frequency of custodial sentences, times spent in a prison cell, appearances in court).

Caregiver burden was measured using the 12-item Zarit Burden Interview. 31 This captures the psychological and social impact of caring and asks respondents to rate the extent to which they agree or disagree with statements regarding their feelings about the results of caring on a five-point scale, with possible responses ranging from 0 (‘never’) to 4 (‘nearly always’). A total score (0–48) was calculated by summing the scores for each question. 31

We used a modified version of the Client Service Receipt Inventory (CSRI) to capture service use. 32 In particular, participants were asked to state whether or not they were under the formal care of child or adult health services, with those answering yes recorded as currently being in touch with services (coded as 1). The CSRI was completed jointly by the parent and young person (where they took part).

Obstacles and solutions

Our main obstacles related to five main factors:

1. delays with REC approval and the need to submit separate ethics applications for both groups

2. reaching young people

3. attrition, a problem common to all longitudinal studies but of particular concern for the ADHD group

4. the online completion of the Development and Well-Being Assessment (DAWBA) by parents of the ASD group

5. the lack of detail on prescribing patterns in an epidemiological sample that could be accessed in the time scale of this programme.

First, approval of the ASD research project for the purposes of the Mental Capacity Act 200533 led to significant delays due to the higher prevalence of LDs. As there is currently no standardised or accepted protocol for assessing whether or not a person lacks capacity to consent to take part in research it was necessary to develop a protocol for assessing capacity for this research project. Thus, a mental capacity assessment was adapted from one used in an earlier project under the same NIHR programme grant (REC 09/H0807/72). As this took some time to develop and to be approved by the local REC, the ASD study began 1 year after the ADHD study. This meant that ASD participants were followed for only 2 years (instead of 3 years, as for ADHD participants).

Second, our original design for the ADHD sample was based on contacting parents (as participants were contacted as children we only had contact details for parents). However, we soon amended the research design to be more inclusive in approach to young people, rather than approaching only via parents. To this effect, initial letters of invitation were sent both to parents and the adolescents/young adults at their parents’ homes.

Third, in order to reduce attrition (and non-response), especially among those with ADHD, we implemented a variety of measures. As participants were initially recruited through the IMAGE sample, the newsletter was revived and sent on an occasional basis to participants with summaries of the research and key findings (this has been led by Professor Asherson and his colleagues).

Fourth, we encountered difficulties in getting parents to complete the online DAWBA for their children with an ASD. This is because this online assessment is relatively time-consuming. We adopted several solutions to address this issue. Our researchers repeatedly contacted those with incomplete DAWBAs, reminding them of the need to provide this information. Furthermore, our researchers offered to assist parents in completing the forms. Finally, we offered parents a gift voucher of £20 in order to compensate them better for time taken to complete the DAWBAs.

Furthermore, we were concerned by the lack of accurate prescribing data and an epidemiological sample that we could access – as this raises issues about the generalisability of our results. Hence, we developed a collaboration with Professor Ian Wong (who was then at the School of Pharmacy) to examine medication use in a nationally representative data set based on general practitioner (GP) records: The Health Improvement Network (THIN). The data for young people with ADHD have already been published and so we examined psychotropic drug prescribing and neuropsychiatric comorbidities among 0- to 24-year-olds with ASD between 1992 and 2008 in a nationally representative primary care database. Further details outlining the rationale for this component and our results are provided later in this report (see Chapter 20).

Instruments specific to ADHD study

We used the DIVA,34 a diagnostic interview measure for adults with ADHD, recommended by the European ADHD Consensus Group,1 to measure current inattentive and hyperactive/impulsive symptoms (that is in the past 6 months). Using the DIVA, three subtypes of ADHD can be identified: predominantly inattentive, predominantly hyperactive–impulsive and combined inattentive and hyperactive–impulsive. 34

Given that there is currently no measure of ADHD symptoms and impairments that is validated for use across the whole age range used in this study (i.e. 14–21 years), the DIVA was chosen for several reasons. First, despite being a relatively new measure, it was used in preference to existing published diagnostic interviews such as the Conners’ Adult ADHD Diagnostic Interview for DSM-IV (CAADID),35 because it is briefer, permitted greater freedom in responses and is used increasingly throughout Europe. Second, compared with the CAADID, which has items that are similar to the DIVA, the DIVA is also publicly available. Third, the items in the DIVA were considered to be more realistic for the diagnostic assessment of ADHD in adults by the European ADHD Consensus Group. 1 Finally, it was judged important to keep the outcome data the same across the whole sample, rather than choose one measure for adolescents and one for adults.

We also used the BAARS-IV (informant version),36 a standardised and widely used rating scale for the assessment, diagnosis and monitoring of treatment of ADHD in adults. It assesses 18 symptom items from the diagnostic criteria for ADHD in the DSM-IV and 10 items relating to impairment across different areas of functioning. The informant is asked to rate the frequency and severity of impairments with scores on a four-point scale ranging from 0 to 3 capturing the severity and frequency of the behaviours representing ‘0, not at all or rarely’; ‘1, sometimes’; ‘2, often’; and ‘3, very often’. Although both parent and young person ratings of impairments were obtained only the parent ratings are used here.

Psychiatric-associated conditions were assessed using the CIS-R. 37 It is a standardised, valid and reliable structured diagnostic instrument used for rating psychiatric symptoms across 14 domains (e.g. anxiety, depression), and is widely used in both clinical and the general populations. 37 Scores on each section can range from 0 to 4 (0–5 for the section on depressive ideas), with a total score ranging from 0 to 57. A total score of ≥ 12 is regarded as clinically significant. 37

Sample characteristics

We included 82 individuals with ADHD. The average age of ADHD group was 17.5 years (see Table 1), and most were male (89%), still living at home (87%), unmarried (96%) and still in education (66%). Half reported having been excluded from school. Thirty-seven per cent of parents reported having achieved an educational level higher than A level (Advanced level).

The majority (73%) still met DSM-IV criteria for ADHD. Overall, there was no significant difference between the younger (14–17 years) and the older (18–21 years) age groups in the percentage who met DSM-IV diagnostic criteria for ADHD (χ² = 0.108; p = 0.742). However, a higher percentage of the younger group met criteria for combined ADHD (41% vs. 24%), whereas the predominantly inattentive subtype was more common in older individuals (44% vs. 24%) (Figure 1).

FIGURE 1.

Distribution of ADHD diagnoses by age according to DSM-IV criteria. HYPER, hyperactive–impulsive; INN, inattentive.

In addition, nearly two-thirds scored above the cut-off score on the AUDIT, suggesting high levels of alcohol consumption. Forty-eight per cent had used illegal drugs at some point and close to one in four had used drugs in the past month. Finally, 25% reported having been in trouble with the police within the past year.

Impairments

In both age groups nearly all were impaired in at least one life activity (Figure 2). Overall, around two-thirds of parents reported significant impairments in their child’s management of daily responsibilities (68%), home life (67%), educational activities (62%) and money (61%); half of parents reported impairments among their children related to work (49%). There were no significant differences in percentages of impairments reported between the two age groups apart from in social interactions – where parents reported more impairments among those in the 14–17 years age group in comparison with those aged 18–21 years [χ² = 10.48, degrees of freedom (df) = 3; p = 0.013]. In summary, impairments were not restricted to younger individuals with ADHD – both groups demonstrated marked impairments.

FIGURE 2.

Distribution of impairments in daily activities by age.

Associated psychiatric symptoms

Twenty-seven per cent scored above the cut-off score (> 12) on the CIS-R, indicating the likelihood of comorbid psychopathology. For example, over one-third of participants reported significant levels of fatigue and irritability and around one-quarter reported significant anxieties (Figure 3). There was no significant difference between the two age groups in the prevalence of associated mental health difficulties – including neurotic symptoms (t = –0.452, df = 79; p = 0.652).

FIGURE 3.

Distribution of neurotic symptoms by age according to CIS-R.

Needs

Parents reported that their child had a mean number of five needs (on average 2.58 of which were met and 2.48 of which were unmet; see Table 1). There was no significant difference between the younger and older age groups in the mean number of reported needs (5.25 vs. 4.89, t = 0.47; p = 0.64), met needs (2.78 vs. 2.41, t = 0.71; p = 0.48), or unmet needs (2.47 vs. 2.48, t = –0.01; p = 0.99). There was also no significant difference by age group in the percentages who reported needs within individual domains. The only exception being in the social contact domain, with parents of young people aged 14–17 years reporting more needs than parents of young people aged 18–21 years (χ² = 6.20, df = 1; p = 0.01) (Figure 4). Most of the help received towards meeting these needs came from family or friends rather than from formal services (Figure 5). Overall, formal services provided little or no help towards needs other than those linked to the participant’s physical health, whereas families or friends or young people provided most of the help towards meeting young people’s needs. For example, whereas 88% of participants received some help from family and friends towards their reported need in the safety to others domain, only 12% reported receiving some help from formal services towards this reducing this need.

FIGURE 4.

Distribution of needs by age.

FIGURE 5.

Distribution of help received from friends or family and services towards meeting needs.

Service use

Despite a high prevalence of associated mental health symptoms, 41% of the sample reported that they were no longer in contact with services. Thirty-three per cent of the sample were in contact with children’s services and 16% reported regularly attending an ADHD service without age boundaries (that is one specialising in the management of children, adolescents and adults with neurodevelopmental difficulties). Only eight (10%) of the participants were seen by adult services (Table 2).

Those more likely to be in contact with services were younger, reported more impairments and higher parent–carer burden (Table 3 shows the results of the logistic regression model used to analyse correlates of service use). The key outcome measure was whether or not the participant reported being currently seen by child, adult or tertiary services. The predictors captured predisposing (young person’s age), enabling (parental education) and need factors [ADHD symptoms, impairments, psychiatric comorbidities, parental caregiver burden and childhood conduct disorder (CD)]. Needs were not included in the model given their significant association with impairment (r = 0.51, df = 79; p < 0.01). Age was the only significant correlate of service use [odds ratio (OR) 0.62, 95% confidence interval (CI) 0.46 to 0.83; p = 0.001], with a 1-year increase in the young person’s age reducing the odds of being seen by services by 38%. None of the other factors considered was significantly correlated with service use. The proportion of variance explained by the models was 0.24 using Cox and Snell R² and 0.38 using Nagelkerke R², indicating that between 24% and 38% of the variance is explained by this model. In summary, we found that in ADHD age (and not level of impairment) determines the chance of being seen by services.

Health-care transition

As reported above, only 10% (n = 8) of participants reported having experienced a transfer to adult care. However, among this group, seven (≈85%) families reported not having received a written transition plan as recommended in clinical guidelines, and approximately 75% had not been spoken to by a professional about the move from child to adult services (and even the 25% who had said that this occurred reported that it been only a brief conversation). Around half of young people reported that they needed more help during the transition process to help them prepare for the move to adult services. Young people reported a need for more information to help them plan for the future (47%), someone to show them which services are available as they grow up (51%) and someone to explain the transition process to them (49%). In addition, around 70% of parents reported that their child needed more help in relation to information about future options (72%), having someone to talk to about transition (70%) and having someone to explain the transition process to them (69%). In addition, the majority (57%) of the those few young people and parents who had received some help in terms of transition reported that their transition had been poorly managed, with two-thirds reporting that they would have wanted more emotional and practical support.

Strengths and weaknesses

To our knowledge, this is the first analysis of needs and services use among those diagnosed with ADHD in childhood who are now at transition to adolescence and young adulthood. A key strength of our study includes the use of face-to-face interviews with both young people and their parents using reliable diagnostic and outcome measures. Moreover, our use of separate face-to-face interviews with young people and their parents enabled us to gain a more comprehensive picture of the wide range of needs among this group and to gauge the extent to which these needs are currently being met by services and family or friends. Another key strength of our study lies in our consideration of multiple potential factors that may correlate with service use among this group. This allowed us to better quantify the relative, and independent, contribution of factors that influence service use among this clinical group. Finally, our exploration of young people’s and their parents’ experiences of health-care transition, enabled us to identify key areas where improvements are necessary in health-care delivery. These are required from the perspective of both the affected person and their carers (e.g. a lack of access to sufficient information, problems with co-ordinating transition and lack of sufficient attention to the individual needs of young people and parents).

However, this was not an epidemiologically based sample and so our findings may not generalise. Nonetheless, we suggest that it is likely that the situation may be even worse in the general population, as our sample included individuals who had been in prior research studies of ADHD, and given that our sample had been in contact with CAMHS in childhood. Previous contact with such services may also have resulted in a bias towards presenting with higher levels of psychological problems at follow-up (e.g. in comparison with those in contact only with paediatric services in childhood). Nevertheless, rates of psychiatric comorbidity in our study are consistent with those reported in previous studies of ADHD. 44–48 Furthermore, as only those with a childhood intelligence quotient (IQ) of > 70 were included (with two exceptions), it is possible that our findings represent an underestimate of needs among those with intellectual disabilities. Likewise, as this study involved participants who were mostly males and from a Caucasian background it is not possible to say how representative the current findings are for women and those of ethnic minorities. Finally, due to our sample size we were likely underpowered to show significant effects of some of the predictor variables.

ASD-specific instruments

Current severity of ASD symptoms were captured using the Autism Quotient (AQ) – informant version. 50 It contains 50 items covering five different areas: social skills, attention switching, attention to detail, communication and imagination. For each question respondents report whether they ‘slightly agree’, ‘definitely agree’, ‘slightly disagree’ or ‘definitely disagree’ that a described behaviour is exhibited. Each question is scored 1 if the respondent reports the autistic symptom (by either ‘definitely’ or ‘slightly’ endorsing the symptom) and 0 if not. A total score (0–50) is calculated by summing the scores for each question. Scores over a cut-off point of 32 are indicative of an ASD. 50

Information on psychological co-occurring mental health symptoms was assessed using the DAWBA. 51 We did not use the CIS-R (as we did in the ADHD study) because there is no informant version of this instrument. The DAWBA assesses psychiatric symptoms and their impact across a number of domains through structured and open-ended questions, which cover the criteria required for a diagnosis according to both DSM-IV and ICD-10 criteria. The domains assessed were specific phobia, social phobia, panic, agoraphobia, OCD, generalised anxiety, depression, deliberate self-harm, impulsivity/hyperactivity and tics/Tourette syndrome. The DAWBA was completed online (www.dawba.net) by parents and young people when possible. Two trained raters (both psychiatrists) reviewed each case and produced consensus diagnoses based on the information provided by both parent and young person (where available). All conditions were rated according to ICD-10 criteria except ADHD, which was rated using DSM-IV criteria (as ADHD is not covered by ICD-10). An additional variable was derived from this measure stating whether or not the young person met criteria for one or more comorbid conditions [e.g. ADHD, Generalised Anxiety Disorder (GAD)].

Mental capacity issues

The ASD (but not ADHD) sample included individuals with intellectual disability (ID). Given the nature of the research, some of those who were eligible to participate lacked capacity to consent. We are aware that the baseline position is that capacity to consent is assumed unless there is reason to suppose otherwise. However, some neurological, psychological and/or behavioural markers may suggest lack of capacity for some things at some times. We also recognise that capacity is fluid in time and with respect to subject.

Indicators of potential lack of capacity include (but are not restricted to) the diagnosis of a LD, an IQ of < 70, significant deficits in adaptive functioning or other signs or symptoms of mental disorder that might compromise capacity to give informed consent. This was of particular concern for the ASD group as we did not exclude participants who had a LD (as was the case for the ADHD group, see above). All individuals with capacity gave informed consent.

Where there were doubts about the capacity of a young adult to give consent the study introduced an easy-read version of the information sheet and a capacity assessment was carried out by a trained researcher. The capacity assessment tool addressed the following questions, which are based on the principles outlined in Section 3 (1) of the Mental Capacity Act 2005:33

• Can the person understand the information relevant to the decision?

• Can the person retain that information?

• Can the person use or weigh that information as part of the process of making the decision?

• Can the person communicate his/her decision (whether by talking, using sign language or any other means)?

When it was established that the participant did not have capacity, but showed willingness to take part in the study, a relative or close friend (other than the carer) was approached to act as a personal consultee. The role of the consultee is to consider what the wishes and feelings of the participant would be if they had capacity, and if they judge that the person would want to take part, to consent on their behalf. We did our best to ensure that the consultee had no conflict of interest. If there were concerns about a conflict of interest then the parent was asked to contact the research team, at which stage a member of the research team spoke to them about the possibility of finding an alternative personal consultee.

The researchers who interviewed the ASD participants all received training from Dr Dene Robertson (consultant psychiatrist at the Mental Impairment and Evaluation Service, Bethlem Hospital) in assessing mental capacity to consent. In addition, they also undertook additional training; for example, they attended short courses on the Mental Health Act 200533 and Deprivation of Liberty Safeguards awareness training, as well as training on protection of vulnerable adults and the Protection of Children Act 1978. 52 When possible, the researchers conducted assessments in collaboration with a member of the health-care team, known to the potential participant, with experience in assessing mental capacity.

Service use (and needs) among those with an ASD in adolescence and young adulthood

As with the ADHD group, our focus so far has been on addressing the first research question, the correlates of service use. In addition, similar to ADHD, there is a lack of adult services for those with an ASD [and in particular among those with high-functioning autism (HFA) who have ongoing needs and functional impairments]. Appropriate and effective planning of care programmes requires knowledge of the needs of this population (and their carers).

Consequently, our aim was to investigate needs and correlates of service use (e.g. medical and demographic) among those diagnosed with ASD in adolescence and young adulthood (aged 14–24 years), to help identify how they are supported while they move from adolescence to young adulthood.

Similar to the analysis of the ADHD group, we used the Andersen–Newman Behavioural Model of Health Service (see Chapter 3, Results) to investigate the correlates of health service use among young people with an ASD.

ASD sample characteristics

Similar to the ADHD sample the young people were predominantly male (92%) and still living at home (87%). Seventy-three per cent met criteria on the DAWBA for an associated psychiatric condition and 13% had a clinically defined LD. Most of the total sample were still in education (77%) and had received a statement of special educational needs (70%); 40% had been excluded from school at some point in their education. Parental education was assessed and 60% reported a high level of achievement – that is they had received A-level qualifications or above.

Service and medication use

Fifty-six per cent of participants (Table 4) reported currently being in contact with child (33%), adult (22%) or tertiary services (1%). The services most commonly seen in the past 3 months were a GP (used by 43% of participants), psychiatrist (26%), key worker (22%) or psychologist (17%). Those who stated they were in contact with child, adult or tertiary services were significantly more likely than those who were not in contact with formal services to report having seen a psychiatrist (χ² = 3.66; p = 0.001) or a mental health key worker in the past 3 months (χ² = 7.92; p = 0.005).

Thirty-eight per cent of participants were currently taking at least one prescribed medication (Table 5). The most common prescribed medication was for ADHD (taken by 9% of participants), followed by antidepressants (8%) and sleep medication (8%). On average participants were prescribed 0.64 medications.

Psychiatric associated symptoms

Figure 6 shows the number of ASD participants who also met criteria on the DAWBA for common psychiatric conditions. Figure 6 shows, for those who met the criteria for each condition, the numbers who were in touch with services; had saw a psychiatrist, psychiatric nurse, psychologist or counsellor in the past 3 months; and whose parents reported that their children had previously received a clinical diagnosis for an associated psychiatric disorder. Results show that 43% of the sample met clinical research diagnostic criteria for ADHD (38/89), 33% for GAD (29/89), 19% for a specific phobia (17/89) and 20% for tic/Tourette syndrome (18/89). In addition, a large proportion met criteria for more than one mental health problem. For instance, nearly half (47%) of participants met criteria for two or more co-occurring mental health disorders. However, relatively few of the parents whose children met the criteria for a co-occurring mental health disorder reported having previously been told that their child had a clinical diagnosis of the disorder. The most common additional diagnoses that parents had been made aware of are ADHD [36% (14/39)], agoraphobia [11% (1/9)], OCD [14% (1/7)] and tic/Tourette syndrome [6% (1/18)]. Young people with an ASD who met criteria on the DAWBA for at least one psychiatric disorder were more likely than those who did not to be in touch with services (χ² = 10.67; p = 0.001). However, of those who currently met criteria for at least one additional psychiatric condition, one-third stated that they were not currently being seen by any services and over half had not seen any psychological or psychiatric services in the past 3 months (see Figure 6).

FIGURE 6.

Number of participants meeting criteria for one or more psychiatric conditions on the DAWBA (n = 89).

Need

Parents reported that their child had a mean number of 10 needs (on average six of which were met and three of which were unmet; see Table 4).

Figure 7 shows the needs most commonly reported by parents and the number of participants who were receiving help from family or services. Eighty-three per cent of parents reported that their child had a need relating to exploitation risk, 73% reported a need related to food, 72% reported a need related to money budgeting and 63% reported social needs. Few of the young people were receiving help for these needs from services; in contrast, almost all were receiving some help for them from friends or families.

FIGURE 7.

Met and unmet needs most frequently reported by participants and help received from family and services (n = 101).

As expected, younger people with an ASD with higher levels of need were more likely to be in contact with services than those who were not [mean 10.54, standard deviation (SD) 3.60, in contact vs. mean 8.14, SD 3.57, not in contact, t = –3.13; p = 0.002].

Correlates of service use

Table 6 shows the results of the logistic regression model used to analyse correlates of service use. The dependent variable was whether or not the participant stated that they were currently being seen by child, adult or tertiary health-care services. The predictors captured predisposing (young person’s age), enabling (parental education) and need factors (that is the level of ASD symptomology, total level of need, whether or not they met criteria on the DAWBA for any associated psychiatric conditions and the presence of a LD). The only significant correlate of service use was whether or not the young person met criteria on the DAWBA for one or more psychiatric conditions. For instance, the odds of young people with an ASD and an associated psychiatric condition being in touch with services are four times higher than for those without such a condition. The proportion of variance explained by the model was 0.18 using Cox and Snell R² and 0.24 using Nagelkerke R², indicating that between 18% and 24% of the variance in service use is explained by this model.

Service use, diagnosis and treatment of comorbid psychiatric conditions

Table 7 examines the relationship between being in touch with services and diagnosis and treatment (in terms of reporting any psychiatric input in the past 3 months and receiving relevant medications), for the two most prevalent psychiatric symptoms (ADHD and GAD). We would expect that those who met criteria for a psychiatric condition and who were in touch with services would be more likely to report having received a clinical diagnosis and treatment. However, we found no such relationship (albeit with one exception). Among those with an ASD meeting DAWBA criteria for an anxiety disorder, 98% (43/44) reported that they had not been informed of (or diagnosed with) this additional problem by any service, 52% (23/44) had received no psychiatric input in the past 3 months and 86% (38/44) reported that they were receiving no relevant medication for anxiety. Only for those with an ASD who met criteria for ADHD was being in touch with services significantly associated with receiving psychiatric input in the past 3 months (see Table 7). However, even among this group, 65% (22/34) reported not having received a clinical diagnosis and 85% (29/34) were not being prescribed relevant medication.

Strengths and weaknesses

This study has a number of limitations. First, the sex balance is not representative of the ASD population as > 90% of our sample is male, compared with a sex ratio of approximately 4 : 1 typically found by others. Second, our sample included relatively high-functioning individuals compared with the ASD samples investigated in other studies. For example, approximately 15% of our sample was diagnosed with a LD compared with a prevalence of 56% in a population cohort of children. 56 Even so, there is increasing recognition that many individuals with an ASD do not have an ID, but still report high prevalence of functional impairments, psychological comorbid disorders and needs. 57,58 Finally, this initial report is from a cross-sectional study and is therefore able to examine associations of need and service use at only one point in time and so causality cannot be inferred.

Nevertheless, this study has significant strengths in comparison with earlier research. Our study is one of the few to focus on adolescents and young adults with an ASD – whereas most studies concentrate on children. 59,60 In addition, our diagnostic assessment, the ADI-R, is a widely recognised and well-validated informant-based (so-called ‘gold standard’) measure of ASD and so the sample contains ‘true’ cases.

Our study has shown a high prevalence of unrecognised psychiatric associated conditions and needs among adolescents and young adults with an ASD. Many of the young people in this clinical group have needs that are met not by services but largely by family and friends. Moreover, our findings show that only severity of psychiatric conditions is associated with service use, and not ASD symptomology. Even those young people with an ASD who met criteria for an additional psychiatric condition and were in touch with services are no more likely to report appropriate diagnosis and treatment. Yet, further research is required into the appropriate use, efficacy and long-term safety of antipsychotics and stimulants in the autistic population before moving towards standardised pharmacological treatments (see Chapter 19). Too few services are ‘autism focused’ and geared towards programmes to help those with an ASD to maintain independence and sustain employment. 61 Our data suggest that needs-led services are required which can meet the complexities of ASD, including the core symptoms (e.g. difficulties with reciprocal social interaction and communication, repetitive behaviours and interests), but which can also diagnose and treat severe associated conditions such as ADHD, anxiety and depression, and thus reduce both the challenges facing the young person and the burden they place on caregivers.

Sample characteristics

We included 89 individuals with ASD and 81 individuals with ADHD. Young people in the two groups were similar in age, sex (≈90% male), education (≈70% in education) and accommodation (≈90% lived at home) (Table 8). Caregiver sociodemographic characteristics in the two groups were also broadly similar (see Table 8). All of the caregivers were either a biological or an adoptive parent. Caregivers from the two samples showed similar sex and marital status distributions (72% of ASD caregivers were married vs. 78% of ADHD caregivers, χ² = 0.77; p = 0.38). In addition, around 35% of parents in both groups provided care for someone else aside from their child with ASD/ADHD (χ² = 0.28; p = 0.60). Caregivers from the two samples were also comparable in terms of age (ASD mean 49 years, SD 6 years vs. ADHD mean 48 years, SD 5 years, t = 1.65; p = 0.10). There were, however, significant differences in occupation of the two caregiver groups: parents of ASD were in higher occupational groups (managerial/professional and associate professional/administrative) (55% ASD vs. 42% ADHD, χ² = 12.54; p = 0.01).

Caregiver burden and study variables

Caregiver burden was high in both68,69 groups, but it was significantly higher in ASD than ADHD (ASD mean 22.66, SD 8.84 vs. ADHD mean 17.80, SD 9.18, t = 3.52; p = 0.001). Family background (parental education) was significantly different, with 62% of parents in the ASD group being in the highest educational category, compared with 35% in ADHD (χ² = 12.58; p < 0.001). Primary stressors (i.e. symptoms, LDs and psychiatric comorbidities) also showed significant differences between the two groups. The mean current severity of ASD as measured by AQ score for the ASD group was 36 (SD 6), with 73% of this group showing AQ scores above the suggested cut-off point of 32. The mean BAARS-IV current symptom score for the ADHD group was 10.63 (SD 4.73). Using a cut-off point of > 5 on the BAARS-IV current symptom score, 58% met threshold for inattentiveness, 42% for hyperactivity and 36% for combined ADHD. Fifteen per cent of the ASD group had been previously diagnosed with a LD, compared with none of the ADHD group. In terms of psychiatric comorbidities, the mean Strengths and Difficulties Questionnaire (SDQ) score for the ASD group was 21 (SD 7), with 70% scoring in the abnormal range (> 16). Fifty-five per cent of the autism group also scored in the abnormal range on the hyperactivity subscale of the SDQ (> 6). The mean CIS-R score for the ADHD group was 8 (SD 6), with 27% per cent of the ADHD group scoring ≥ 12, indicating the presence of a common mental disorder.

Need, our measure of primary appraisal, also showed significant differences between the two groups, with the needs of those with an ASD (as rated by caregivers) being significantly higher (ASD mean 9.90, SD 3.55, vs. ADHD mean 5.01, SD 3.12, t = 9.51; p < 0.001). We used service use to conceptualise resources. There was little difference in this indicator between the two groups; 58% of the ASD group and 56% of the ADHD group were currently being seen by health services.

Tables 9–11 present the sequential multiple regression models used to examine the correlates of caregiver burden. Separate models were produced for each condition, as well as a model combining both conditions. Variables representing family background characteristics (parental education) were first entered into the model, followed by those capturing primary stressors (symptoms, a LD for ASD and psychiatric comorbidities), primary appraisal (need) and resources (use of services). For both samples, caregiver-rated need was a significant correlate of burden even after other factors were controlled for (see Tables 9–11). In addition, once need was controlled for, severity of disorder symptoms was no longer a significant predictor of burden for ASD, but remained significant for ADHD (see Table 10, model 3). Conversely, although psychiatric comorbidities were a significant predictor of burden for ASD, they were not for ADHD. Given that approximately half of the ASD group also met the threshold on the SDQ hyperactivity scale, we also investigated whether or not ADHD symptomology was associated with burden for the ASD group. When the SDQ hyperactivity subscale was included as a separate variable in the model (along with a variable comprising the total score from the other three subscales), the hyperactivity subscale was not significantly associated with burden.

In the combined model need and comorbid psychopathology were significant predictors of burden, whereas having a diagnosis of either an ASD or ADHD was not (see Table 11). The models show 34% of the variance in caregiver burden among those caring for a young person with an ASD (see Table 9) in comparison with 23% in the ADHD group (see Table 10). Further models were run for both disorders using the young person’s assessment of their needs; however, no relationship was found with parental burden.

In order to further explore the relationship between need and burden the models above were run with unmet and met need instead of total need (not shown). The results showed that only unmet need was significantly associated with caregiver burden. The models using unmet need instead of total need accounted for 42% of the variance in burden for the ASD group and 26% for the ADHD group. To investigate these results further, additional bivariate correlations were calculated between individual CANDID items and caregiver burden. Table 12 shows the individual domains of unmet needs, which were significantly correlated with burden in the two groups. Unmet needs, related to depression/anxiety, inappropriate behaviour, exploitation risk, aggression/violence and daytime activities, were significantly correlated with burden for both samples. Specific to the ASD sample were significant correlations between burden and unmet needs relating to social relationships, mental health problems, safety of self and communication.

Strengths and weaknesses

This study has a number of limitations. First, we did not have a direct measure of maladaptive behaviour, although it is likely that the measures used (e.g. of need and psychiatric comorbidities) also captured behavioural difficulties. Second, our ASD sample appears to be higher functioning than the ASD populations that have been investigated in other studies. For example, approximately 15% of our sample was diagnosed with a LD, in comparison with a prevalence of 56% found by others in a population cohort of children. 7 In addition, there is increasing recognition that many individuals with ASD do not have ID. Even so, caregivers of these higher-functioning individuals still describe a very significant level of burden. Hence, public health policies should not assume that relatively higher-functioning ASD individuals and their caregivers do not also have significant (and unmet) needs. Third, the sex balance of the young people in both the ASD and ADHD samples was not representative of the respective populations. A male to female ratio of 3.3 : 1 was reported in a study of children with ASD62 and a male to female ratio of 1.6 : 1 was reported in a study of adults with ADHD. 63

By contrast, our samples had ratios of approximately 9 : 1 (male : female). In addition, almost all caregivers in the study were mothers. Given that studies in ASD and ADHD have found having a male child to be associated with higher levels of burden,68,69 and that levels of burden have been reported to be higher in mothers than in fathers,74 our results may overestimate the level of burden in this population. Fourth, although a measure of ADHD symptomology was included for the ASD sample (the hyperactivity/inattention subscale of the SDQ), no corresponding measure of ASD symptomology was included for the ADHD sample. Although the ADHD sample was screened for ASD in childhood, mild autism traits could still have been present; however, we were unable to consider the effect of these traits on caregiver burden. Fifth, this initial study was cross-sectional and is therefore able to examine associations of need and caregiver burden at only one point in time. Future longitudinal research is required to determine whether or not input from formal services significantly reduces parental burden. Sixth, there was some reduction in sample size due to missing data. As the CIS-R in the ADHD group was taken from the young person’s interview, analyses focused on those families with both a parent and young person interview (i.e. out of 86 paired interviews, 81 contained complete information). For the ASD group, analyses focused on the parent interviews as these contained the necessary measures. The sample size for this group dropped from 101 to 89, largely as a result of missing data on the SDQ. Seventh, differences in the relationship between comorbidities and caregiver burden in the two groups may be due to differences in the way that this information was collected: through self-reports in the ADHD group using the CIS-R and through informant reports in the ASD group using the SDQ. Finally, our initial analyses were cross-sectional – but we need to also examine change. As all data have now been collected, our future focus will be examining changes over time as outlined in research questions 2 and 3. Thus, we will investigate how needs and associated comorbidities have changed over time in this clinical group and examine the demographic and medical factors associated with this change. Our final analysis, which has not yet begun, will be to examine the main family carer’s role in meeting needs and how this has changed over time and whether or not this is related to changes in service use. We will assess correlates associated with this change and the consequences of changes in the family carer’s role for their own well-being.

Sample

Participants were recruited from the IMAGE study,85 a genetic study of ADHD carried out by researchers at the IoP between June 2003 and January 2006. Families were originally recruited to the IMAGE study by referral from child and adolescent clinics in south-east England on the basis that they had received a clinical diagnosis of combined-type ADHD (as defined in the DSM-IV) and had at least one surviving biological sibling aged 5–17 years. Both participants and their siblings were included in the IMAGE study if they were between the ages of 5 and 17 years, had an IQ of ≥ 70, were of European or Caucasian descent and had at least one biological parent willing to provide DNA samples. Participants were excluded from the IMAGE study if they had been diagnosed with autism, epilepsy, an IQ of < 70, brain disorders and/or any genetic or medical disorder associated with externalising behaviours that might mimic ADHD based on both history and clinical assessment.

To compare mental health outcomes with population norms, comparison data were acquired from the Adult Psychiatric Morbidity Study (APMS). 86 Data were included from 411 participants aged 16–23 years who did not meet criteria for ADHD.

Baseline measures

Follow-up procedure

Participants were recruited for the present study if they were aged between 14 and 24 years by 1 March 2009. Parents and young people were initially contacted by letter and followed up by telephone. Participants were interviewed either at our research offices or at the participant’s home. Of the original IMAGE sample of 204 families, 118 ADHD probands and their parents were successfully followed up. Mean follow-up duration was 6 (range 3–8) years.

Follow-up measures

Analyses

Descriptive statistics were produced for the sample at baseline and follow-up. To examine the persistence of ADHD, participants were classified into two groups using DIVA scores:

1. persistent ADHD, which included combined type (six or more inattentive and six or more hyperactive/impulsive symptoms), inattentive subtype (six or more inattentive symptoms and fewer than six hyperactive symptoms), and hyperactive–impulsive subtype (six or more hyperactive/impulsive symptoms and fewer than six inattentive symptoms)

2. subthreshold ADHD, which included partial remission (four or five inattentive or hyperactive/impulsive symptoms) and full remission (fewer than four inattentive and hyperactive symptoms).

To investigate the outcomes of persistent ADHD, comparisons were made between participants with persistent ADHD, subthreshold ADHD and the comparison sample from the APMS 2007 study. Comparison data from the APMS study were available for psychiatric comorbidity, drug and alcohol use. To explore correlates of these outcomes, multiple linear regression and logistic regression models were produced for each outcome, including both childhood and adult characteristics as predictors. All analyses were conducted in IBM SPSS Statistics for Windows version 20 (IBM Corporation, Armonk, NY, USA).

Sample characteristics

We included 118 individuals with ADHD. The average age of the ADHD group at follow-up was 17.5 years (Table 13); most were males (89%) and still living at home (88%). Fifty-three per cent had used illegal drugs in their lifetime, 20% had used drugs in the past month and 22% had been in trouble with the police within the past year. Over half of the sample met the cut-off score on the AUDIT, suggesting high levels of alcohol consumption. Twenty-seven per cent exceeded the cut-off score on the CIS-R, which is indicative of comorbid psychopathology (see Table 13). Over half of the sample were in full-time education and 12% were in full-time employment. As expected, the proportion of the sample in education decreases with age and the percentage in employment increases with age. However, 46% of those aged 19–23 years reported being unemployed (Table 14).

Persistence of ADHD at follow-up

At follow-up, 80% of the sample met full diagnostic criteria for DSM-IV ADHD: combined-type ADHD (37%), inattentive subtype (36%) and hyperactive–impulsive subtype (7%). Of the group who did not meet full DSM-IV criteria for ADHD, 15% were in partial remission (subthreshold ADHD), with only 5% in total remission (Tables 15 and 16). Age at follow-up was unrelated to the number of inattentive (r = –0.6; p = 0.54) or number of hyperactive symptoms (r = –0.12; p = 0.21) (Figure 8).

FIGURE 8.

Number of individuals by ADHD outcome in adolescents (aged 14–17 years) and young adults (aged 18–23 years).

Psychiatric comorbidity at follow-up

Significant differences were detected between persistent ADHD, subthreshold ADHD and normative groups in the rate of co-occurring mental health problems as assessed by scores on the CIS-R (F = 8.95; p < 0.001). Total scores on this measure were significantly higher in the persistent ADHD group than in the normative sample, indicating higher rates of comorbidity than in the non-ADHD control sample. No other group differences were present (Figure 9). The CIS-R contains some domains (memory and concentration) that overlap with ADHD diagnostic criteria. These analyses were therefore repeated with the memory and concentration items excluded, with no overall change in the finding for higher CIS-R scores in the ADHD group. We conducted further analyses by defining cases as above or below the recognised cut-off score of 32 on the CIS-R. There was a significant difference between the three groups [χ²₍₂₎ = 10.24; p = 0.06], with 17% of the control group above the threshold for other mental health disorders, compared with 40% of the ADHD persistent group. Thus, these results indicate that the individuals with persistent ADHD may be more at risk of psychiatric comorbidity compared with both the subthreshold ADHD group and the general population control sample. The subthreshold and control samples had similar rates of other mental health problems.

FIGURE 9.

Mean CIS-R scores for the normative, subthreshold and persistent ADHD groups. ***p < 0.001. A, persistent vs. normative control groups.

Next, we explored scores on the individual domains of the CIS-R to examine differences in the type of mental health symptoms between the three groups. The persistent ADHD group had significantly higher scores than both the subthreshold ADHD and normative group in the ‘fatigue’ and ‘concentration/memory’ domains and significantly higher scores than the normative group in the ‘anger’ domain. No differences were found in any other clinical domains reflecting a wider range of mental health problems (Figure 10).

FIGURE 10.

Mean CIS-R domain scores by ADHD group (n = 517). *p < 0.05, **p < 0.001, ***p < 0.001. A, persistent vs. normative control groups; B, persistent vs. subthreshold groups; C, subthreshold vs. normative control groups.

Drug use and police contact

The persistent ADHD group showed significantly higher rates of lifetime drug use [χ²₍₂₎ = 24.26; p < 0.001] relative to the normative group, but not the subthreshold ADHD group (Figure 11). Both the persistent and subthreshold ADHD groups showed significantly higher rates of police contact than the normative group, but no differences were found between the two ADHD groups [χ²₍₂₎ = 0.03; p = 8.72]. No differences were detected between any of the three groups in dangerous alcohol use [χ²₍₂₎ = 0.11; p = 0.95].

FIGURE 11.

Percentage of sample exceeding AUDIT cut-off point, reporting drug use and police contact. *p < 0.05, **p < 0.001, ***p < 0.001. A, persistent vs. normative controls groups; B, persistent vs. subthreshold groups; C, subthreshold vs. normative control groups. a, Data were unavailable in the normative control group for police contact over the past year, so data for lifetime police contact were used instead.

Emotional lability

Normative data were not available for measures of emotional lability or functional impairment, so comparisons were made only between the persistent and subthreshold ADHD groups. Significant differences between the two groups were observed on the emotional lability scale, with the persistent ADHD group scoring higher levels of emotional instability [t_(41.10) = –3.93; p < 0.001]. Functional impairment did not differ between the two groups [t₍₉₃₎ = –1.84; p < 0.07].

Predictors of outcome

To explore which child and adulthood characteristics were associated with outcomes, multiple regression and logistic regression models were produced for the main outcome variables. Predictor variables investigated included the number of adult and childhood ADHD symptoms (derived from DIVA and PACS interviews), age, IQ and sex. Emotional lability scores and a variable indicating whether or not they exceeded the cut-off point on the CIS-R were entered for the models for which those variables were not the dependent variable. Childhood medication status and the presence of a comorbid oppositional defiant disorder and CD in childhood were included in preliminary models but were not found to be significantly associated with any outcome measure, so were excluded from the final analysis.

Tables 17–20 list the findings from regression models investigating predictors of comorbid symptoms, behavioural problems and impairment scores. The main findings were as follows:

• Total functional impairments as rated by informants on the BAARS-IV impairments scale at follow-up was significantly associated with childhood inattentive symptoms and sex (see Table 17). None of the follow-up measures showed any association with impairment.

• Using a binary variable stating whether or not they met threshold scores for comorbid symptoms on the CIS-R (> 11) as the dependent variable, current level of hyperactive–impulsive symptoms, age, sex and emotional lability were all significant predictors (see Table 18). None of the childhood measures of ADHD showed any association with the adult comorbidity scores.

• Emotional lability was associated with current comorbidity scores. Childhood characteristics did not predict current emotional lability (see Table 19).

• Table 20 shows the findings from a linear regression model for alcohol use. Being of male sex was significantly associated with increased alcohol use, but none of the clinical characteristics was significant.

• The remaining models examined predictors of drug use and police contact. In both models, the only significant predictor was the child’s age; none of the measures of adult or childhood ADHD symptom severity predicted current drug use and police contact.

Strengths and weaknesses

In drawing conclusions from this study, we need to consider several limitations. First, different interview measures were used to assess ADHD in childhood and at follow-up. However, both of these measures are based on the DSM-IV and are recommended for use in the respective age bands as diagnostic instruments. Second, all of our ADHD sample was in touch with services at baseline and had combined-type ADHD. Hence, our ADHD sample is not representative of the broader spectrum of ADHD cases, although in most clinical settings around two-thirds of ADHD cases meet criteria for the combined subtype. The relatively homogeneous sample does, however, bring advantages by reducing clinical heterogeneity and focusing on what are widely considered the most impaired group of children with ADHD. Further work could usefully be completed in the UK to follow groups of children (and girls) with predominantly inattentive symptoms, where outcomes may potentially differ. Finally, due to our relatively small sample size, we were unable to examine separately participants with partially remitted or completely remitted ADHD. Indeed, this problem arose from the very high rate of persistence for the full diagnosis of ADHD that we saw in this sample at follow-up.

Study sites

Patients in the following local primary, forensic and adult mental health services were screened and assessed:

• primary care patients currently being treated for other common mental health disorders, mainly anxiety and depression

• patients in the anxiety disorders unit at the Bethlem Hospital, which is an inpatient facility for the treatment of severe anxiety disorders

• patients in the secure forensic psychiatry inpatient units at Broadmoor Hospital, River House and Bethlem Hospital, which are ‘prison hospitals’ caring for two main groups of patients: around two-thirds with a primary diagnosis of severe mental illness (usually schizophrenia) and one-third with a primary diagnosis of severe and dangerous PD

• patients in addiction inpatient facilities at the Bethlem Hospital and Maudsley Hospital and a community outpatient drug treatment service

• sentenced young male prisoners aged 18–25 years, held at Her Majesty’s Prison (HMP) Isis (south London)

• sentenced adult male prisoners aged 25–65 years, held at HMP Brixton (south London).

Sample

We screened 1604 individuals for ADHD/ASD: 253 individuals from primary care, 153 individuals from hospital anxiety services, 159 individuals from hospital forensic services and 226 individuals from hospital addiction services. Within the prison settings we investigated the rates of ADHD in 592 young people aged 18–25 years and 221 older prisoners aged 25–65 years. As the study of young prisoners was completed as part of a clinical trial of stimulants for ADHD and related behaviours (see Results and discussion), we were unable to complete Autism Diagnostic Observation Schedule (ADOS) interviews in addition to ADHD interviews, because of the already complex protocol and constraints on interview time as advised by the prison.

Diagnosing ADHD

Participants completed the DSM-IV 18-item self-report screening questionnaires for current and childhood ADHD symptoms using either the BAARS-IV44 or, in primary care, the Adult ADHD Self-Report Scale. 110,111 Positive screening for ADHD consisted of six or more ADHD symptoms during childhood, plus four or more symptoms during adulthood. Participants who screened positive in the first stage were invited to a diagnostic interview.

Research diagnosis of ADHD was established using the DIVA,71 conducted by trained research assistants (RAs). Consultant psychiatrists experienced in the diagnosis of ADHD further reviewed the cases. The DIVA interview systematically evaluates each of the DSM-IV symptom items for both current and childhood symptoms and asks additional questions to establish impairment from ADHD symptoms (impairment criteria), in two or more settings (pervasiveness criteria) and the age at onset of symptoms (age at onset criteria). The diagnostic algorithm for ADHD used the following criteria: (1) six or more ADHD symptoms in either domain from retrospective childhood rating scales, (2) four or more ADHD symptoms from either domain from current ADHD symptom rating scales and (3) research diagnosis of DSM-IV ADHD following application of diagnostic interview for DSM-IV ADHD.

Diagnosing ASD

In the screening step, the AQ was used to screen for ASD. 50 Participants with above threshold AQ scores (> 31) were invited to complete the Autism Diagnostic Observation Schedule – Generic (ADOS-G). 27 The ADOS-G algorithm scores participants as below threshold, autism spectrum or autism. The ADOS-G is a structured play and conversational interview that includes a series of social presses and other opportunities to elicit symptoms of ASD. Thirty-one accompanying ratings are divided between the domains of social interaction, communication, imagination/creativity, stereotyped behaviours and restricted interests. According to the coding algorithm, scores in the domains of social interaction, communication and the sum of these two, contribute to the ADOS-G classification of ‘below threshold’ (≤ 6), ‘autism spectrum’ (≥ 7), or ‘autism’ (≥ 10).

Overview of findings for ADHD

In total, we screened 1468 participants for ADHD from adult mental health and prison mental health services (Table 21). Across all the participants, we estimated an overall prevalence of ADHD of 15.1%, which is around a three- to fivefold increase compared with the population rate of 2.5–4.3%. However, there were marked differences depending on the clinical group being evaluated. The highest rate of ADHD was seen among prison inmates (20.8% for older prisoners; 16.9% for younger prisoners), followed by primary care mental health patients (11.6%), addiction patients (11.7%), forensic psychiatry patients (7.3%) and anxiety disorder patients (6.1%).

The higher rate of ADHD in the older prisoners is not significantly greater than the rate estimated for the younger prisoners.

The rates in the forensic psychiatry unit appear to be only slightly raised above that in the general population; however, further analysis found that this depended on the primary diagnosis. No cases of ADHD were identified among 81 patients diagnosed with a severe mental illness (schizophrenia in most cases), whereas the estimate was 23.5% among the smaller group of 52 forensic patients with a primary diagnosis of severe and dangerous PD. When the criteria were broadened to include those with high levels of ADHD symptoms (four or more symptoms in either domain) persisting from childhood and still causing impairment (ADHD in partial remission), the prevalence within this subgroup was 35.3%.

The increased rate of ADHD in the primary care setting is particularly interesting, as we screened only patients who were currently being treated for common mental health conditions, such as anxiety and depression. This contrasted with the lower rate of ADHD in the specialist anxiety disorder unit, which consisted of patients with a sufficiently severe form of anxiety of OCD to require inpatient treatment.

For the addiction inpatient unit, when patients were entering detoxification programmes, we noted a significant decline in the numbers screening positive for ADHD within 1 week of their admission to the unit. Current ADHD scores dropped significantly between screening assessment session 1 (T1) and screening assessment session 2 (T2) ratings (1 week part) with a mean change of 8.6. We found that 52% of participants with T1 and T2 data had a clinically meaningful reduction of ≥ 8 points on the self-completed ADHD rating scales, which is equivalent to a one-level drop in the Clinical Global Impression (CGI) scale. 112 This finding means that care must be taken when evaluating the level of ADHD symptoms in patients actively addicted to drugs and/or alcohol.

Across the entire set of clinic samples, none of the 90 cases identified as currently meeting diagnostic criteria for ADHD was currently recognised as having ADHD or being treated for the disorder. A small number had been diagnosed during childhood, and this was particularly notable among the young prison study group, who were all aged between 18 and 25 years. None of the cases of ADHD identified in the other services had received a diagnosis during childhood.

Overall, these data find high rates of undiagnosed and untreated ADHD in adult mental health and forensic services, particular among young male prisoners (even though they are routinely screened for other mental health conditions), among the subgroup of male forensic psychiatry patients with a primary diagnosis of severe and dangerous PD, among patients with an addiction disorder and among primary care patients being treated for other common mental health conditions.

A total of 842 participants completed screening for ASD; we did not screen for ASD in young offenders as there were local concerns (expressed by the prison) about the inclusion of the necessary additional screening and diagnostic steps in this population. We were, however, able to carry complete screening in an adult category C prison (Brixton).

An overall rate of ASD was estimated to be 5.5%, which is around eightfold higher than the general population prevalence of 0.6%. However, obtaining accurate diagnoses of ASD was difficult because of the limited access to parent reports of early childhood and developmental onset and course. This was a particular concern within the forensic psychiatry group, in which there was an estimated 17% of patients with ASD according to the ADOS-G assessment, which evaluates current autistic-like behaviours. Many of these patients had a primary diagnosis of schizophrenia and it is possible that ADOS-G scores are explained by the schizophrenic disorders and not autism. Further investigation would require autism diagnostic interviews to be completed with the parents of the forensic psychiatry cases, which was not possible owing to the low rate of agreement to contact families within this patient group. Higher rates were also found in the older adult prisoners (5.4%) and the specialist anxiety unit (5%), but only a relatively small increase in rates within primary care (2.0%) and the addiction service (1.2%). The association of severe anxiety disorders with ASD is interesting because it suggests that in at least some cases ASD may be an important risk factor, leading to sustained difficulties in controlling high levels of anxiety. The rate of ASD was higher in the anxiety unit than within the primary care setting, suggesting that the link is with severe anxiety that has not responded to other forms of treatment, rather than the more common forms of anxiety seen in general practice. ASD, as well as ADHD, appears to be linked to criminal behaviour, leading to incarceration in either forensic psychiatry or prison settings.

In addition to providing an overview of the findings in the section above, we have completed further, more detailed ,analyses of the forensic psychiatry and addiction subpopulations. Here we not only focus on the prevalence of undiagnosed ADHD and ASD, but also start to map out the associated impairments. We have mainly focused on the analysis of impairment data in the ADHD data.

Recruitment of participants

General practitioner surgeries for recruitment were identified in the local area by the Primary Care Research Network (URL: www.nihr.ac.uk/research-and-impact/nihr-clinical-research-network-portfolio/), and approached by the research team to take part. In accordance with Primary Care Research Network standards, the surgeries/GPs were paid a fee for participating in the study and a ‘per-head’ cost for each participant recruited to the study. Eligible participants included primary care patients aged > 18 years receiving a mental health intervention, including prescription of selective serotonin reuptake inhibitor medication and/or referral to Improving Access to Psychological Therapies services. A list of eligible patients was generated from the surgeries’ computerised databases. These lists were checked by GPs themselves to exclude any patients inappropriate to approach. Researchers prepared invitation letters to post to potential participants. These were signed by the GP before posting on paper or electronically. The letter included information on the study and consent forms to allow participants to opt in. The consent form included permission for researchers to notify GPs of the outcome of their patients’ participation. Inclusion criteria included all patients with non-psychotic disorders attending the clinic. Exclusion criteria for most settings will include major mental disorders such as schizophrenia, bipolar I disorder and dementia. Inclusion/exclusion criteria may differ between units.

ADHD treatment outcome in primary care (qualitative feasibility/pilot study)

From the primary care sample, 15 patients were started on treatment for ADHD. We took baseline measures and after 6 months reviewed their progress. We are currently completing this stage of the project; however, all patients reported improvements in symptoms of ADHD and associated impairments.

We conducted review meetings with the patients on ADHD medication. During the meetings some interesting comments were made about the effect of medication and the treatment process. Patient LL (60-year-old female) reported being able to ‘cope with things better’ and can ‘sit still and watch full programme on TV’, which she had never been able to do before starting medication. Overall, she felt her quality of life had greatly improved. Patient AW (57-year-old male) reported feeling ‘much better’ on medication as it helps get his ‘head together’ and he was getting many more things done. Patient KC (28-year-old female) reported that the medication made her feel ‘more functional’. Medication ‘helped 50% and the rest was just having the diagnosis as it did wonders for my self-esteem . . . used to be down on myself but now I know there is a reason behind how I am’. Others all reported finding the treatment and improved understanding of their condition an enormous help, with some praising the team for finding solutions to long-standing difficulties. For example, one man often had arguments with his daughter who was also diagnosed and treated for ADHD. He now found that his behaviour was much better controlled and he had time to stop and think rather than starting a shouting match with his daughter.

ASD in primary care

One of the most striking findings within the primary care sample was the high rate of false positives. A total of 62 out of 77 patients who were AQ positive had a further ASD assessment (ADOS-G). Only four cases were above threshold (6.45%) and 58 cases were below threshold (93.6%). This suggests that the AQ may not be an effective screening measure in individuals with associated mental health conditions such as generalised anxiety or depression. However, it is important to note the lack of information on developmental history within the primary care sample. The mean age of individuals who were AQ positive was 43 (range 20.6–65.1, SD 11.6) years and many individuals were unable to provide informants to complete the ADI-R assessing key behaviours during early development. Without this essential information it is hard to be say definitively whether the individuals who scored above threshold on the ADOS-G actually had ASD and/or the ADOS-G failed to identify some cases.

Analysis of impairment data revealed that patients who screened positive for ASD had significantly greater social impairments than ASD-negative individuals. There were no other significant group differences on the remaining five scales of the WFIRS (Table 23).

When analysis was repeated using a continuous approach, there was a significant positive correlation between AQ total and family (r = 0.28; p = 0.03) and social (r = 0.31; p = 0.01) impairment scores on the WFIRS. The results indicate a significant association between a higher number of autistic traits and greater self-reported impairment in family and social domains (Figure 12). There were no significant differences between ASD screen-positive and ASD screen-negative individuals on the Pittsburgh Sleep Quality Index scale, measuring sleep impairments, or on the Affective Lability Scale, measuring affective liability.

FIGURE 12.

Scatterplot displaying the relationship between AQ total score and self-report family impairment on WFIRS.

It was not possible to compare those above and below the threshold for ASD due to the large difference in numbers in each group (4 vs. 172).

Analysis of impairments in populations

Further analyses on impairments have yet to be completed.

Participants

Sample characteristics are listed in Table 24. Participants (n = 226) were recruited from two inpatient alcohol and drug detoxification and stabilisation units in south-east London. Consecutive admissions to the inpatient units were invited to take part in the study over an 18- (clinic 1) or 11-month (clinic 2) period. Of these, 74% had previously attended a detoxification treatment programme. One participant already had a current adult diagnosis of ADHD and two had a recorded childhood diagnosis of ADHD. We excluded individuals with a history of current or recent psychosis (n = 4), current serious physical illness (n = 3), insufficient understanding of English to give informed consent (n = 6) and lack of capacity to consent (n = 2). Overall, 51% of patients approached agreed to participate. Study participants received no payment for participation.

Study procedures

The study of people undergoing inpatient detoxification included two screening stages using ADHD rating scale data, followed by formal evaluation of the diagnosis of ADHD using a semistructured interview for DSM-IV ADHD. By including two screening steps (T1 and T2) we were able to evaluate the difference between the level of ADHD symptoms on admission (T1) and 1 week later (T2), when they had been detoxified or stabilised on treatments such as methadone.

Screening assessment session 1 (T1) took place soon after admission (mean 5 days, SD 4 days). The second screening assessment session (T2) was completed as close as possible to 7 days after the T1 assessment (mean 8 days, SD 6 days). During the T1 session, patients completed current and childhood symptom checklists for DSM-IV ADHD; however, during the T2 session patients completed current symptoms only, as it was assumed that only the report of current levels of symptoms would be affected by state of detoxification and current drug use and needed to be measured at both time points. In addition, whenever possible, we obtained contact details from informants for childhood and current ADHD symptoms who were asked to complete an informant version of the screening questionnaire. Current informant ratings were obtained for 72 participants and childhood ratings for 48 participants. The self-rated screening data were used to allocate participants into a screen-positive group and a screen-negative group for ADHD.

Participants screening positive for ADHD were invited to complete a research diagnostic interview assessment. Participants who failed to attend the diagnostic interview were offered up to three further appointments. Following three missed appointments, a final letter invited participants to contact the research team to reschedule an appointment before the assessment was recorded as ‘missing’.

Childhood ADHD screeners

Mean total childhood ADHD symptom scores were highest in the group treated for drug dependency (mean 29, SD 14), followed by the combined drug and alcohol group (mean 24, SD 15) and the alcohol dependency group (mean 17, SD 14). A one-way analysis of variance (ANOVA) showed significant effects of primary diagnosis (drug vs. alcohol vs. combined) on childhood screener total score (p < 0.001). Planned contrasts showed significantly higher ratings of ADHD symptoms for drug than for alcohol participants (p < 0.001), but not between drug and combined drug and alcohol groups (p = 0.12).

T1 compared with T2 ratings

These data are summarised in Table 25. Current ADHD scores dropped significantly between T1 and T2 ratings, with a mean change of 8.6. We found that 52% of participants with T1 and T2 data had a clinically meaningful reduction of ≥ 8 points on the self-completed ADHD rating scales, which is equivalent to a one-level drop in the CGI scale. 112

There is a significant reduction in ADHD symptoms following 1 week or more of treatment in the addiction units. A 2 × 3 mixed model factorial ANOVA on the time (T1 vs. T2) and primary diagnosis (drug vs. alcohol vs. combined drug and alcohol) found a significant main effect of time (p < 0.001), indicating that self-rated ADHD symptom scores are influenced by the detoxification treatment process and/or withdrawal states following admission. A significant effect of primary diagnosis was also found (p < 0.001), but no significant interaction between time and primary diagnosis (p = 0.34). Between-subjects contrasts showed significant differences for ADHD ratings between the alcohol and combined drug and alcohol groups (p < 0.001), but not between drug and combined (p = 0.28) or alcohol and drug (p = 0.11), indicating higher levels of ADHD symptoms at T1 and T2 among those with drug abuse compared with those with alcohol abuse alone.

Estimated prevalence of ADHD

There was an estimated prevalence rate for ADHD of 12.2%. Of these, 73% met criteria for the combined subtype, 18% for the hyperactive–impulsive subtype and 9% for the inattentive subtype. The prevalence was calculated using the proportion of participants completing each step that did and did not fulfil ADHD screening and assessment criteria. This in turn allowed us to calculate an estimated prevalence rate of positive screens at each stage, which is not affected by attrition rates between screening and assessment stages.

To evaluate whether or not rates of ADHD might differ in the groups that screened positive for ADHD at T1 and T2, but did (or did not) attend for diagnostic interviews, we compared the attendees to the non-attendees. The participants who attended the diagnostic assessments showed no significant differences from those that did not at T1. However, they had significantly higher score for inattentive (but not hyperactive–impulsive) symptoms at T2 (6.1 vs. 4.3; p = 0.025), suggesting that the prevalence estimate based on available data might be slightly inflated.

Broader definitions of ADHD

Broadening the diagnostic threshold, based on the DIVA (diagnostic interview), had only a small impact on the prevalence estimate. For example, reducing the threshold for diagnosing ADHD from six or more symptoms in childhood to four or more increased the overall prevalence estimate only to 14.4%. In addition, no additional cases were identified by decreasing the threshold for current symptoms to four or more.

Informant rating scales

Sixty-four per cent of participants provided contact details for informants of their current behaviour and 43% for childhood behaviour. In addition, 50% of both sets of contacts returned screening questionnaires. Informant ratings of current ADHD symptoms correlated moderately with self-reported symptoms and were similar to the equivalent correlations in the control sample (r = 0.47; p < 0.001). However, the participants with positive screens for ADHD agreed far more strongly with their informants than those who did not screen positive for ADHD (correlations 0.62 to 0.66). Correlations between childhood self and informant ratings were similar to those for current behaviour, but were non-significant. Overall, participants and informants were concordant for the 4+ screening criteria in 64% of cases at T1 and in 50% of cases at T2 and 81% were concordant for the childhood 6+ criteria (Table 26).

Impairment and substance use

As expected, the SUD sample rated themselves as significantly more impaired than controls (p < 0.05). Among the individuals with SUD there was a significant positive relationship between self-rated impairment and the severity of ADHD symptoms in childhood and at both T1 and T2 (respectively, r = 0.52, p < 0.001; r = 0.73, p < 0.001; r = 0.51, p < 0.001). Furthermore, impairment was significantly negatively correlated with change scores between T1 and T2 (r = –0.23; p < 0.01), indicating that participants with greater impairment at T1 showed a smaller drop in ADHD symptoms between T1 and T2.

Finally, within the SUD population we compared the diagnosed ADHD cases with the non-ADHD cases. Those with ADHD had significantly higher rates of self-reported impairments and were significantly more likely to have reported using cocaine or amphetamines (Table 27), consumed more units of alcohol per day and had more prior suicide attempts. In addition, they had a trend towards higher rates of depression, previous convictions and previous drug or alcohol dependency treatment (see Table 27).

ASD

Only four participants out of 113 who completed the AQ were above the threshold and, of these, only one met criteria for possible ASD. As the rates of ASD in this population, based initially on self-report data, appear to be very low, we did not pursue this line of investigation further. Future studies should consider informant reports of autistic behaviours at the screening stage, to exclude the possibility of low reporting of ASD behaviours using self-report questionnaires.

Impairments associated with comorbid ADHD and substance use disorder

The finding of higher levels of impairment among ADHD cases within the SUD sample appears to be robust, as we found indicators of impairment from both self-report and more objective (case note) measures. Appropriate assessment and management of ADHD in SUD patients would therefore seem to be potentially important to improve the general level of functional impairments, particularly given our finding that ADHD is associated with increased frequency of suicide attempts and depression. Whether or not treating ADHD in the context of SUD improves depression has yet to be adequately studied; however, we know that in people with ADHD symptoms such as mood instability and low self-esteem respond well to treatments for ADHD. 127–130

Another finding was that the ADHD group had significantly higher use of stimulants, such as cocaine and amphetamine (although not crack cocaine). One possible (but untested) explanation for this, given that stimulants are routinely used to treat ADHD in the general population, is that people with ADHD are using stimulants as a form of self-treatment. However, an additional mechanism could be that people with ADHD have a preference for drugs that are more ‘stimulating’, especially when injected or taken at high dose. Further work is needed to address this issue.

Definition of ADHD and methodological issues

Our work attempted to address, as best we could, the inevitable methodological difficulties that might impact on accurate estimation of prevalence rates for ADHD within people with SUD. These include the potential for unreliable information from self-report questionnaires, difficulties with retrospective recall of childhood symptoms and the direct impact of drug and alcohol intoxication and withdrawal states on ADHD-like symptoms. We therefore attempted to measure ADHD symptoms before and after detoxification, included independent informant ratings whenever possible, compared these with population control data and completed research diagnostic interviews. One additional confounder might arise if the study participants considered a positive diagnosis of ADHD as a way to obtain stimulant medication. However, the study was not linked directly to the treatment of ADHD, and expectations for treatment with stimulants are currently low in the UK, because in most cases ADHD in adults goes unrecognised and untreated. 102

Based on the percentage of patients who screened positive for ADHD in childhood and adulthood and the proportion of those interviewed, we estimated an overall rate of ADHD of around 12% in our sample, much higher than the equivalent estimated worldwide prevalence of around 2.5–3.4% in non-SUD populations. 97,101,113 Using slightly broader criteria for ADHD to reflect different approaches to the diagnosis of adult ADHD being taken by different investigators and clinicians, we found the estimated prevalence in our sample increased to a maximum of 15%. Overall, the estimated prevalence of ADHD in our sample is far higher than population rates, yet lower than those cited in some previous studies of ADHD in SUD populations. There are several potential reasons for these differences in the findings from this and previous studies.

One potential reason was our stringent application of the DSM-IV criteria, which might have led to an underestimate of the true rate of ADHD in the SUD population for several reasons. First, the inherent problem in collecting childhood data retrospectively might mean that some participants who met current criteria for ADHD may have been unable to recall sufficient examples of childhood symptoms. We recorded ‘unknown’ ratings in the diagnostic interview assessments if participants could not provide sufficient information to conclude that a symptom was present or absent and found that on average 15% of childhood symptoms could not be scored. Second, DSM-IV criteria only require that ‘some’ symptoms and impairments were present during early childhood and do not specifically require six or more symptoms from childhood so long as the symptom count criteria are currently met as an adult. Nevertheless, we decided to take the more stringent approach because of the lack of prospective data from childhood and to guard against inclusion of ADHD-like syndromes that might arise as a result of chronic drug abuse in the absence of an underlying ADHD diagnosis. Finally, we know that the current DSM-IV criteria are not adjusted to take into account age-related changes in the development of ADHD, and there is evidence that four or more symptoms in adults, rather than the current six or more symptoms, is sufficient. Indeed, this change is being considered for the fifth revision of the DSM that is currently in preparation. 102,131,132 However, taking all of these alternative thresholds into account had only a minor impact on our estimate of the prevalence of ADHD in the SUD population.

The impact of drug detoxification on ADHD symptoms

We investigated the impact of drug intoxication and/or the detoxification process by evaluating self-rated ADHD symptom scores a few days after admission to the detoxification unit and 1 week later, when the participants were detoxified or stabilised on long-term medication and no longer in a withdrawal state. We found significant decreases in ADHD symptoms of around 8 points (15% of the total score), which is a clinically significant reduction in ADHD symptoms and comparable to a one-level drop of the CGI scale. 112 In terms of our screening criteria, this led to 40% of patients no longer meeting screening criteria for ADHD at T2 compared with T1. Hence, prior studies may have reported a higher prevalence than we found due to the confounding effect of drug use and/or withdrawal symptoms. Other researchers have noted mood disturbances during alcohol detoxification133 and it is therefore possible that ADHD-like symptoms are also part of the withdrawal syndrome. Despite this, we found that in 60% of cases self-rated ADHD symptoms remained clinically significant following completion of the detoxification process. One implication is that, although the withdrawal process may impact on the level of ADHD symptoms (and this should be taken into account when evaluating ADHD in SUD patients), there remain a significant number of individuals with clinically significant symptoms of ADHD – and these require treatment.

Clinical implications

Our findings suggest that clinical evaluations for ADHD are probably best completed once detoxification or stabilisation for drug or alcohol dependency has been completed. However, this suggestion may be difficult to implement in community patients. Furthermore, once diagnosis of ADHD has been established it will be important to offer treatment. However, use of some pharmacological treatments (such as stimulants) is complex in those with a current SUD. Nevertheless, therapeutic nihilism is not an option, as treatment of underlying ADHD may be important to the success of drug treatment programmes for some individuals. It might be advisable to use the more stringent screening criteria of six or more symptoms in either domain to identify those that need full clinical evaluation for ADHD, or to recognise the need for detailed ADHD assessment in a higher proportion of cases. Moreover, it may be more appropriate to use non-stimulant medications, such as atomoxetine, as a first-line treatment. Further work is required to address this issue and the potential for using CBT-based therapies for ADHD in this complex population.

Obtaining informant data on ADHD symptoms for childhood and for current symptoms proved difficult in this population. The reasons for this were not investigated here, but probably reflect the often poor relationships that many SUD patients have with their family and friends. Therefore, in clinical practice it will also often be the case that the diagnostic assessment of ADHD will depend on self-report alone. We were able to investigate the validity of these self-reported data by comparison of self-report with informant-reported data in a subset of our sample and a comparison control sample. This investigation revealed moderate correlations between raters which were similar to that seen in non-SUD control populations. For the most impaired subgroup, however, who were screening positive for ADHD based on their self-report, there was a far higher correlation with informant report for current ADHD ratings of around 0.62–0.65. We therefore suggest that, although discrepancies between raters exist for ADHD rating scales, this does not appear to be different for SUD compared with control populations. Furthermore, ratings showed moderately high levels of agreement for the group of patients with ADHD.

Previous research has shown that in general people with ADHD tend to rate their symptoms lower than informants,134 perhaps reflecting difficulties in self-evaluation of ADHD symptoms. Our research does not support this finding within the SUD population investigated here. When we completed diagnostic interviews with 26 people who screened positive for ADHD on the basis of their self-rated symptoms, we found that only 61% met full criteria for ADHD. The clinical implication of this finding is that the diagnosis of ADHD in patients with SUD should depend not solely on rating scale data, but rather on more objective examples of symptoms characteristic of ADHD, as applied here using the DIVA interview. Rating scales are a valuable tool for screening for ADHD, but should not be used as a replacement for a full diagnostic assessment by clinical interview. Furthermore, although informant reports are helpful in supporting the diagnosis, the moderately high correlations with self-report suggest that in most cases self-report alone should be sufficient. For inpatient SUD units it should also be feasible to observe patients for level of restlessness, problems with self-organisation, inattentiveness, impulsive responses and poor emotional regulation, which are characteristic of ADHD in adults.

Participants

Adult patients participated in the study from both high- and medium-security establishments in the Greater London area. Three hundred and forty-one mentally disordered offenders, all of whom were detained under the UK Mental Health Act 2007,135 were resident at these two secure services [241 (71%) at the high-secure service and 100 (29%) at the medium-secure service]. The majority of patients in these settings have a primary diagnosis of either serious mental illness (SMI; e.g. schizophrenia, schizoaffective disorder, bipolar disorder) or PD. Exclusion criteria included age > 65 years and those patients who were too mentally unstable to participate, had severe cognitive deficits due to neurological illness or head injury, posed a risk of violence to the researcher and/or who lacked capacity to consent to participate in the study. Of the 341 patients, 93 (27.3%) were identified as ineligible to participate; 68 (20%) did not meet study criteria and a further 25 (7%) were on trial leave or discharged prior to the assessment. Of the 68 patients who did not meet exclusion criteria, 28 lacked capacity to consent and 40 were mentally unstable. Thus, a total of 248 patients were eligible to take part in the study. From this sample of 248 patients, a further 115 (46%) patients refused to participate in the research. Hence, 133 participated in the study [92 (69.2%) from the high-secure psychiatric service and 41 (30.8%) from the regional medium-secure service].

Assessment procedure

The study took place over a 16-month period. An information sheet describing the study and specifying inclusion and exclusion criteria was sent to clinical teams. Patients meeting these criteria were referred to the researchers, who subsequently approached the patients to explain the study and obtain informed consent. Once consent had been given, patients met with a RA to complete screening measures for ADHD and data were extracted from the clinical records. Four patients in the SMI category consented to participate in the study and met with researchers, but did not consent for their records to be accessed. Hence, there were missing data for these participants.

If a patient screened positive on the BAARS-IV scales (self-rated scales for current and retrospective ADHD symptoms), they were invited to participate in a diagnostic interview. These interviews were administered by a RA who had received training, including observation of these assessments by qualified staff in a clinical setting; specific training by qualified and experienced clinicians who routinely administer the interviews; and practice ratings of recorded clinical interviews to a point of reliability and convergence of ratings. For patients who were positive on the clinical diagnostic interview for ADHD, an informant interview was conducted with the patient’s primary nurse to supplement information from the self-report interview. Once completed, materials for each case were reviewed by a consultant psychiatrist (PA) from the Maudsley Hospital Adult ADHD service.

Behavioural function

Critical incidents were obtained from clinical records recorded by staff over the previous 12-month period. The data were grouped based on harmful impact on others: (1) ‘no harm incidents’ including security issues, verbal aggression and physical injury not requiring treatment or (2) ‘harmful incidents’, classified as any physical injury requiring further assessment or treatment. There were no critical incidents involving grievous bodily harm or homicide during the 12-month period. Seclusion data were obtained from clinical records recorded by staff over the previous 12-month period and categorised by number of episodes and duration of time in seclusion.

Comorbid diagnoses

All participants had a DSM-IV primary diagnosis of SMI (n = 81, 60.90%) or PD (n = 52, 39.10%). Four patients in the SMI category denied us access to records so we were unable to specify their primary diagnosis, offence history, ethnicity, age, critical incidents, episodes and duration of seclusion. The SMI category consisted of patients with a primary diagnosis of psychotic (n = 71, 92.21%) or bipolar disorders (n = 6, 7.79%). The PD category consisted of patients with a primary diagnosis of borderline (n = 24, 46.15%), antisocial (n = 23, 44.23%), schizoid (n = 3, 5.77%), narcissistic (n = 1, 1.92%) or histrionic (n = 1, 1.92%) PD. In common with all forensic mental health settings, there were high rates of comorbidity within the sample; however, only six patients had a history of ADHD recorded in their records (three in the SMI category and three in the PD group), and none had a current diagnosis or was receiving medication for ADHD at the time.

Age, ethnicity and history of offences

Most SMI patients were male (n = 79, 97.53%), with ethnicity classified as white (n = 39, 50.65%), black (n = 32, 41.56%), Asian (n = 2, 2.60%) or mixed race (n = 4; 5.19%). In the PD category, all patients were male (n = 52), with 47 (90.38%) classified as white, two (3.85%) as black, one (1.92%) as Asian and two (3.85%) as mixed race. Participants were aged between 19.3 and 64.3 years with no significant difference in age between the SMI (mean 37.10 years, SD 10.69 years) and PD categories (mean 40.26 years, SD 11.19 years) [t₍₁₂₉₎ = –1.62; p = 0.11]. Nearly all the participants had a history of violence (n = 96, 74.42%) and/or sexual violence (n = 26, 20.16%). Other offences (n = 7, 5.43%) included arson, burglary and driving offences.

Prevalence estimates

One hundred and thirty-three patients completed the ADHD child and current screeners (SMI, n = 81; PD, n = 52). Compared with patients in the SMI group, in the PD group patients had significantly higher BAARS-IV current total scores [SMI: mean 12.00, SD 9.93; PD: mean 15.92, SD 15.92, t₍₁₃₃₎ = –2.11; p = 0.04] and childhood scores [SMI: mean 16.59, SD 13.89; PD: mean 27.41, SD 15.07, t₍₁₃₃₎ = –2.11; p = 0.00].

Twenty-six patients (19.55%) screened positive for ADHD on the BAARS-IV scales and were offered a clinical diagnostic interview. Seven (8.64%) patients had a primary diagnosis of SMI and 19 (36.53%) had a primary diagnosis of PD. Three patients refused further assessment with a clinical diagnostic interview; thus, 23 of the screen-positive participants completed a DIVA (SMI, n = 6; PD, n = 17). Following the DIVA assessment, four participants (23.52%) in the PD category showed persistent ‘syndromatic’ ADHD (one with hyperactive–impulsive type and three with combined type). When applying the more relaxed ‘symptomatic’ criteria, six participants (35.29%) in the PD group had persisting symptoms (three hyperactive–impulsive type and three combined type). Of the six participants in the SMI category who screened positive, none met criteria for either ‘syndromatic’ or ‘symptomatic’ ADHD. Hence, subsequent analysis was conducted on only the PD group.

Self-rated impairments

The forensic patients who met clinical and symptomatic criteria for ADHD (n = 6) and those who screened positive for ADHD (n = 19) rated themselves as significantly more impaired than forensic psychiatry patients who screened negative for ADHD, with a large effect size for all items except for money management (Table 28). Significant positive correlations were also seen between ADHD symptom scores (in adulthood and childhood) and all domains of self-reported impairment (Table 29).

Critical incidents and seclusion

Critical incident and seclusion data taken from prison records did not show a significant difference (albeit we found trend to higher rates in ADHD cases; p = 0.07). This perhaps reflects the serious nature of the comorbid disorders and behavioural problems, other than those related to ADHD, in this high-risk population and the small sample of identified ADHD cases. Using the broader category of screen-positive cases, there was a significant increase in hours spent in seclusion compared with the screen-negative group (see Table 28).

Strengths and weaknesses

The study is limited by its small sample size, which is partly due to the high rate of refusal (46%), something that is difficult to avoid in the settings investigated. Hence, the sample may be biased as some patients may have refused to participate in a lengthy clinical assessment (including the completion of screening questionnaires) because of problems with attention span and other symptoms associated with ADHD, such as irritability. 94 Alternatively, those with ADHD symptoms might have been more interested in taking part in the study. These considerations might have resulted in a non-representative sample, which was underpowered to detect some of the impairments, such as critical incidents, in within-group analysis. A further limitation is that we did not interview those individuals who screened negative for ADHD, so the specificity of the screening tool in this population remains unknown. Despite these limitations, the use of the DIVA diagnostic interview gives confidence in the finding of high rates of ADHD in the PD group. DIVA is designed to establish the presence or absence of each of the 18 DSM-IV symptoms for ADHD during both childhood and adulthood, in addition to the essential age at onset, situational pervasiveness, symptom chronicity and impairment criteria. As such, it is expected to give lower prevalence rates than the use of symptom rating scales alone. A potential limitation is the ability to accurately identify ADHD symptoms retrospectively when relying on self-report data alone, although in the context of this study this would lead to more conservative estimates of prevalence. Future studies would benefit from diagnostic interviews for ADHD in a wider range of patients and further work to improve on effective screening tools in forensic mental health populations.

Step 1: electronic database search

A search of electronic databases (including the Research Autism website, MEDLINE, PsycINFO, EMBASE and PubMed) identified just four relevant articles that focused specifically on ASD in prisons. One was a study in Scotland on the evaluation of a screening tool for ASD in prisons;143 another was on knowledge of ASD among prison staff144 and one reported two case studies of prisoners with Asperger syndrome. 145 These papers are included in the review below. The final article was a Swedish study that could not be obtained in full text. 146

Step 2: rates of ASD and ADHD

Having completed the search of available published data (see Step 1: electronic database search), our next aim was to examine how many adults in a local prison have NDs such as ASD and/or ADHD and to examine rates of associated ID and mental health problems in these individuals in a local prison (HMP Brixton, a category C resettlement prison with capacity for up to 798 category C and D prisoners).

The following screening tools were used: ID, Learning Disability Screening Questionnaire; ASD, Autism Spectrum Quotient; and ADHD, ARSR. Those screening positive were assessed further using the Quick Test (ID) and the so-called ‘gold standard’ research diagnostic tools for ASD and ADHD [Autism Diagnostic Observation Schedule – 2nd Edition/ADI-R (for ASD) and DIVA (ADHD)]. All prisoners screening positive for ID, ASD or ADHD (and all control group participants) were further assessed for mental health diagnoses using the Mini International Neuropsychiatric Interview, which covers 22 DSM-IV/ICD-10 diagnoses.

Step 1

Step 2

As of 16 May 2013 we had approached 351 prisoners, and, of these, 221 were screened. Fifty-nine participants (27%) were housed on the VPs wing. Mean age was 36 years, 41% were white British (around 50% of prisoners in Brixton are from black and minority ethnic backgrounds) and before coming into prison 52% were single, 11% homeless, 47% unemployed and 27% reported past or current mental health problems. Eighty-one prisoners have screened positive for a ND (ASD, n = 39; ADHD, n = 60; ID, n = 32). For the ND group we aim to have a full data set (follow-up diagnostic assessment and the Mini International Neuropsychiatric Interview) on 70 participants. Currently this stands at 60 participants, with eight being followed up.

Comorbidity

Our preliminary analysis suggests that adult prisoners with ASD and/or ADHD have extremely high rates of associated mental health symptoms (Table 30).

ADHD and transition

Our analyses were based on the follow-up interviews with young people aged 14–24 years and their parents/partners who were recruited from the IMAGE research database through a childhood clinical diagnoses of ADHD. Participants were included if they provided CSRI data at either wave 1 (n = 86) or wave 2 (n = 79): 66 participants provided data at both waves. Data were also available from the CANDID, the DIVA, the ADHD-specific BAARS-IV rating scale, the DAWBA, the CIS-R, the Zarit Burden Interview and the AUDIT.

Additional questions, developed in the Social Policy Research Unit at the University of York, asked about participants’ experiences of transition processes. 163

Our participants were aged, on average, 17 years; most were male and more than half were in full-time education, although this figure had reduced by the wave 2 interviews (Table 31). Most remained living with their parents and, at wave 1, 43% were taking ADHD medication. The young adults used a wide variety of services, but, in most cases, few people used any one particular service.

The most common source of support in both waves was GPs, both for ADHD-specific problems (23% in wave 1 and 15% in wave 2) and for other reasons (33% in both waves). Many participants had repeat prescriptions (52% in wave 1 and 44% in wave 2). Use of specialist mental health services was sparse, with the exception of psychiatrists in wave 1 (26%), but this had dropped to 15% of the sample by the wave 2 interviews. In addition to these formal services, websites and helplines appear to play a fairly important role in supporting young adults, with 19% in wave 1 and 13% in wave 2 reporting their use in the 3 months prior to the interview.

The costs associated with service use and additional education supports (i.e. excluding the costs of mainstream education) are shown in Table 32. In both waves, education was the highest contributor to total costs, accounting for 46% at wave 1 and 34% at wave 2. Mental health services accounted for just 15% and 13% of costs, respectively. Although the average cost of primary care was similar in both waves, the proportion it contributed to total costs was 16% in wave 2 compared with just 8% in wave 1. There was a statistically significant reduction in total costs from wave 1 to wave 2 (£665 vs. £390), with significant differences in the costs (subtotals) for hospital and for mental health-care services. Costs for advice services and self-help supports were similar at each wave.

Our analyses of these data will explore what individual characteristics and needs are associated with these costs at each wave and across both waves. We will also look in more detail at the group of young adults who have transferred from children to adult services (n = 18) over the period; those who do and who do not (n = 13, £0) receive any support; and those who do or do not use specialist mental health services. These analyses will help address the questions identified in Background.

ASD and transition

Young people aged 14–24 years with a diagnosis of ASD were recruited from CAMHS and adult clinics in south London, voluntary organisations and research databases held at the IoP through their child’s childhood clinical diagnosis of ASD. Data for 81 participants were available. As well as the CSRI, data were collected on the CANDID, the AQ – Informant version, the DAWBA, the AUDIT and the Zarit Burden Interview.

The majority of participants were male and white British (85% each). At wave 1, ages ranged from 14 to 25 years, with an average of 17.9 years. Most participants had an International Classification of Diseases (ICD) diagnosis (94%), with > 90% diagnosed as on the autistic spectrum (autism, Asperger syndrome). On average, parents reported needs in nine of the CANDID domains, six of which parents considered to be met needs.

Fifty six (69%) of these young adults were still in full-time education, with a further six in part-time education (Table 33). Two-thirds had a statement of SEN. Additional support in school was common in both waves 1 and 2, with around one in three young adults receiving individual help in class, some lessons taught in smaller classes, support from the special educational needs co-ordinator, or additional meetings with teachers.

The most commonly used service was the GP, seen by 36% of the sample in wave 1 and 35% in wave 2, although by just 9% in each wave for ASD-related reasons. Repeat prescriptions were made for nearly half the sample at both waves (47% and 46%, respectively). Use of mental health care was similar to the proportion found among the ADHD sample: < 5% of participants used ASD-specific outpatient clinics, therapists or counsellors; 27% had seen a psychiatrist (17% in wave 2); 17% had seen a psychologist (10%); and 6% (2%) had seen a psychiatric nurse specialist. Fewer than 5% of the young adults used community-based therapy services such as occupational or speech and language therapists, but social workers were seen by 16% of participants and 21% had a keyworker.

The costs of support over 3 months are shown in Table 34; nearly half of total costs are absorbed by additional supports provided by the education authorities with higher costs due to attendance at special needs schools. A further one-fifth was accounted for by provision of social care services; higher costs here are for one young adult spending time in a residential care unit and three who used respite care services. Mental health services contributed just 5% to total costs in wave 1 and only slightly less in wave 2 (4%).

Total average costs were lower in wave 2 than in wave 1 (£4557 vs. £2919), although there was a wide range of costs at both periods; for a number of young adults, total costs were £0 indicating no services or support were used. Mean costs for additional education, hospital services and social work/social care all were lower in wave 2, although again showing wide ranges for each. The only significant difference between wave 1 and wave 2 is mental health service costs (p = 0.02).

As with the ADHD sample, our analyses of these data will explore what individual characteristics and needs are associated with these costs at each wave and across both waves. We will also look in more detail at the group of young adults who have transferred from children to adult services over the period, those who do and who do not receive any support, and those who do or do not use specialist mental health services. These analyses will help address the questions identified in the earlier section (see Background).

Selection and diagnostic assessment of ASD sample

Screening for ASD was carried out using the Social Communication Questionnaire (SCQ),191 the Childhood Autism Spectrum Test (CAST)192 and additional questions about whether or not children had previously been given an autism or Asperger syndrome diagnosis. In addition, some families also spontaneously disclosed that one or both of their children was suspected of having ASD during data collection unrelated to ASD. Data were collected via questionnaires mailed to parents when their children were aged 5–8 years (SCQ), 8 years (CAST) and 9 years (questions about previous ASD diagnosis), and via telephone interview when children were aged 7 years (questions about ASD diagnosis). A total of 14,797 families were invited to complete questionnaires or a telephone interview and 8941 (60.4%) returned data at least once. This sample was broadly representative of the UK population and of the TEDS cohort as a whole (maternal ethnicity, 93.2% white; maternal education, 40.1% with A levels or higher).

Children were considered to be at risk of ASD if they met any of the following criteria:

• They scored ≥ 15 on the SCQ.

• They scored ≥ 15 on the CAST.

• Their parents had endorsed questionnaire items at ages 7, 8 or 9 years indicating that their child had been given a diagnosis of ASD.

• Their parents had spontaneously indicated at other times that they were concerned that their child might have ASD. Families where one or both children were considered to be at risk of ASD were then invited to complete the ASD module of the DAWBA.

Families were then visited at home for further assessment. The ADI-R and the ADOS-G were carried out by trained interviewers. Two interviewers assessed each family. One interviewer carried out the ADI-R for the first twin, while the other interviewer carried out the ADOS-G. The interviewers then reversed roles for the second twin so that the same interviewer did not carry out the ADI-R and the ADOS-G for the same child and did not carry out either for both twins in a pair. A consensus diagnosis was assigned by the study team after reviewing ADI-R and ADOS-G data and other relevant data, such as parent reports of previous clinical diagnosis. Children were classified as ASD (i.e. met criteria for childhood autism, Asperger syndrome, ‘other’ PDD) or unaffected. One hundred and twelve children with a diagnosis of ASD were included in analyses (child sex, 83.0% male; maternal ethnicity, 96.4% white; maternal education, 50.4% with A levels or higher).

Selection of comparison sample

The ‘at-risk’ sample consisted of children likely to have ASD and their co-twins, which included unaffected co-twins. However, as unaffected co-twins were expected to be at higher risk of subthreshold autistic traits,193 parents of twins considered to be low risk (score of < 12 on the CAST) were invited to complete the DAWBA. These families were selected to be matched to the ‘at-risk’ group on sex, zygosity, age and socioeconomic status. Home visits were carried out with the families, during which parents completed the family history interview and children were administered a battery of cognitive measures. Researchers also completed an observation sheet for each child, which included observations about the child’s social interaction with the researcher. Suspicion about ASD was raised for only one child in the control sample and an ADI-R was completed with their mother. No further data were available for this child and so they were excluded from analyses in which a consensus diagnosis was required.

Non-cases included 101 children who were unaffected co-twins of children with ASD and a further 68 children from the low-risk group. For analyses using ADI-R or ADOS-G data, non-cases included unaffected co-twins (n = 101; child sex, 52.5% male; maternal ethnicity, 96.0% white; maternal education, 52.0% with A levels or higher). For analyses using consensus diagnosis data, non-cases included unaffected co-twins and low-risk twins (n = 169; child sex, 60.4% male; maternal ethnicity, 96.4% white; maternal education, 53.0% with A levels or higher).

Measures

Statistical analysis

First, the DAWBA ASD module was compared with the ADI-R by examining the correlation between the DAWBA total impairment score and ADI-R total algorithm score using Spearman’s r. The distribution of scores for both measures was also compared in cases and non-cases.

Second, the optimal cut-off point for DAWBA probability bands in this sample was determined using receiver operating characteristic (ROC) curve analysis and by examining the sensitivity, specificity, PPV, NPV and overall percentage of children correctly classified at different cut-off points. The OR for cases versus controls was used to determine if there was a significantly increased risk of ASD associated with probability bands higher than the cut-off point. Robust standard errors, clustered by family, controlled for the relatedness of twins in a pair.

Third, the sensitivity, specificity, PPV, NPV and overall percentage of children correctly classified using diagnosis assigned by clinical rater were compared with the probability bands, to determine to what extent prediction was improved by the clinical rater.

Finally, ADOS-G data were added to the DAWBA clinical rater diagnosis and the sensitivity, specificity, PPV, NPV and overall percentage of children correctly classified were compared for children meeting criteria for ASD on one or both measures.

Performance of Development and Well-Being Assessment in comparison with the Autism Diagnostic Interview – Revised

The distribution of continuous measures of ASD symptoms, the DAWBA ASD total impairment score and the ADI-R algorithm total, were compared with each other in cases and non-cases. Figure 13 shows that the distribution using the DAWBA mirrors that of the ADI-R. Both measures are positively skewed in non-cases, with the majority of children having low scores and roughly normally distributed in cases. There is a strong correlation between the DAWBA ASD total impairment score and ADI-R algorithm score (in the whole sample, r = 0.82, p < 0.001; in cases, r = 0.55, p < 0.001; and in non-cases, r = 0.64, p < 0.001).

FIGURE 13.

Comparison of DAWBA and ADI-R. (a) Distribution of DAWBA ASD total impairment scores (cases n = 112); (b) distribution of DAWBA ASD total impairment scores (non-cases n = 101); (c) ADI-R algorithm total scores (cases n = 112); and (d) ADI-R algorithm total scores (non-cases n = 101). Adapted from McEwen et al. 3 © 2016 McEwen et al. ; licensee BioMed Central Ltd. This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Adapted from McEwen et al. 3 © 2016 McEwen et al. ; licensee BioMed Central Ltd. This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Prediction of ASD by Development and Well-Being Assessment computer-generated probability bands

Receiver operating characteristic curve analysis was used to determine how well DAWBA computer-generated probability bands predicted consensus research diagnosis of ASD (Figure 14). The area under the ROC curve was significant (0.90, 95% CI 0.87 to 0.94; p < 0.001) and the OR, corrected for relatedness of twins in each pair, showed a threefold increase in the odds of an ASD diagnosis for each increasing probability band (OR 3.39, 95% CI 2.60 to 4.42; p < 0.001).

FIGURE 14.

Receiver operating characteristic curve of DAWBA computer-generated probability band predicting consensus research diagnosis of ASD. Area under the curve = 0.90 (95% CI 0.87 to 0.94; p < 0.001). OR 3.39 (95% CI 2.60 to 4.42; p < 0.001) (robust standard error used, clustered by family to control for relatedness of twins in pairs). Adapted from McEwen et al. 3 © 2016 McEwen et al. ; licensee BioMed Central Ltd. This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Adapted from McEwen et al. 3 © 2016 McEwen et al. ; licensee BioMed Central Ltd. This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Receiver operating characteristic curve

Classification statistics for probability bands are shown in Table 36. The optimal cut-off point in this sample was low (3%) – taking those in this band or higher to be cases – with 86.1% of children correctly classified. This achieved a good balance between sensitivity (0.88) and specificity (0.85). The PPV of 0.80 indicated that 8 out of 10 children identified as cases at this cut-off point had a research diagnosis of ASD, whereas 2 out of 10 were false positives. The NPV of 0.91 showed that 9 out of 10 children identified as non-cases were truly non-cases, whereas 1 in 10 identified as non-cases were cases of ASD that had been missed.

Ability of DAWBA clinical rater-assigned diagnosis to predict ASD

The numbers of children rated as having autism or Asperger syndrome, other ASD, or not ASD by DAWBA clinical rater are cross-tabulated with consensus research diagnosis of ASD in Table 37. The majority, 86.9%, of children without a research diagnosis of ASD were correctly rated as non-cases by the DAWBA clinical rater.

Similarly, 85.7% of children with a research diagnosis of ASD were identified by the DAWBA clinical rater as having an ASD diagnosis. Table 35 shows the classification statistics for cases defined as those with any ASD diagnosis by DAWBA clinical rater, and for cases more narrowly defined as those with a diagnosis of childhood autism or Asperger syndrome. Using the broader category of any ASD diagnosis, there was only minimal improvement over computer probability bands in the overall percentage of children correctly classified, from 86.1% to 86.4%, and minimal change in the sensitivity, specificity, PPV and NPV. Using the narrower category of autism or Asperger syndrome, specificity improved to 0.98 and PPV to 0.95, meaning that virtually all non-cases were correctly classified and there were very few false positives. Conversely, sensitivity decreased to 0.54 and NPV to 0.76, meaning that only around half of the cases of ASD were identified and less confidence was warranted by a negative result.

Does combining the DAWBA with the ADOS improve the classification of cases?

In practice, assessment of children suspected of having ASD should include both a detailed developmental history and direct observation of the child using a tool such as the ADOS-G. Table 38 shows the classification statistics when DAWBA clinical rater-assigned diagnosis was used in conjunction with ADOS-G diagnostic status. Those with DAWBA and ADOS-G data were included (cases, n = 112; non-cases, n = 101; note that this does not include the low-risk non-cases, see Table 37). Children who met criteria according to the DAWBA and/or met criteria for ASD according to the ADOS were assumed to be positive for ASD, and this diagnosis was compared with the research diagnosis of ASD. The DAWBA was used at both the broader category of any ASD and the narrower category of autism and Asperger syndrome.

Results of this analysis are shown in Table 38. In summary:

• Using ADOS-G diagnosis in conjunction with a DAWBA diagnosis of any ASD increased the percentage correctly classified.

• Using ADOS-G diagnosis in conjunction with the narrower DAWBA diagnosis of autism or Asperger syndrome decreased the percentage correctly classified if cases were taken to be those positive on DAWBA and ADOS-G, but increased it if cases were taken to be those positive on DAWBA or ADOS-G.

• When a positive result on either the DAWBA or the ADOS-G was used to define cases, sensitivity was very high (nearly all cases were detected), but specificity was lower (some non-cases were incorrectly flagged as having ASD), and the PPV showed that one out of five children identified as cases were false positives.

• When cases were taken to be those with positive results for both DAWBA and ADOS-G, sensitivity was lower (some cases were missed), but specificity was very high (very few non-cases were flagged as ASD), and PPV was very high (there were very few false positives).

• Children negative on both the DAWBA and ADOS-G were very unlikely to have ASD.

Sensitivity, specificity, PPV and NPV for diagnosis assigned by DAWBA clinical rater in combination with ADOS-G diagnostic status (if met criteria for ASD or autism according to revised algorithm198)

The level of confidence in positive results is conveyed in Figure 15, which shows the percentage of true and false positives in cases identified through various combinations of DAWBA and ADOS-G results. For example, although a DAWBA rating of any ASD resulted in around 80% true positives, the proportion of true positives increased in those with a narrower DAWBA rating of Asperger syndrome or autism. The proportion also increased in those who met both DAWBA and ADOS-G criteria for ASD, and was 98% in those with a DAWBA rating of Asperger syndrome or autism and who met ADOS-G criteria for ASD.

FIGURE 15.

Percentage of true and false positives for cases identified using DAWBA and ADOS-G. Adapted from McEwen et al. 3 © 2016 McEwen et al. ; licensee BioMed Central Ltd. This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Adapted from McEwen et al. 3 © 2016 McEwen et al. ; licensee BioMed Central Ltd. This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

The level of confidence in negative results can be seen in Figure 16, which shows the percentage of true and false negatives using the DAWBA and ADOS-G. Nearly 20% of children rated as not having ASD using the DAWBA were false negatives. However, this dropped to only 3% for children who did not meet either DAWBA or ADOS-G criteria for ASD.

FIGURE 16.

Percentage of true and false negatives for cases identified using DAWBA and ADOS-G. Adapted from McEwen et al. 3 © 2016 McEwen et al. ; licensee BioMed Central Ltd. This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Adapted from McEwen et al. 3 © 2016 McEwen et al. ; licensee BioMed Central Ltd. This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Strengths and weaknesses

The strengths of this study include our sample being drawn from a large UK general population sample and so our findings should be reasonably representative of the types of cases seen in community/first opinion settings. The use of staged screening and assessment for ASD at multiple points through childhood and adolescence should also have ensured that cases from across the autism spectrum were detected, including those with more subtle difficulties that may not have been detected until late childhood.

However, our work also has limitations. For instance, the mothers in our sample had a higher level of education than the UK average (≈50% vs. ≈40% with A level or above), and this may have resulted in better-quality data (e.g. more detailed answers given to open questions). However, 50% of the mothers in our sample were of a lower educational level so, despite the slight over-representation of more educated mothers, all educational levels were included. It is therefore unlikely that this would have led to significant bias.

The use of a twin sample has the potential to introduce bias by (1) the relatedness of twins in each pair inflating correlations and (2) contrast effects, whereby parents contrast twins with differing levels of impairment and exaggerate the differences between them. We did control statistically for relatedness of twins in pairs to try to reduce bias introduced by twin similarity. It remains possible that parent rating bias could have exaggerated the difference between discordant pairs, increasing the contrast between affected and unaffected twins. However, in the case of concordant pairs of which one is less severely affected than the other, the less affected twin could be falsely rated by the parent as unaffected. Rater bias could, therefore, result in an increase or decrease in the apparent accuracy of the DAWBA and it is not clear to what extent, or in which direction, this could have acted on the results. However, the benefits of utilising this longitudinal study of a nationally representative cohort, from which results can be generalised to community health settings, are likely to outweigh the potential uncertainties in interpretation of twin data.

Although not necessarily a limitation, it is worth bearing in mind that, in many cases, the DAWBA was administered using telephone interview by RAs. This could have improved the quality of data collection beyond what might be expected if online administration had been used (e.g. interviewers may have asked for clarification or provided guidance if parents had misinterpreted questions). This does not invalidate the results, but it is possible that using an interviewer with some experience of ASD would result in better-quality data than online administration. This needs to be tested as online tools are potentially much more cost-effective than interviewers. Nonetheless, it is likely that some aspects of clinical interview will still be required to be undertaken by trained clinicians to confirm diagnosis.

Sample

We included 145 families consisting of adolescents and adults with ASD and their parents. Families were recruited through the patient’s clinical diagnosis of ASD from CAMHS, adult clinics, charities and research databases that form part of our clinical research networks. Clinical diagnosis of autism was confirmed using the ADI-R. 27 When a parent was unavailable or unwilling to complete the ADI-R, the ADOS-G197 was used.

Strengths and Difficulties Questionnaire

The SDQ is a brief, 25-item questionnaire that can be administered as self-report, or to parents and teachers of children and adolescents. Information about the SDQ is available online (URL: www.sdqinfo.com). It covers common areas of emotional and behavioural difficulties and also asks whether or not the informant thinks that the young person has a problem in these areas. It includes five subscales (emotional symptoms, conduct problems, hyperactivity, peer problems and prosocial scales) and an impact score. Scores for each subscale are classified as ‘normal’, ‘borderline’ or ‘abnormal’. Previously validated computerised algorithms were used to predict psychiatric disorder by bringing together information on symptoms and impact from the SDQ completed by multiple informants. The algorithm makes separate predictions for three groups of disorders: (1) conduct-oppositional disorders, (2) hyperactivity–inattention disorders and (3) anxiety–depressive disorder. Each is predicted to be ‘unlikely’, ‘possible’ or ‘probable’. Predictions from these groups are combined to generate a single prediction of the presence of any psychiatric disorder. For a prediction of probable hyperactivity–inattention disorder it is essential to have evidence of pervasiveness through both parent and teacher ratings. As only parent- and self-rated SDQ were available in this study, only unlikely or possible predictions could result from these data.

Psychiatric diagnosis

As noted previously (see Chapter 15), DAWBA51 is a detailed computer-based psychiatric interview which can be completed by parents and patients. Each section of the DAWBA uses skip rules, which start with structured questions that cover the operationalised diagnostic criteria of the DSM-IV. 206 A computer algorithm uses the responses from both informants to generate probability bands for the presence of particular psychiatric morbidity. A clinical rater then reviews all information and assigns a diagnosis for each disorder. Where available, clinical notes were reviewed and the presence or absence of psychiatric disorders was recorded. A patient was considered to be at high risk of particular clinical disorders if they had been assigned a diagnosis by a DAWBA clinical rater, or if they had received a diagnosis in the clinic where they were assessed.

Screening questionnaires

In addition to the DAWBA, we assessed the external validity of the SDQ against standard screening questionnaires and diagnostic measures. Symptoms of ADHD were measured using the BAARS-IV. 207 The BAARS-IV is a commonly used, validated 18-item rating scale of DSM-IV criteria for ADHD.

Anxiety and depression were measured by the relevant subscales of the Hospital Anxiety and Depression Scale (HADS). 208 The HADS is a commonly used, 14-item screening tool for anxiety and depression. It has demonstrated excellent psychometric properties and is used routinely in clinical practice.

The Obsessive–Compulsive Inventory – Revised (OCI-R)209 was used to measure symptoms of OCD. The OCI-R is an 18-item, self-report measure with excellent psychometric properties. 209

Data analysis

Internal reliability was measured by examining the inter-rater reliability, construct validity and temporal stability of the measure. Inter-rater reliability was tested by examining the correlation (Pearson’s r) of total SDQ scores between parent and respondent. Construct validity was tested using multitrait–multimethod analysis (MTMM). MTMMs are based on producing a correlation matrix of multiple traits (e.g. the subscales of the SDQ) measured by multiple methods (e.g. parent and respondent). Construct validity is assessed through comparisons across raters and across subscales. For example, correlations between parent and respondent emotional subscale (a convergent correlation coefficient) would be expected to be higher than between parent emotional and respondent hyperactivity subscales (a discriminant correlation coefficient). Where poor convergent and discriminant correlations are observed, it is inferred that these subscales may not be tapping into the same, distinct constructs across raters. MTMM was performed using Spearman’s correlations calculated in SPSS for Windows version 20. We also present the Cronbach’s alpha for each subscale according to each rater.

Temporal stability was tested by calculating the Pearson’s correlation between baseline and follow-up for each SDQ subscale and the total SDQ score for each rater.

External validity was tested initially by examining the correlation (Spearman’s r) between SDQ subscale scores and other measures of psychiatric disorder, namely DAWBA probability bands, clinical diagnosis (coded as no, possible or definite diagnosis) and standard screening questionnaires. It was hypothesised that SDQ emotional problems would be associated with measures of anxiety disorders, depression and OCD, and that SDQ hyperactivity would be associated with measures of ADHD. Correlations were repeated, controlling for SDQ impact scores (partial r), to ensure that associations were not driven by severity of symptoms (e.g. patients with more severe ASD symptoms being rated highly across SDQ subscales).

Parent- and self-rated SDQ emotional problems and hyperactivity subscales, coded into normal, borderline and abnormal bands, were compared with risk of clinical disorder. Borderline and abnormal bands were collapsed into one category and taken to indicate higher risk. Multi-informant predictive algorithms were also used, with possible and probable categories collapsed for emotional disorders (probable category was not available for hyperactivity, see Strengths and Difficulties Questionnaire). Risk of clinical disorder was measured using DAWBA clinical rater-assigned diagnoses and evidence of diagnoses from clinical notes. Sensitivity, specificity, PPV and NPV were calculated for each rater and for the predictive algorithm. The overall percentage of individuals correctly classified and the OR was also calculated in each case.

Inter-rater correlations

Pearson’s correlations between parent and respondent total SDQ scores were 0.418. This is comparable with the 0.48 inter-rater correlation between parents and youth previously reported in a large-scale epidemiological sample on the SDQ. 210

Construct validity of the Strengths and Difficulties Questionnaire subscales across informants

Table 39 presents a MTMM of the five hypothesised SDQ subscales. The Cronbach’s alpha coefficients ranged between 0.484 and 0.810. The cross-method correlations of the same traits are shaded green; all were significantly different from zero (p < 0.001) and in the moderate magnitude range (0.423–0.573).

In the majority of cases, the convergent correlations were significantly larger than the other correlation coefficients in the same row or column (the discriminant correlations). However, there are some notable exceptions. First, in both informant-rated pairs, the emotional problems subscale demonstrated poor discriminant validity relative to peer problems and parent-rated behavioural problems (relevant cells are in bold). Second, the youth-rated behavioural subscale did not show good discriminant validity relative to the parent-reported hyperactivity subscale and the parent-rated behavioural subscale, which did not show good discriminant validity relative to the youth-reported emotional subscale (relevant cells italicised).

The behavioural, emotional and peer problems subscales therefore demonstrate poor discriminant validity; thus, these findings suggest that the five subscales do not tap into the same, distinct aspects of the youths mental health problems across both informants. However, the three-factor (internalising–externalising–prosocial) contrast provided much better convergent and discriminant validity. Furthermore, Cronbach’s alphas for each subscale were higher on average in the three-factor model (Table 40).

Temporal stability

The SDQ was re-administered to parents and youths after an interval of ≈11 months (parents, n = 69; youths, n = 35). The mean retest stability was 0.64. Stability did not differ by informant (parent = 0.62; youth = 0.66) (Table 41).

External validity

Tables 42 and 43 show correlations between SDQ subscales and other measures of comorbid disorders (DAWBA probability bands, clinical diagnosis and standard screening questionnaires). Both parent- and self-report SDQ scores showed the predicted pattern of associations for emotional symptoms (correlating with anxiety disorders, OCD and depression) and hyperactivity (correlating with ADHD). This was true even when the effects of SDQ Impact scores were controlled (using partial correlations), suggesting that these associations are not being driven by severity of problems regardless of domain.

Parent- and self-rated SDQ emotional problems and hyperactivity subscales, coded into normal, borderline and abnormal bands, were then compared with risk of clinical disorder.

Emotional disorder

Figure 17 shows ratings for SDQ emotional problems or disorder in those at high and low risk of clinical diagnosis of any emotional disorder. For both parent- and self-rated emotional problems, more than half of those rated as borderline or abnormal were at high risk of emotional disorder and the majority of those rated as normal were at low risk of emotional disorder. Parents were more likely to give abnormal ratings, resulting in more true and false positives, and patients were more likely to rate themselves as normal, resulting in more true and false negatives.

FIGURE 17.

Strengths and Difficulties Questionnaire emotional problems (parent and self-rated) and emotional disorder (multi-informant predictive algorithm) scores in patients at high or low risk of clinical diagnosis of any emotional disorder.

Table 44 shows the sensitivity, specificity, PPV and NPV of each SDQ subscale in predicting risk of relevant clinical disorders. Parent-rated emotional problems had high sensitivity (90% of cases would have been detected), whereas only around half (55%) of cases would have been detected using self-ratings. The PPV showed that 60–70% of cases identified were true positives and 30–40% were false positives. The NPV showed that 60–75% of normal ratings reflected low risk of disorder, whereas 25–40% represented missed cases. The odds of clinical disorder were significantly elevated in those identified using the SDQ, using parent or self-ratings. The multi-informant predictive algorithm performed similarly to parent-rated SDQ, with only slightly improved ability to correctly identify risk of clinical disorder.

Hyperactivity

A similar pattern was observed for SDQ hyperactivity. Figure 18 and Table 44 show that the majority (60–65%) of those with borderline or abnormal ratings were at high risk of ADHD and the majority (67–84%) of those with normal ratings were at low risk. ORs were significant for parent ratings, self-ratings and multi-informant predictive algorithm. Parent ratings detected more cases (86%) than self-ratings (60%).

FIGURE 18.

Strengths and Difficulties Questionnaire hyperactivity (parent and self-rated) and hyperactivity disorder (multi-informant predictive algorithm) scores in patients at high or low risk of clinical diagnosis of any type of ADHD/hyperkinesis. NB. Probable band for multi-informant predictive algorithm requires evidence of pervasiveness across settings from parent and teacher SDQ, which was not available in this sample.

Strengths and weaknesses

The strengths of this study include our comparison of the SDQ with the some of the most widely used screening tools for psychopathology in the UK (BAARS-IV, HADS and the OCI-R). In addition, we compared the SDQ against clinical diagnoses assigned by chartered psychiatrists.

However, our work also has limitations. Our sample primarily consisted of high-functioning individuals, which is not representative of ASD as a whole. For example, approximately 8% of our sample was diagnosed with a LD, in comparison with an estimated prevalence of 56%. 211 Thus, we can generalise our findings only to higher-functioning adults with ASD.

Data used to test external validity were limited in various ways. For example, it was clear from clinical notes that, despite having evidence of comorbid emotional disorder or hyperactivity, some patients had not received further assessment. This means that the absence of clinical diagnosis in these patients did not necessarily reflect the absence of disorder. Furthermore, in some patients the time gap between assessment with the SDQ and with other measures (screening questionnaires, clinical assessment) was relatively long. Consequently, disagreement between the SDQ and other measures may have reflected genuine change in mental state rather than poor performance of the SDQ. Despite these limitations of the clinical data, a consistent pattern of significant associations emerged across measures. Indeed, it is possible that these associations would have been stronger with better-quality data.

Participants

Participants included 158 individuals consecutively assessed for an ASD diagnosis in a specialist national tertiary ASD diagnostic clinic for adults between January and May 2011. The clinic is situated within the South London and Maudsley NHS Foundation Trust. People are typically referred by their local family physician/GP or consultant psychiatrist for a second opinion. In eight cases diagnosis was inconclusive due to a history of acquired head injury or the presence of severe psychotic symptoms during assessment. Data from these cases were excluded from the study. The remaining 150 participants were aged 18–65 years, with a mean age of 31 years. There were 110 males (mean age 32 years) and 40 females (mean age 31 years). Seventy-three patients already had a diagnosis of a mental health condition, although only seven of these had previously been diagnosed with ASD.

Clinical assessment

Assessment included a detailed psychiatric assessment using ICD-10 research diagnostic criteria and, when possible (i.e. when parents were available, able and willing), an ADI-R. In the event that no parent was available for the ADI-R, we asked the person seeking diagnosis to undergo the ADOS-G. In some cases both assessment tools were required to gather enough relevant information. Seventy-one individuals were assessed using the ADI-R, 62 were assessed with the ADOS-G and 17 were assessed using both the ADI-R and the ADOS-G.

All information obtained was compiled by the multidisciplinary clinical team – a consultant psychiatrist, a junior doctor and an ADI-R/ADOS-G administrator (nurse or psychologist) – who together decided whether or not each criterion on the ICD-10 algorithm was fulfilled. If a patient met the full ICD-10 criteria (a total of at least six symptoms must be present – either currently or by history – with at least two from the first domain and one from each of the second and third domains) and the symptoms were noted before the age of 3 years, they were diagnosed with childhood autism (if they exhibited a language delay) or Asperger syndrome (if there was no evidence of a language delay). In line with ICD-10 guidelines, if a patient exhibited some autistic symptoms but did not meet full ICD-10 diagnostic criteria they were diagnosed with PDD, unspecified (PDD-unspecified), or atypical autism. For the purposes of this study these two subthreshold diagnostic groups were collapsed into a single ‘PDD-unspecified’ group. In some cases, it was not possible to decide confidently whether or not a symptom was present due to a lack of information, or because information obtained from patient and parent was contradictory. In this event the criterion was coded as ‘unclear’ and the team made the clinical diagnostic decision based on the gestalt of the information received. Of the 150 consecutive assessments, 113 were diagnosed with an ASD using ICD-10 criteria. Of these, 28 participants were subtyped as having childhood autism, 48 Asperger syndrome and 37 PDD-unspecified.

Additional mental health conditions were also diagnosed in accordance with the ICD-10 (with the exception of adult ADHD which, in keeping with UK guidelines, was assessed using DSM-IV-TR). Supplementary self-report questionnaires were used to help inform the assessment of OCD, ADHD, depression and anxiety. These were (respectively) the OCI-R,209 the Barkley Current and Childhood Symptom Scales44 and the HADS. 208

Where there was clinical suspicion that a participant might have ID (F70–73 in ICD-10) or a significant lacuna in a neuropsychological function, they were referred for testing of general intellectual and executive functioning. These participants (n = 35) were tested using the Wechsler Adult Intelligence Scale-III,228 which revealed a mean performance IQ of 87 (SD 16, range 55–132) and a mean verbal IQ of 95 (SD 18, range 60–133). Mean full-scale IQ could not be calculated in 16 participants due to large discrepancies between performance IQ and verbal IQ. In the remaining 19 participants mean full-scale IQ was 90 (SD 17, range 53–129) and only two participants had an IQ of < 70. The remaining 115 participants, not referred for such testing, were estimated to be in the normal range of general intellectual function based on education attainment, employment and informant report.

Procedure

For each participant the diagnostic outcome (ASD/not ASD; ICD-10 subtype of ASD; ADI-R and ADOS-G scores; presence of additional mental health problems), was reviewed by the research team. Information from the ICD-10 algorithm was used to determine whether or not each criterion on the proposed DSM-V algorithm would be satisfied and this was supplemented by anonymised reports from the ADI-R, ADOS-G and the psychiatric interview. Each criterion could be coded as ‘yes’, ‘no’ or ‘unclear’. A participant was considered to meet criteria for ASD on the DSM-V only when all three criteria in ‘A’ and at least two out of four criteria in ‘B’ were coded as ‘yes’, as suggested in the criteria last posted by the American Psychiatric Association. If a participant did not meet criteria for ASD on the DSM-V it was determined whether or not they would meet criteria for the alternative diagnosis of SCD (using last posted draft criteria). Recoding was completed by pairs of researchers; and for 40 sets of participant data the recoding was reviewed at consensus meetings with the whole team (10 researchers) to ensure agreement.

Data were also recoded to complete the DSM-IV-TR algorithm, on which participants could be diagnosed as either ‘ASD’ (autistic disorder/Asperger syndrome) if they fulfilled at least six criteria, with at least two in domain A and one from each of domains B and C, or ‘not ASD’.

Factors affecting agreement between ICD-10, DSM-IV-TR and DSM-V were also investigated. Participant characteristics (age, sex, IQ, additional mental health conditions) were compared between ASD-positive and ASD-negative groups. The effect of altering the DSM-V algorithm was examined in two ways: (1) by relaxing the number of criterion required for diagnosis and (2) by considering criteria that were coded as ‘unclear’ as either met or not met. To examine the effect of relaxing the number of criteria required for diagnosis we reduced the thresholds in criteria A (from three to two), or in criteria B (from two to one), or both. To examine the effect of including or excluding the ‘unclear’ items we recoded the DSM-V algorithm by considering ‘unclear’ items to be present. This was relevant because the DSM-V allows criteria to be met by history and allowing or disallowing the ‘unclear’ criteria is likely to be crucial in many adult cases where multiple informants are not available.

We hypothesised that (1) prevalence of childhood autism or Asperger syndrome diagnosed using ICD-10 criteria would be similar to that using DSM-IV-TR and DSM-V, (2) most participants diagnosed with ASD on the ICD-10 but not the DSM-V would be diagnosed with SCD and (3) altering the DSM-V algorithm would have significant effects on the rate of positive DSM-V ASD diagnosis.

Conclusions of initial diagnostic assessments

Factors affecting agreement between ICD-10 and DSM-V

Participant characteristics: age, sex, intelligence quotient, diagnostic subtype and additional mental health conditions

Among the participants who were ASD positive using the ICD-10, there were no differences with respect to age, sex or IQ (where available) between those individuals who were ASD positive and those who were ASD negative on the DSM-V.

There was, however, a significant difference in rate of DSM-V ASD-positive diagnosis between ICD-10 subtypes (χ² = 31.58; p < 0.001). Significantly more participants diagnosed with ICD-10 childhood autism or Asperger syndrome met DSM-V criteria for ASD than those in the PDD-unspecified group (Table 47, column 1). The difference between ICD-10 defined childhood autism and Asperger syndrome was not significant (χ² = 1.64; p = 0.2).

TABLE 47 - Percentage of participants who would be diagnosed with (A) ASD on DSM-V, (B) SCD on DSM-V or (C) ASD on DSM-V if the number of criteria A required was reduced from three to two and/or the number of criteria B required was reduced from two to one

Difference between full criteria and relaxed threshold: McNemar’s p < 0.001.

Difference between full criteria and relaxed threshold: McNemar’s p < 0.05.

ICD-10/DSM-IV-TR diagnosis	A	B	C
	DSM-V ASD	DSM-V SCD	DSM-V ASD: relax A	DSM-V ASD: relax B	DSM-V ASD: relax A and B
ICD-10, % (n)
Not ASD (total n = 37)	0	0	0	0	0
ASD (total n = 113)	56 (63)	19 (21)	69 (78)a	70 (79)a	87 (98)a
Childhood autism (total n = 28)	82 (23)	11 (3)	86 (24)	93 (26)	96 (27)
Asperger syndrome (total n = 48)	69 (33)	15 (7)	83 (40)b	83 (40)b	98 (47)a
PDD-unspecified (total n = 37)	19 (7)	30 (11)	38 (14)b	35 (13)b	65 (24)a
DSM-IV-TR, % (n)
Not ASD (total n = 70)	3 (2)	13 (9)	14 (10)b	10 (7)b	27 (19)a
ASD (total n = 80)	77 (61)	15 (12)	85 (68)b	90 (72)a	99 (79)a

With respect to additional mental health conditions, a significant difference was found only for OCD: higher rates of OCD were found in the group that were ASD positive on the DSM-V than in the group that were ASD positive only on the ICD-10 (χ² = 4.58; p = 0.03; Figure 19).

FIGURE 19.

Participants meeting criteria for mental health conditions. *p < 0.05, **p < 0.01. ‘None’, no additional/alternative mental health condition.

Strengths and weaknesses

Although this is the first study to examine the draft DSM-V criteria in able adults with ASD, some limitations should be noted. Like other studies comparing existing criteria with the DSM-V draft, we analysed existing clinical notes and instruments that were used primarily for allocating current (ICD-10) diagnostic categories. It remains to be seen whether or not using the DSM-V criteria in the clinic during assessment in a prospective fashion would result in different findings. For example, sensory sensitivities are not part of current clinical criteria and might be more thoroughly assessed in future in clinics where DSM-V criteria are adopted. Nonetheless, one potentially useful feature of the present study, in contrast to many others, is the relatively large number of participants (25%) who were assessed for ASD but found to be negative using current criteria. This allowed examination of specificity of the different diagnostic measures. Finally, it should be noted, of course, that it remains unclear whether or not existing ICD-10 or DSM-IV-TR criteria should be considered the ‘gold standard’ against which new criteria are compared; in the absence of biomarkers, the exact definition of ASD and its boundaries remains a matter for debate.

Participants

Twenty control adults were recruited locally by advertisement and 20 adults with ASD were recruited through a clinical research programme at the Maudsley Hospital/IoP, London. All volunteers (Table 48) gave informed consent (as approved by the IoP and Bethlem and Maudsley Hospital Trust REC), and had a full-scale IQ of > 75 (WASI244).

All volunteers were right-handed males and were between 20 and 68 years of age. None had a history of major psychiatric disorder or medical illness affecting brain function (e.g. psychosis or epilepsy). All participants underwent a psychiatric interview and physical examination. Blood tests were used to exclude biochemical, haematological or chromosomal abnormalities (including fragile X syndrome). All participants with ASD were diagnosed with autism according to ICD-10 research criteria245 by a trained and qualified clinician in our adult autism specialist clinic at the South London and Maudsley Hospital. The initial clinical diagnosis was then confirmed using the ADI-R,27 whenever possible (i.e. where parents agreed to take part, n = 17). In 3 out of the 20 included cases, ADI-R scores could not be obtained because informants were unavailable. In these three cases, the clinical diagnosis was confirmed using the ADOS. 196 Both ADI-R and ADOS-G scores were available for two participants. All cases had to reach the ADOS-G or ADI-R algorithm cut-off points in the three domains of impaired reciprocal social interaction, communication and repetitive behaviours and stereotyped patterns, but we did allow failure to reach cut-off point in one of the domains (by one point). Sixteen ASD individuals did not have a delay in development of phrase speech at the age of 36 months (and so may be subtyped by some as having the autistic subtype of Asperger syndrome); the remaining four individuals had a history of delayed phrase speech (and so may be subtyped by some as having the autistic subtype of high-functioning autism). However, due to the small sample size it was not possible to reliably investigate putative differences between people with HFA and Asperger syndrome, and all subjects were analysed in a combined group of individuals with ASD.

Furthermore, we recruited a group of 19 individuals with ADHD from the adult ADHD services at the Maudsley Hospital, London. This group served as a neurodevelopmental control group and was matched to the ASD group in sex, age (mean 31 ± 8.5 years), full-scale IQ (mean 1.7 ± 6) and handedness. Of these, eight individuals fulfilled the criteria for ADHD combined type and 11 for ADHD inattentive type. The diagnosis of ADHD was made by a trained and qualified clinician based on a structured clinical interview according to DSM-IV criteria. ADHD symptoms were also assessed using Barkley Current and Childhood Symptoms Scales207 and the Wender Utah Rating Scale. 246

Magnetic resonance imaging data acquisition

Magnetic resonance imaging (MRI) data were acquired using a 1.5-T GE Signa Neuro-optimized System (General Electric Medical Systems, Milwaukee, WI, USA) fitted with 40 mT/m high-speed gradients at the Maudsley Hospital, London. Whole-brain spoiled gradient recalled acquisition in the steady-state T1-weighted series were collected in the coronal plane with repetition time = 13.8 months, echo time = 2.8 months, yielding 124 contiguous 1.5-mm² axial slices of 256 × 192 voxels with an in-plane resolution of 1 mm². Similar T1-weighted MRI scans were acquired for the ADHD group on a 1.5-T GE Signa Neuro-optimized System at the Maudsley Hospital, London, using repetition time = 10.72 months, echo time = 4.86 months, yielding 146 contiguous 1.1-mm² axial slices. A quadrature birdcage head coil was used for radiofrequency transmission and reception. Foam padding and a forehead strap were used to limit head motion. All scans were visually inspected by a radiologist to assess (1) image contrast, (2) movement artefacts and (3) the existence of clinical abnormalities. Scans displaying low image quality or clinical abnormalities were excluded from this study.

Image processing

Cortical reconstruction and volumetric segmentation was performed with the FreeSurfer image analysis suite (Laboratory for Computational Neuroimaging, Charlestown, MA, USA), which is documented and freely available for download online [URL: http://surfer.nmr.mgh.harvard.edu/ (accessed 31 May 2018)]. The technical details of these procedures are described in prior publications. 247–254 Briefly, this processing includes motion correction and averaging of multiple volumetric T1-weighted images (when more than one is available), removal of non-brain tissue using a hybrid watershed/surface deformation procedure,249 automated Talairach transformation, segmentation of the subcortical white matter and deep grey matter volumetric structures (including hippocampus, amygdala, caudate, putamen, ventricles),251,252 intensity normalisation, tessellation of the grey matter–white matter boundary, automated topology correction,255,256 and surface deformation following intensity gradients to optimally place the grey/white and grey/cerebrospinal fluid borders at the location where the greatest shift in intensity defines the transition to the other tissue class. 247 Once the cortical models are complete, a number of deformable procedures can be performed for further data processing and analysis, including surface inflation,247 registration to a spherical atlas (which utilised individual cortical folding patterns to match cortical geometry across subjects248), parcellation of the cerebral cortex into units based on gyral and sulcal structure253,257 and creation of a variety of surface-based data, including maps of curvature and sulcal depth. This method uses both intensity and continuity information from the entire three-dimensional magnetic resonance volume in segmentation and deformation procedures to produce representations of cortical thickness, calculated as the closest distance from the grey/white boundary to the grey/cerebral spinal fluid boundary at each vertex on the tessellated surface. 250 The maps are created using spatial intensity gradients across tissue classes and are therefore not simply reliant on absolute signal intensity. The maps produced are not restricted to the voxel resolution of the original data, thus are capable of detecting submillimetre differences between groups. Procedures for the measurement of cortical thickness have been validated against histological analysis258 and manual measurements. 259,260 FreeSurfer morphometric procedures have been demonstrated to show good test–retest reliability across scanner manufacturers and across field strengths. 261 All reconstructed surfaces were visually inspected for gross anatomical topological defects. Brains that did not reconstruct or showed geometric inaccuracies were excluded from the study.

A set of five morphometric parameters per vertex were used as input to the multimodal classifier. Three parameters [1–3: average convexity or concavity, mean (radial) curvature and metric distortion] accounted for geometric features at each cerebral vertex and two parameters (4–5: cortical thickness and surface area, respectively), measured volumetric features.

The average convexity or concavity was used for quantifying the primary folding pattern of a surface. Convexity/concavity captures large-scale geometric features and is insensitive to noise in the form of small wrinkles in a surface. 248 At each vertex, convexity/concavity indicates the depth/height above the average surface and measures sulcal depth or gyral height, respectively. Differences in sulcal depth have previously been investigated as one important aspect of the cerebral geometry implicated in ASD238 and have been suggested to reflect abnormal pattern of cortical connectivity. Mean (radial) curvature was used to assess folding of the small secondary and tertiary folds in the surface. The selection of this feature was motivated by early findings of polymicrogyria (i.e. excessive number of small convolutions on the surface of the brain) in autism. 262 These would not be captured by the average convexity measure reflecting large-scale geometric features only. Metric distortion (i.e. Jacobian) indicating the degree of cortical folding was calculated as degree of displacement and convolution of the cortical surface relative to the average template. 248,263 Although sulcal depth and radial curvature measure specific aspects of the cortical geometry, metric distortion is a wider measure of the overall degree of cortical folding and thus captures geometric distortions otherwise not specified. There is a large body of evidence to suggest that individuals with ASD display abnormal patterns of cortical gyrification reflecting cerebral development and connectivity (e.g. Levitt and colleagues239 and Hardan and colleagues264). Cortical thickness (see above) and pial area (i.e. the area of a vertex on the grey matter surface, calculated as the average of the area of the triangles touching that vertex) were used to quantify volumetric differences. Although these two features are generally combined to measure regional brain volume, a recent study has shown that cortical thickness and surface are influenced by distinct genetic mechanism240 and we therefore added as separate features into the model. A summary of the set of parameters is displayed in Figure 20.

FIGURE 20.

Summary of the five morphometric parameters measured at each cerebral vertex. (a), Average convexity; (b) cortical thickness; (c) pial area; (d) metric distortion (Jacobian); and (e) mean (radial) curvature. g/w, grey/white. Adapted from Ecker et al. 4 © 2010 Ecker et al. ; licensee BioMed Central Ltd. This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Adapted from Ecker et al. 4 © 2010 Ecker et al. ; licensee BioMed Central Ltd. This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

These included average convexity, cortical thickness, pial area, metric distortion (Jacobian) and mean (radial) curvature.

Group differences in intracranial volume, total brain volume and grey matter volume as estimated by FreeSurfer were assessed using t-tests for independent samples prior to classification. There were no significant differences between groups in any of these parameters at a level of p < 0.05 (Table 49). There were also no significant differences in total brain volume between controls and individuals with ADHD, nor between ASD and ADHD group on p < 0.05. Bartlett’s test for homogeneity of variances was used to examine parameter variability across homologue regions in different hemispheres using a threshold of p < 0.0014 (corrected for multiple comparisons).

Classification and support vector machine

A linear SVM was used to classify between patient and control group on the basis of their brain morphology. SVM has previously been applied to MRI data241–243 and a detailed description can be found in Schoelkopf and Smola265 and Burges. 266 Briefly, SVM is a supervised multivariate classification method that treats each image as a point in a high-dimensional space. If SVM is applied to images coming from different modalities, as in the current study, the number of dimensions equals the number of voxels/vertices per image multiplied by the number of modalities. Input data were then classified into two classes (e.g. individuals with ASD and controls), by identifying a separating hyperplane or decision boundary. The algorithm is initially trained on a subset of the data <x,c> to find a hyperplane that best separates the input space according to the class labels c (e.g. –1 for patients, +1 for controls). Here, x represents the input data (i.e. feature vector). The feature vector was generated by concatenating the five image modalities for each subject. Once the decision function is learned from the training data, it can be used to predict the class of a new test example. SVM267 is a maximum margin classifier, which identifies the optimal hyperplane by finding the hyperplane with the maximum margin (i.e. maximal separation between classes). The margin is the distance from the separating hyperplane to the closest training examples. The training examples that lie on the margin are called support vectors. The hyperplane is defined by a weight vector and an offset. The weight vector is a linear combination of the support vectors and is normal to the hyperplane. The linear kernel SVM used in the present study allows direct extraction of the weight vector as an image (i.e. the SVM discrimination map). A parameter C, that controls the trade-off between having zero training errors and allowing misclassifications, was fixed at C = 1 for all cases (default value). The LIBSVM toolbox for MATLAB (ACM Transactions on Intelligent Systems and Technology, Taiwan) was used to perform the classifications [URL: www.csie.ntu.edu.tw/∼cjlin/libsvm/ (accessed 31 May 2018)].

Discrimination maps

The weight vector has the same dimension as the feature vector and is normal to the hyperplane. It can be thought of as a spatial representation of the decision boundary and thus represents a map of the most discriminating regions. Here, the feature vector had n × m dimensions, where n is the number of vertices and m denotes the number of modalities (i.e. five). Given two groups (ASD vs. controls), with the labels +1 and –1, respectively, a positive value in the discrimination map (red/yellow colour scale) indicates relatively higher parameter values in patients than in controls with respect to the hyperplane, and a negative weight (blue/cyan colour scale) means relatively higher parameter values in controls than in patients with respect to the hyperplane. As the classifier is multivariate by nature, the combination of all voxels as a whole is identified as a global spatial pattern by which the groups differ (i.e. the discriminating pattern). To enable visualisation of the discriminating pattern for each image modality, the weight vector was cut into its constituent parts, which were then mapped back onto the average white matter surface. In the present study we coloured all voxels that have values > 30% of the maximum value of the discrimination map. 241,268 This threshold, although ultimately arbitrary, eliminates noise components predominantly (< 30%), thus enabling a better visualisation of the most discriminating regions (31–100%).

Intraregional morphometric profiles

To identify the relative contribution of specific parameters in discriminating between groups at different locations on the cortical surface, we displayed so called intraregional morphometric profiles. These profiles were derived by calculating the average weights of vertices within regions of interest (ROI) for the five different parameters. ROIs were based on contiguous weight clusters from the overall discrimination maps. The choice of the ROIs, although ultimately arbitrary, was motivated by (1) the relevance of a region to the current literature (e.g. sulcal depth differences in intraparietal sulcus238) and (2) generally high parameter weights for a specific morphometric feature (e.g. high weights for cortical thickness in medial temporal sulcus). The ROI analysis aimed to illustrate that different regions can display distinct differences in one or more parameters, rather than displaying the same pattern of feature weights. As all weights were scaled, means and SDs are directly comparable across regions and parameters.

Cross-validation and permutation testing

The performance of the classifier was validated using the commonly used leave-two-out cross-validation approach. This validation approach provides robust parameter estimates particularly for smaller samples. In each trial observations from all but one subject from each group were used to train the classifier. Subsequently, the class assignment of the test subjects was calculated during the test phase. This procedure was repeated S = 20 times (where S is the number of subjects per group), each time leaving observations from a different subject from each group out. The accuracy of the classifier was measured by the proportion of observations that were correctly classified into patient or control group. We also quantified the sensitivity and specificity of the classifier defined as:

where TP is the number of true positives (i.e. the number of patient images correctly classified); TN is the number of true negatives (i.e. number of control images correctly classified); FP is the number of false positives (i.e. number of controls images classified as patients); and FN is the number of false negatives (i.e. number of patients images classified as controls).

Classifications were made, first, by including all five morphological modalities into the feature vector and, second, on the basis of each individual parameter. Different classifiers were trained for each hemisphere. This enabled us to assess the overall classification accuracies for individual parameters and hemispheres. Classifier performance was evaluated using basic ROC graphs as well as permutation testing. Permutation testing can be used to evaluate the probability of getting specificity and sensitivity values higher than the ones obtained during the cross-validation procedure by chance. We permuted the labels 1000 times without replacement, each time randomly assigning patient and control labels to each image and repeated the cross-validation procedure. We then counted the number of times the specificity and sensitivity for the permuted labels were higher than the ones obtained for the real labels. Dividing this number by 1000 we derived a p-value for the classification.

Finally, the established classifier was used to predict group membership of individuals with ADHD. To achieve this, the classifier was first retrained using all ASD–control pairs to provide the best possible prediction model for discriminating the two groups. This trained classifier was then applied to the multiparameter data from the ADHD group and predictions for group membership for these subjects were derived.

To validate the binary classification of the subject groups on a quantitative level and to identify the degree to which the classification is driven by autistic symptoms rather than confounds unrelated to autism, the test margin for each subject coming from the all included classifier was correlated with the level of symptom severity measured by the ADI-R subscales. A similar approach has previously been employed by Ecker et al. 269

Overall classifier performance

Classification accuracies as well as sensitivity and specificity for each classifier are listed in Table 50. On the whole, strong hemispheric asymmetry was observed with regard to the overall classification accuracy. The left hemisphere provided consistently higher and above chance prediction accuracies across all morphometric parameters. Here, individuals with ASD were correctly assigned to the appropriate diagnostic category in 85.0% of all cases when all parameters were considered simultaneously (Figure 21c). The sensitivity of the multiparameter classification in the left hemisphere was 90.0% (i.e. if a volunteer had a clinical diagnosis of ASD, the probability that this participant was correctly assigned to the ASD category was 0.9). The specificity was 80.0%, meaning that 80.0% of the controls subjects were correctly classified as controls. High classification accuracies were also obtained when individual morphological parameters were considered. Figure 21a shows the receiver operator characteristics or ROC graph for the six classifiers in the left hemisphere. In general, one point in ROC space is better than another if it is to the north-west (true-positive rate is high, false-positive rate is low, or both), with the point (0,1) representing perfect classification. Classifiers appearing on the left-hand side of the ROC graph, near the x-axis, may be considered ‘conservative’ (i.e. make positive classifications only with strong evidence but often have low true-positive rates), whereas classifiers on the upper right-hand side of the graph may be thought of as ‘liberal’ (i.e. make positive classifications with weak evidence so they classify nearly all positives correctly but often have high false positives). The diagonal line y = x represents the strategy of randomly guessing a class. Best discrimination was obtained when cortical thickness measures were used to classify between groups with sensitivity and specificity values as high as 90.0%, followed by a classification on the basis of the metric distortion parameter (sensitivity 80.0%, specificity 80.0%). The average convexity, pial area and mean curvature displayed accuracies in the range of 70.0–80.0%. The classification p-value resulting from the permutation test was very low across, as well as within, modalities (< 0.004). The probability of obtaining specificity and sensitivity values higher than the ones obtained during cross-validation procedure by chance is thus extremely low.

FIGURE 21.

Receiver operating characteristic graphs for classifiers in the left hemisphere and right hemisphere. (a) Left hemisphere; (b) right hemisphere; (c) classification plots for the left hemisphere; and (d) classification plots for the right hemisphere. A, all parameters; B, cortical thickness; C, metric distortion/Jacobian; D, average convexity; E, pial area; F, mean (radial) curvature. Adapted from Ecker et al. 4 © 2010 Ecker et al. ; licensee BioMed Central Ltd. This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Adapted from Ecker et al. 4 © 2010 Ecker et al. ; licensee BioMed Central Ltd. This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

A very similar profile of parameter importance was observed in the right hemisphere – despite it having generally lower classification accuracies. Here, individuals with ASD were correctly assigned to the appropriate diagnostic category in 65.0% of all cases (sensitivity 60.0%, specificity 70.0%; p < 0.03), when all parameters were considered simultaneously (see Figure 21a and d). As in the left hemisphere, cortical thickness as well as metric distortion displayed the best classification performance, although only cortical thickness measures reached statistical significance (p < 0.01). The discrepancy in overall classification accuracy between the two hemispheres was not due to differences in parameter variability, as there were no significant differences in parameter variance between homologous regions in different hemispheres (calculated across the whole sample).

Thus, although all parameters provided statistically significant predictions in the left hemisphere, only the full model and cortical thickness displayed significant predictions in the right hemisphere. Highest classification accuracy (90.0%) might thus be obtained by using cortical thickness measures exclusively in the left hemispheres. Correlation coefficients between ADI-R subscales and the test margin coming from the combined model are listed in Table 51. The test margin in the left hemisphere was positively correlated with the ADI-R scores for individuals with ASD in the social (r = 0.414; p < 0.05) and communication domain (r = 0.62; p < 0.01). Therefore, individuals with higher values on these ADI-R subdomains are located on the extreme right relative on the hyperplane, whereas the individuals with a lower level of impairment are mostly located in close proximity to the hyperplane overall.

Classification of individuals with ADHD using the established ASD classifier

In order to establish the degree of clinical specificity of the discrimination algorithm to ASD, rather than to NDs in general, the established classifier was used to predict group membership of individuals with ADHD. To predict group membership, the model including all parameters was chosen as it provided best classification accuracy in the right hemisphere and high accuracy in the left hemisphere. On the basis of the neuroanatomical information available for the left hemisphere, 15 out of the 19 individuals with ADHD were allocated to the control group (78.9%) and four individuals with ADHD were allocated to the ASD group (21%). Using the right hemisphere, the classifier allocated cases with ADHD with approximately equal frequencies (47% allocated to the ASD category; 52% allocated to the control category). The classification plots for individuals with ADHD are shown in Figure 22 for both hemispheres.

FIGURE 22.

Classification plots showing group allocation of individuals with ADHD (blue squares) in the (a) left and (b) right hemisphere using the ASD classifier. Reproduced from Ecker et al. 4 © 2010 Ecker et al. ; licensee BioMed Central Ltd. This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Adapted from Ecker et al. 4 © 2010 Ecker et al. ; licensee BioMed Central Ltd. This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Discrimination maps of ASD-specific abnormalities

The spatial maps of brain regions identified as outlined above are shown in Figure 23 and a summary description can be found in Table 52. Each of these maps is a spatial representation of the SVM weight vector, and although they do not directly quantify the information content of each region, each map forms a spatially distributed pattern showing the relative contribution of each voxel to the decision function. Although ‘knockout’ techniques may be used to directly quantify the information content of particular regions, this approach may be hampered by substantial information redundancy across brain regions. Nonetheless, the spatial distribution of the weight vector does provide information about which brain regions contributed to classification and in this case is suggestive of a distributed pattern of relative deficit or excess in ASD with respect to controls. We emphasise that due to the multivariate character of SVM (i.e. it considers inter-regional correlations), each region in the discrimination maps should be interpreted in the context of the entire discriminating pattern and should not be considered in isolation. Furthermore, individual regions may display high classification weights for several reasons [e.g. there is a large difference in volume between groups in that region, or the region is highly intercorrelated with other components of the network (i.e. pattern)]. This is of particular relevance to ASD, as individuals with the disorder most likely have abnormalities in the development of neural systems, in addition to differences in isolated regions. Thus, it is important to highlight that individual network components are not necessarily different between groups, but should be considered as constituent parts of a neuroanatomical network discriminating between groups.

FIGURE 23.

Discrimination maps for the five different morphometric features in the left and right hemispheres. Colour maps represent the weight vector on the basis of the five modality classification for (a) cortical thickness; (b) average convexity; (c) metric distortion; and (d) pial area. Weights for the mean (radial) curvature did not exceed the set threshold. Positive weights (i.e. overall excess patters in ASD relative to controls) are displayed in red, negative weights (i.e. overall deficit patterns in ASD relative to controls) are displayed in blue. Reproduced from Ecker et al. 4 © 2010 Ecker et al. ; licensee BioMed Central Ltd. This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Adapted from Ecker et al. 4 © 2010 Ecker et al. ; licensee BioMed Central Ltd. This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

As can be seen in Figure 23, the five investigated parameters resulted in different spatially distributed patterns of regions with highest contribution to the discrimination. Although cortical thickness provided best classification accuracy and maximum weight values, differences between groups were observed in both volumetric as well as geometric features.

Colour maps represent the weight vector on the basis of the five modality classification for cortical thickness (see Figure 23a), average convexity (see Figure 23b), metric distortion (see Figure 23c) and pial area (see Figure 23d). Weights for the mean (radial) curvature did not exceed the set threshold. Positive weights (i.e. overall excess patters in ASD relative to controls) are displayed in red, negative weights (i.e. overall deficit patterns in ASD relative to controls) are displayed in blue (see Figure 23).

Morphometric profiles

The discrimination maps show that different morphometric parameters elicited different spatial patterns of weights with some overlap between particular parameter pairs. The weights for individual parameters are thus not uniform across the cortex, but vary from region to region (i.e. vertex to vertex). This can also be seen on the basis of the morphometric profile for individual regions, which are displayed in Figures 24–27 and listed in Table 54. These profiles visualise the mean parameter weights for individual features within ROI.

FIGURE 24.

Visualisation of the morphometric abnormalities in the right intraparietal sulcus. (a) Colour maps represent the weight vector score; (b) outlines of the cortical surface for ASD (red) and control (blue) group (this main discriminating factor in this group was an increase in sulcal depth in ASDs relative to controls); (c) differences in sulcal depth for this ROI are shown for both groups; and (d) morphometric profile for this region. Profiles were derived by averaging the weight vector scores across vertices within this ROI and for the different morphometric parameters. Weights were identified on the basis of the concatenated SVM model, thus showing the relative contribution of parameters in this ROI. Reproduced from Ecker et al. 4 © 2010 Ecker et al. ; licensee BioMed Central Ltd. This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Adapted from Ecker et al. 4 © 2010 Ecker et al. ; licensee BioMed Central Ltd. This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

FIGURE 25.

(a) Visualisation of the morphometric abnormalities in the left inferior parietal lobe (BA39); (b) outlines of the cortical surface for ASD and control group; (c) differences in metric distortion for this ROI are shown for both groups; and (d) morphometric profile. Reproduced from Ecker et al. 4 © 2010 Ecker et al. ; licensee BioMed Central Ltd. This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Adapted from Ecker et al. 4 © 2010 Ecker et al. ; licensee BioMed Central Ltd. This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

FIGURE 26.

(a) Morphometric abnormalities in the middle temporal sulcus; (b) visualisation of cortical thickness measures within the middle temporal sulcus; (c) and (d) in this region, only cortical thickness discriminated between ASD and control groups, for which individuals with ASD displayed an increase in thickness, relative to controls. Reproduced from Ecker et al. 4 © 2010 Ecker et al. ; licensee BioMed Central Ltd. This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Adapted from Ecker et al. 4 © 2010 Ecker et al. ; licensee BioMed Central Ltd. This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

FIGURE 27.

(a) Morphometric abnormalities in the posterior cingulate gyrus; (b) and (c) a combination of cortical thickness and folding pattern lead to a high contribution to the classification in that region. Reproduced from Ecker et al. 4 © 2010 Ecker et al. ; licensee BioMed Central Ltd. This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Adapted from Ecker et al. 4 © 2010 Ecker et al. ; licensee BioMed Central Ltd. This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

In the intraparietal sulcus for instance, high discrimination weights were observed for differences in sulcal depth with lower weights for volumetric features such as cortical thickness or surface area (see Figure 24). A similar profile was also seen in the inferior parietal lobe (BA39). Here, the weights for volumetric parameters such as cortical thickness or surface area were very low in comparison with the weights associated with the pattern of cortical folding (i.e. metric distortion) (see Figure 25).

Other regions, however, displayed high weights for volumetric parameters exclusively. For instance, in the temporal sulcus (BA21), high weights were observed for cortical thickness exclusively in the light of low weights in all other parameters (see Figure 27). There were also regions with equally high weights in volumetric as well as geometric features (e.g. the anterior cingulate gyrus). Here, cortical thickness as well as local cortical folding displayed high parameter weights.

In summary, we observed a spatially distributed pattern of regions that can be used to differentiate individuals with ASD from controls. Overall, highest weights were observed for measures of cortical thickness, followed by geometric features such as sulcal depth and metric distortion, which were predominantly observed in parietal regions. This suggests that both volumetric and geometric features play an important role in distinguishing between ASD and control group. Finally, we demonstrated that the parameter weights for individual cortical features were region dependent.

Strengths and weaknesses

Our findings should be interpreted in the context of a number of methodological limitations.

First, the classification algorithm is highly specific to the particular sample used for ‘training’ the classifier, namely high-functioning adults with ASD. The advantage of this approach is that the classifier offers high specificity with regard to this particular subject group, but is less specific to other cohorts on the spectrum. Owing to the small sample size it was also not possible to reliably investigate differences between HFA and Asperger syndrome. Evidence275 suggests that by adulthood these groups are largely indistinguishable at the phenotypic level. However, the extent to which these groups differ at the level of brain anatomy is unknown and may be worth investigating using SVM in the future. Second, 85% of ASD participants in our sample were diagnosed using the ADI-R and 15% were diagnosed using the ADOS-G. As both diagnostic tools measure autistic symptoms at different developmental stages, the classifier may be biased towards individuals with an early diagnosis of ASD. Although it is not expected that classifier performance on the basis of ADOS-G and ADI-R differ drastically, diagnostic heterogeneity may be a potential limitation. Finally, SVM is a multivariate technique and hence offers a limited degree of interpretability of specific network components. Additional analysis such as ‘searchlight’ or ‘virtual lesions’ approaches276–278 may therefore be combined with SVM in the future to establish the relative contribution of individual regions/parameters to the overall classification performance.

Study design

A descriptive cohort study was conducted using THIN database to investigate the incidence and prevalence of ASD diagnoses, psychotropic medication prescribing and neuropsychiatric-related comorbidities of children, adolescents and young adults in UK primary care.

Data source

The Health Improvement Network contains anonymised computerised information systematically recorded by GPs in the UK for patient management. The database provider collates and organises this information in order for it to be used for research. In the UK, almost all patient care is managed by GPs in primary care, who act as gatekeepers of the UK NHS. When required, GPs will refer patients to hospital consultants or specialists in secondary care for diagnosis and initiation of treatment, and GPs usually continue to monitor their patients and issue prescriptions. The diagnosis and management of ASDs and the initiation of drug treatment for symptoms of ASDs (or comorbidities) is usually the responsibility of specialist teams and GPs manage patients in primary care through shared care arrangements with these teams.

The Health Improvement Network covers approximately 5.7% of the UK population with 3.6 million active patients from 464 general practices. 305 The demographic distribution of THIN is broadly representative of that of the UK; therefore, analysis of the clinical and prescribing data will provide information on trends that is representative of national trends. This is particularly useful in evaluating treatment rates, rates of specific diagnoses and changes in these rates.

The Health Improvement Network is a rich source of clinical primary care data and contains information on patient demographics, prescriptions, diagnoses and referrals. Drugs are coded in the database using Multilex^® codes (First Databank; THIN). Diagnoses, symptoms and referrals are coded using Read Clinical Terms, a comprehensive hierarchical system. 306 The database has been previously used to study trends in disease incidence and prescribing patterns in paediatric and adult populations. 307–309 A previous study using a similar UK primary care database (the former General Practice Research Database; GPRD), has validated diagnoses of ASDs in GP records by expert review and by a computerised diagnostic algorithm of DSM-IV symptoms ratings, showing they are appropriate for the identification of ASD cases. 310 Practices who contribute to THIN and/or GPRD use the same practice management software (Vision, In Practice Systems Ltd, London, UK), so data from both databases are collected in a similar manner and structure. In fact, a large proportion of practices contributing to THIN also contribute to GPRD (66%, 327/495 practices in 2001–8). 311