CollAborative care and active surveillance for Screen-Positive EldeRs with subthreshold depression (CASPER): a multicentred randomised controlled trial of clinical effectiveness and cost-effectiveness

Approvals obtained

This study was approved by the NHS Leeds East Research Ethics Committee (REC) on 28 September 2010 (reference number 10/H1306/61). Research management and governance approval was obtained for each trial centre thereafter (see Appendix 1). This trial was assigned the number ISRCTNO2202951.

Trial centres

Four centres in the north of England were selected as trial sites: York centre (core study centre) covering the cities of York, Harrogate and Hull and the surrounding areas; Leeds centre and the surrounding area; Durham centre and the surrounding area; and Newcastle upon Tyne centre, including Northumberland and North Tyneside. Each centre was responsible for co-ordinating the recruitment of participants into the study (trial and epidemiological cohort).

Duration of follow-up

All participants were followed up at 4 and 12 months.

Participant eligibility

Sample size

We estimated that to detect a minimum effect size of 0.3 (with 80% power and a two-sided 5% significance level) would require 352 patients, 176 in each arm. Although this was an individually randomised trial, there may have been potential clustering at the level of each collaborative-care case manager, hence we needed to inflate the sample size to account for this. Based on an intracluster correlation coefficient (ICC) of 0.02 and a projected caseload size of 20, the design effect would be 1.38 {1 + [(20 – 1) × 0.02]}, which meant that we required 486 patients or 243 in each group. Allowing for a potential loss to follow-up of 26%, the final sample size needed to be at least 658 patients or 329 in each group.

Epidemiological cohort

The CASPER study was originally designed to assemble an epidemiological cohort of people aged ≥ 75 years; after the pilot phase the age threshold was reduced to include adults aged ≥ 65 years (see Chapter 4). Through our broad inclusion criteria we successfully recruited a total of 4668 patients aged ≥ 65 years into the CASPER cohort, from which we identified those eligible to participate in the CASPER trial, a trial of collaborative care in older adults with subthreshold depression. The reasoning behind this strategy was twofold: first, to enable us to recruit an adequate number of potential participants who would subsequently be identified as suffering from subthreshold depression, as we believed that this would not necessarily be recorded in GP records; second, to establish an epidemiological cohort of older adults who could be followed up and who would help inform the knowledge base around the health and well-being of older adults. This type of study design is termed a cohort multiple randomised controlled trial. 33

Consenting participants

During the consent stage potential participants were asked to complete the Whooley questions,9 a two-item depression screening/case-finding tool. These questions were asked at two different time points – on the background information sheet at invitation and in the baseline questionnaire – both times as self-reports. At the consent stage, participants were informed about the opportunity of participating in other related studies (e.g. qualitative studies); they were asked to indicate if they agreed to be approached in the future by ticking a box on the consent form (see Chapter 4). All participants who consented to take part in the CASPER study at this stage became part of the CASPER cohort. Participants did not become part of the CASPER trial until they had been subsequently assessed for suitability by completing a standardised diagnostic interview and randomised.

Baseline assessment

On receipt of written consent from participants, baseline data were collected through a self-report questionnaire. All participants who returned a completed consent form to the core study centre were sent a baseline questionnaire (see Appendix 3). Participants were asked to respond to the Whooley questions9 for a second time and to provide self-report medication data. They were also asked to complete a range of health surveys consisting of the PHQ-9,34 a measure of depression severity using a nine-item depression scale in reference to how a respondent has been feeling over the past 2 weeks; the Short Form questionnaire-12 items (SF-12),35 a measure of health-related quality of life; the European Quality of Life-5 Dimensions (EQ-5D),36 a standardised measure of health state utility, designed primarily for self-completion by respondents; the Generalised Anxiety Disorder seven-item scale (GAD-7),37 a severity measure of generalised anxiety used to gauge anxiety in the past 2 weeks; the Patient Health Questionnaire-15 items (PHQ-15),38 a measure of somatic complaints using a 15-item scale in reference to the last month; and the Connor–Davidson Resilience Scale two-item version (CD-RISC 2),39 used to measure an individual’s resilience and ability to bounce back.

Randomisation

Randomisation was carried out by the York Trials Unit Randomisation Service [see www.yorkrand.com/ (accessed 16 September 2015)]. Participants were automatically randomised into the trial on a 1 : 1 basis to either the intervention group or the control group following the completion of a diagnostic interview. All diagnostic interviews were conducted by a trained researcher from the study team over the telephone. The major depressive episode module of the Mini International Neuropsychiatric Interview (MINI)14 was used to ascertain the presence or absence of core depressive symptoms. This allowed potential recruits to be identified as having major depressive disorder (five or more symptoms), subthreshold depression (two to four symptoms) or no depression (one symptom) (Box 1). 14,40,41 All participants diagnosed with subthreshold depression were randomised to either the intervention arm or the control arm.

Once participants had been randomised they were sent a letter informing them of the outcome of their diagnostic interview; if identified as having subthreshold depression they were informed of their group allocation, either collaborative care or usual care. Each participant’s GP was also sent a letter informing him or her that the named patient was eligible to take part in the CASPER trial because of the outcome (subthreshold depression) of the diagnostic interview (MINI) and specifying which arm of the trial he or she had been randomised to.

Ineligible participants

For all those participants whose outcome was not subthreshold depression (either non-depressed or major depressive disorder), a letter was sent informing them that they were ineligible for the CASPER trial but that they would remain in the CASPER epidemiological cohort and continue to be followed up by questionnaire. Their GPs were also informed of this.

Amendment to those eligible for randomisation

The decision to make recruitment to the fully powered trial more targeted had been taken at the design stage of the CASPER trial [see original protocol National Institute for Health Research Evaluation, Trials and Studies (NETS) website: www.nets.nihr.ac.uk/projects/hta/081904 (accessed 24 September 2015)]. This decision was supported by the validation of the Whooley questions, which provided evidence of a negative response being effective at ruling out depression. The randomisation process changed once the pilot was completed. For the main trial, only those participants who responded positively to at least one of the two Whooley questions, at either consent or baseline, were eligible for a diagnostic interview using the MINI. This amendment was necessary to increase the capacity for recruitment of subthreshold participants into the trial. All participants who responded negatively to both the Whooley questions at both time points were informed that they were not eligible for the CASPER trial but would remain in the CASPER study (epidemiological cohort) and would be followed up at 4 and 12 months using the same questionnaires as those used for the trial participants.

Control group

Participants in the control group were allocated to receive usual GP care. They received no additional care to the usual primary care management of subthreshold depression offered by their GP, in line with NICE depression guidance as implemented by their GP and local service provision. 6,7

Intervention group

Participants in the intervention group were allocated to receive a low-intensity programme of collaborative care using behavioural activation, designed specifically for those aged ≥ 65 years with subthreshold depression. Collaborative care was delivered by a case manager [a primary care mental health worker/Improving Access to Psychological Therapies (IAPT) worker] for an intended 8–10 weeks. This took place alongside participants’ usual GP care. The defining feature of collaborative care is a collaboration of expertise to help support the participant: the case manager works alongside the participant, sharing any relevant information with the GP and a mental health specialist (psychiatrist or psychologist). The case manager is a cohesive link between the participant and other professionals involved in his or her care. For example, if a case manager deemed a participant’s depressive symptoms to have deteriorated (moving from subthreshold to above-threshold depression), he or she informed the GP to optimise the management of the patient’s condition.

Collaborative care in the CASPER trial included telephone support, symptom monitoring and active surveillance, facilitated by a computerised case management system [Patient Case Management Information System (PC-MIS); see www.york.ac.uk/healthsciences/pc-mis (accessed 16 September 2015)], and low-intensity psychosocial management (behavioural activation). Participants randomised to the collaborative-care intervention group were contacted by a case manager within a week of their randomisation to arrange their first session of collaborative care. This was carried out face to face, usually at their home unless they preferred an alternative venue. After this initial meeting subsequent sessions were carried out on a weekly basis by telephone unless the participant had sensory impairments or preferred to have face-to-face visits. Case managers worked collaboratively with participants, liaising with GPs and other health professionals involved in their care. They contacted them to discuss issues relating to participants’ mental and physical health, both in routine updates and when any concerns were identified. Case managers worked with participants to identify problems and agree goals for the intervention. They also worked with participants to identify and subsequently provide information about other services that may be helpful, such as voluntary and statutory sector organisations and services.

Refinement of collaborative care/behavioural activation

The delivery of collaborative care and behavioural activation had been established in working-age adults and an appropriate training package and manual already existed. 28 However, these had not been tailored for use with older adults diagnosed with subthreshold depression. Before the study began, necessary changes were made to both the training package and the manual (see following sections) to account for differences that may exist in the older adult population. Further refinement was made following the pilot phase, based on feedback from participants to case managers.

Participant follow-up

All participants in the CASPER trial were followed up with questionnaires at 4 months (see Appendix 5) and 12 months (see Appendix 6). Post-randomisation questionnaires were posted to participants from the York Trials Unit, along with a pre-addressed prepaid envelope. Participants could complete the questionnaires manually and return them by post to the York Trials Unit, or they could complete the questionnaire online; an instruction sheet explaining how to log onto the CASPER website and complete the process was included with each questionnaire. Reminder letters were sent by post to any participants who at 2 weeks had not returned their questionnaire. Telephone follow-up by one of the study team’s researchers was conducted for any participants who had not returned a questionnaire, to enable the primary outcome measure (PHQ-9) to be completed at the very least.

Withdrawals

Withdrawal could occur at any point during the study at the request of the participant. If a participant indicated that he or she wished to withdraw from the study, a researcher would speak to the participant to clarify to what extent he or she wished to withdraw (withdrawal from the intervention, from follow-up or from all aspects of the study). When withdrawal was only from the intervention, follow-up data continued to be collected. Data were retained for all participants up to the date of withdrawal, unless they specifically requested that their details be removed.

Objective data

Once all of a GP practice’s participants had completed their follow-up, objective data were collected for each trial participant. Objective data consisted of details on participants’ prescribed medication and the number of contacts that they had with their GP practice during their time in the trial. The only exception was for those participants who had withdrawn in full, thereby withdrawing consent to access their medical records. Objective data were collected from GP practices by request from the core study centre. A spreadsheet template was emailed to the key contact of each GP practice, which included each of the practice’s trial participants’ identification codes along with pre-written frozen headings; no identifiable data were included. Also listed were the search dates for each participant, from the date that they were randomised until either the date that they completed the study 12 months later or the date that they had died if that was the case. Data were still collected on participants who had withdrawn from treatment or follow-up as they had provided us with consent to access their health records for the 12 months that they would have been in the study. All objective data were collected by e-mail and this method was approved by the University of York’s data protection team on the basis that no identifiable data were transferred; patients were identified by GP practices using the administration code assigned to them at the recruitment stage.

Suicide protocol

A small but elevated risk of suicide and self-harm was inherent in the population studied, who had all been identified as having subthreshold depression. All participants (both usual care and collaborative care) were subject to usual GP care. GPs were responsible for the day-to-day management of subthreshold depression. GPs were accountable for all treatment and management decisions, including prescribing, referral and assessment of risk. This arrangement was made clear to all clinicians and GP practices who agreed to participate in the study. The pragmatic nature of the CASPER trial meant that we did not seek to influence this arrangement. However, we did follow good clinical practice by monitoring for suicide risk during all of our encounters with participants. When a patient expressed a risk through thoughts of suicide or self-harm, we followed the study-specific procedure for suicide risk (see Appendices 7.1 and 7.2).

Patient and public involvement in research

The CASPER trial was informed by the involvement of users of mental health services and carers throughout the research period. An advisory group was established at the end of the pilot phase to review the materials used in the study. This consisted of a number of older adults, some of whom had mental health conditions, along with a carer representative. This group provided valuable insights into the relevance and readability of the study documentation. In the future we plan to engage with patient and public involvement in our dissemination strategies to guide on how best to share the findings.

Further studies

Following validation of the Whooley questions during the CASPER pilot phase42 it was acknowledged that a sizeable number of participants (approximately 5%) were identified with case-level depression after completing the diagnostic interview. The CASPER Plus trial was born out of an aspiration to be able to offer collaborative care and behavioural activation to older adults with more severe depression [see www.trialsjournal.com/content/15/1/451 (accessed 16 September 2015)].

Primary outcome

The primary end point for the trial was patient-reported depression severity as measured by the PHQ-9 at 4 months’ follow-up. Each item is scored between 0 and 3 and thus PHQ-9 scores can range from 0 to 27, with higher scores indicating greater depression. Total scores from 0 (not depressed) to 27 (severely depressed) were calculated based on the nine PHQ-9 items. These data were collected via self-report on the follow-up questionnaire. Any participants who did not return a completed questionnaire were sent a reminder; those participants who did not respond were telephoned by one of the study team’s researchers to ask them to complete the PHQ-9 over the telephone. If one or two items were missing, missing items were replaced with the mean of the remaining items.

The PHQ-9 data were collected at baseline and randomisation as well as at 4 and 12 months’ follow-up. Scores at baseline and randomisation are reported in Baseline characteristics. When analyses were adjusted for initial PHQ-9 score, the score at randomisation was used. The primary end point for the CASPER trial was at 4 months’ follow-up. A standard effect size of 0.3 for mean PHQ-9 difference was sought between treatment arms, equivalent to a PHQ-9 score difference of approximately 1.35 points, assuming a standard deviation (SD) of 4.5. The standard effect size of 0.3 is of moderate size for psychological interventions and in line with collaborative care effects observed in other studies. Cohen48 classifies a standard effect size of 0.3 as a small to medium effect size and this is in line with NICE guidelines for depression,6 in which a similar grading of clinical significance is adopted.

Secondary outcomes

The secondary outcome measures used were:

• depression severity and symptomatology at 12 months (PHQ-9)

• binary depression severity at 4 and 12 months (PHQ-9) using scores of ≥ 10 to designate moderate depression caseness

• quality of life at 4 and 12 months (SF-12 and EQ-5D)

• psychological anxiety at 4 and 12 months (GAD-7)

• mental health medication at 4 and 12 months

• mortality at 4 and 12 months.

Mental health medication

Medication data were captured by self-report on the follow-up questionnaires. Participants indicated prescribed medication by selecting from a list of 10 antidepressants as well as listing any other medications that they had been prescribed. Independent objectively collected medication data from GP records were incorporated in the economic analysis.

Mortality data

A data linkage service was established with the Health and Social Care Information Centre to provide regular updates from the Office for National Statistics mortality data on any trial participants who had died while in the study.

Members of the research team recorded any identified deaths on the study management database.

Other patient questionnaire data collected

Adverse events

The CASPER study was a non-clinical trial of an investigational medicinal product and was therefore not subject to any additional restrictions. Decisions regarding prescription of medications were made by the participants in conjunction with their GP; participation in the study had no bearing on this process. If participants asked a member of the CASPER study team for an opinion on medication issues, they were strongly encouraged to seek advice from their GP.

The study recorded details of all serious adverse events (SAEs); any judged to have been related to the study were required to be reported to the main REC under the terms of the Standard Operating Procedures for RECs. 49 In the context of the CASPER study’s older adult population the occurrence of many of the SAEs was expected; unscheduled hospitalisations, life-threatening conditions, incapacitating illnesses and deaths were not perceived as unexpected events and therefore they would be reported as SAEs only if they appeared to be related to an aspect of taking part in the study (e.g. participation in treatment, completion of follow-up questionnaires, participation in qualitative substudies or participation in telephone contact).

When a SAE was identified, the Trial Manager was informed by e-mail using a participant’s trial identification number, not any identifiable data. The Trial Manager then informed the Chief Investigator and two members of the Trial Management Group (TMG), who jointly decided if the event should be reported to the REC as a SAE. A SAE form was completed and a copy filed securely at the core study centre. Any unexpected SAEs that were also judged to have been related were to be reported to the REC within 15 days of the Chief Investigator becoming aware of them. In the CASPER study none of the SAEs appeared to have been related to the trial.

The occurrence of adverse events during the trial was monitored by an independent Data Monitoring and Ethics Committee (DMEC) and the Trial Steering Committee (TSC). The DMEC/TSC would have seen immediately all SAEs thought to be treatment related.

Data collection schedule

An overview of the time points at which trial data were collected is presented in Table 1.

Statistical assumptions

The CASPER trial was powered to detect a standard effect size of 0.3 for PHQ-9 depression severity between treatment arms, assuming 80% power, 5% two-sided significance, an ICC of 0.02 (based on a caseload of 20 patients) and 26% loss to follow-up (a revised figure following interim data review). The total sample size required based on these criteria was 658 patients (329 in each arm). Participants were randomised 1 : 1 using simple randomisation to either collaborative care or usual care.

Participants, care deliverers and the study team were not blinded to treatment allocation. However, allocations were concealed (group A and group B) for interim study reports, for example for the purpose of independent data monitoring reporting. The trial statistician responsible for the final statistical analysis was kept blind to group allocation until the primary analysis had been completed.

All analyses were conducted on an intention-to-treat basis, using a two-sided statistical significance level of 0.05 unless otherwise stated. All statistical analyses followed a prespecified statistical analysis plan.

Recruitment and flow of participants through the trial

Participants were recruited into the CASPER trial between June 2011 and July 2013 from four UK sites and their surrounding areas in the north of England: York, Leeds, Durham and Newcastle upon Tyne. A total of 32 GP practices consented to screen their practice lists and identify patients who met the initial inclusion criterion: aged ≥ 75 years during the pilot phase and ≥ 65 years during the main trial (see Chapter 4). The exclusion criteria consisted of any known alcohol dependency and/or psychotic symptoms as recorded on GP records, any known adverse comorbidities or any other factors that resulted in GPs deeming it inadvisable to invite patients, such as a recent bereavement.

We were obliged to follow a broad recruitment strategy because we would not have been able to accurately identify patients with subthreshold depression using Read codes from GP records. Therefore, GP practices conducted searches with broad inclusion criteria, which meant that we invited nearly all patients aged ≥ 65 years, except for the few who met the exclusion criteria. This explains why we recruited such a large number of participants into the CASPER cohort and why the number who made it into the CASPER trial with subthreshold depression was far smaller.

A total of 37,134 patients were identified by GP practices between June 2011 and July 2013 and invited by letter to take part in the CASPER study. Of 6693 patients who consented, 4259 were excluded and 2434 patients were assessed for eligibility by diagnostic interview. Based on the diagnostic interview, 705 (29%) patients were identified to have subthreshold depression and were randomised into the CASPER trial. Of those 705 participants randomised, 344 were allocated to collaborative care and 361 to usual care. The remaining patients were classified either as being below the threshold for depression (n = 1558, 64%) or as suffering from a major depressive disorder (n = 171, 7%); they became part of the epidemiological cohort. The randomised number of 705 participants exceeded the planned sample size of 658 participants. The flow of participants is illustrated in a Consolidated Standards of Reporting Trials (CONSORT) diagram in Figure 1.

FIGURE 1.

Consolidated Standards of Reporting Trials (CONSORT) diagram. IQR, interquartile range; max., maximum; min., minimum.

Timing of follow-up

Different patients in the CASPER trial were followed up according to two different time schedules. Follow-up times (4 and 12 months) were initially calculated from baseline (n = 165 randomised participants, 23%), with randomisation expected shortly after. As the volume of participant throughput increased, diagnostic interview and randomisation occurred only sometime after baseline data collection; therefore, follow-up times for participants later on in the trial were calculated from the time of randomisation as the more appropriate reference point (n = 540 randomised participants, 77%). A sensitivity analysis was carried out to assess the impact of having adopted two different reference points (see Table 22).

Table 6 illustrates the average delay between baseline and randomisation, with a median delay of 2 weeks and considerable variability. Table 6 also shows that, for those patients whose follow-up reference time was at baseline, their 4-month follow-up could occur much sooner after (and for four cases before) randomisation.

Response types and response times

Participants had the option to complete questionnaires at baseline and follow-up on paper or online. Only a small proportion of questionnaires were completed online, which reduced over time [2.7% at baseline (3.8% collaborative-care group, 1.7% usual-care group), 1.9% at 4 months’ follow-up (2.6% collaborative-care group, 1.2% usual-care group) and 1.0% at 12 months’ follow-up (1.7% collaborative-care group, 0.4% usual-care group). Patients in the collaborative-care arm were slightly more likely to use the online option.

The average response times to questionnaires are detailed in Table 7 by trial arm for data for which completion dates were provided by participants or dates of return were logged on the trial database. Participants generally completed questionnaires within a week and returned them within 2 weeks. There were no differences in response time between treatment arms.

Descriptives

Complete PHQ-9 responses were available for all participants at randomisation. At follow-up, 495 patients (70%) had valid PHQ-9 scores at both follow-up times, 115 patients (16%) had a valid PHQ-9 score at 4 months or 12 months only, and for 95 patients (13%) no PHQ-9 scores were available at follow-up.

The extent of, and reasons for, missing PHQ-9 scores at each time point are detailed in Table 14 by trial arm. Although there were fewer valid PHQ-9 responses in the collaborative-care arm at both follow-up points, this was mainly because of a higher dropout rate in this group. Of all attempted PHQ-9 completions, only one questionnaire had an insufficient number of questions completed to allow scoring (one patient in the usual-care arm completed only the first five questions at 4 months’ follow-up). The total number of times that a particular PHQ-9 item was missing for any attempted completions is presented in Table 15. Earlier questionnaire items (items 1, 2 and 4) tended to be missed more frequently, but no meaningful pattern emerged.

Score distribution

Figure 2 illustrates the distribution of PHQ-9 scores for each treatment group over time. At randomisation, scores were distributed normally with a slight right skew, which became more pronounced at 4 and 12 months’ follow-up as patients in both arms improved. A floor effect can be observed at the follow-up time points as a larger number of participants reported no or very mild depression on the PHQ-9 scale.

FIGURE 2.

Distribution of PHQ-9 scores by treatment group. (a) Randomisation; (b) 4-month follow-up; and (c) 12-month follow-up.

Unadjusted summary statistics

Summary statistics of the raw PHQ-9 scores are given in Table 16 and are illustrated in Figure 3. Average depression severity as measured by the PHQ-9 was around 7.5 score points at randomisation, which is classed as mild depression. Scores improved for both arms between randomisation and 4 months’ follow-up, more so for the collaborative-care arm (to a score of 5.2) than for the usual-care group (to a score of 6.8). At 12 months, average depression scores increased again by approximately half a PHQ-9 score point, but the difference between trial arms was maintained.

TABLE 16 - Unadjusted PHQ-9 descriptive statistics

max., maximum; min., minimum.

Time point	Collaborative care (n = 344)	Usual care (n = 361)	Total (n = 705)
Randomisation, n (%)	344 (100)	361 (100)	705 (100)
Mean (SD)	7.5 (4.29)	7.6 (4.21)	7.5 (4.24)
Median (min., max.)	7 (0, 23)	7 (0, 20)	7 (0, 23)
4 months, n (%)	262 (76.2)	324 (89.8)	586 (83.1)
Mean (SD)	5.2 (4.17)	6.8 (4.50)	6.0 (4.42)
Median (min., max.)	4 (0, 20)	6 (0, 26)	5 (0, 26)
12 months, n (%)	235 (68.3)	284 (78.7)	519 (73.6)
Mean (SD)	5.7 (4.50)	7.2 (5.01)	6.5 (4.84)
Median (min., max.)	5 (0, 24)	7 (0, 27)	6 (0, 27)

FIGURE 3.

Unadjusted mean PHQ-9 scores (with 95% CIs).

Primary analysis

The primary outcome was analysed using a covariance pattern linear mixed model using PHQ-9 score at 4 and 12 months as the outcome. The model included as fixed effects time, treatment group and time × treatment group interaction, adjusting for PHQ-9 depression at randomisation and physical/functional limitations as measured by the baseline SF-12 PCS score. Patients were included in the analysis if they had a valid PHQ-9 score at either 4 or 12 months’ follow-up and complete covariate data. Patients were analysed as part of the group to which they had been randomised (intention to treat).

The correlation of observations within patients over time was modelled by a covariance structure to describe the random effects. Different types of available covariance structures were investigated for this model (unstructured, independent, exchangeable, autoregressive and exponential). The model did not converge for the autoregressive covariance pattern after 1500 iterations. Of the remaining patterns, the unstructured covariance structure (estimating the two residual variances and covariance over time separately) displayed the lowest and therefore best-fitting log-likelihood values and emerged as the significantly better-fitting model when comparing nested models by chi-square test. Therefore, the unstructured covariance pattern was selected.

Diagnostics of model fit showed an acceptable distribution of standard residuals with slight non-normality at the higher end of the distribution because of the right skew of PHQ-9 scores. There was uniform variance between predicted and actual residuals, and no transformation of PHQ-9 scores was carried out for the analysis.

Adjusted PHQ-9 mean scores and group differences for the primary analysis model as specified above are presented in Table 17. The analysis revealed significant differences between trial arms at each of the follow-up time points in favour of collaborative care: 1.31 score points (95% CI 0.67 to 1.95 score points; p < 0.001) for the primary end point at 4 months’ follow-up and 1.33 score points (95% CI 0.55 to 2.10 score points; p = 0.001) at 12 months’ follow-up. Using the overall residual SD (4.35), the score difference at 4 months equates to a standard effect size of 0.30, the exact value for which the trial was powered.

Secondary analyses

Summary of Patient Health Questionnaire-9 items analysis models

Table 25 provides an overview of group means and treatment effect estimates from the primary analysis and secondary analyses of depression severity at 4 and 12 months as measured by PHQ-9 score. Unadjusted means are presented for reference. Adjusted average estimates of group differences from the different analyses ranged from 1.20 to 1.36 score points in favour of collaborative care (equivalent to a standard effect size of 0.28–0.31).

Binary Patient Health Questionnaire-9 items outcome

Using a cut-off of ≥ 10 PHQ-9 score points, Table 26 presents the numbers and percentages of moderately to severely depressed participants at randomisation and follow-up by treatment arm. This is illustrated in Figure 4. Approximately 30% of randomised CASPER participants were depressed at randomisation. At 4 months’ follow-up, this percentage decreased to 17% in the collaborative-care arm and was maintained to 12 months. In contrast, in the usual-care arm the percentage of moderately to severely depressed participants decreased to only 24% by 4 months’ follow-up and increased again to 28% by 12 months’ follow-up.

FIGURE 4.

Unadjusted percentage of patients (with 95% CIs) with moderate to severe depression.

It was planned to analyse the data by logistic mixed-effects modelling, including moderate to severe PHQ-9 depression (yes or no) at 4 and 12 months as the outcome, predicted by treatment group, time (4 and 12 months), group × time interaction, depression severity at randomisation (PHQ-9 score), baseline physical functioning (SF-12 PCS score) and the additionally identified covariates of PHQ-9 depression (prescribed antidepressants at baseline, GAD-7 anxiety and PHQ-15 physical functioning; see Adjusting for covariates predictive of non-response at 4 months). The correlation of outcomes by the same patient was to be modelled by a suitable covariance structure.

However, such a model did not converge in Stata and data were instead analysed by individual logistic regressions at each follow-up time point using the same predictors as above except for the variable for time. The resulting treatment effect estimates are presented in Table 27.

The relatively greater reduction in moderately to severely depressed cases seen in the collaborative-care arm compared with the usual-care arm was not statistically significant at 4 months’ follow-up (OR 1.35, 95% CI 0.85 to 2.16; p = 0.205) but did reach statistical significance at 12 months’ follow-up (OR 1.98, 95% CI 1.21 to 3.25; p = 0.007).

Secondary outcomes

A summary of the secondary outcomes is provided in Box 3.

Descriptive statistics for secondary outcomes are generally presented in tabular form and also in graphical form where appropriate.

For continuous outcomes, any planned statistical analyses were conducted using the same covariance pattern linear mixed model as described for the primary analysis, adjusting for baseline PHQ-9 score and SF-12 PCS score (see Primary analysis). The models used the secondary outcome in question as the dependent variable and further included any baseline assessment of the outcome as well as the additionally identified significant predictors of the primary outcome depression [baseline anxiety (GAD-7), physical functioning (PHQ-15) and prescribed antidepressants; see Adjusting for covariates predictive of Patient Health Questionnaire-9 items score at 4 months] as covariates. Estimates of the effect of the intervention were derived and are presented for each follow-up time point. Any other planned analyses for different types of outcome are described in the relevant section for each measure.

Quality of life

Quality of life was assessed by patient report on the physical and mental component scales of the SF-12 (PCS and MCS) as well as the five dimensions of the EQ-5D.

Short Form questionnaire-12 items physical component summary score

The SF-12 PCS score ranges from 0 to 100, with higher scores indicating better functioning. Unadjusted means for physical functioning are presented in Table 28 and Figure 5, and the results of the formal statistical analysis by mixed modelling are given in Table 29. The figures indicate that physical functioning was below the average adult physical health status (score of < 50) for participants throughout the trial period, as would be expected in an elderly population. Patients in the usual-care arm maintained physical functioning scores at approximately 35–36 score points over follow-up, whereas patients in the collaborative-care arm improved on average to up to 40 points over follow-up. The differences in physical functioning were statistically significant based on the mixed-effects analysis at 4 months (mean score difference 2.83, 95% CI 1.62 to 4.03; p < 0.001) and 12 months (mean score difference 1.67, 95% CI 0.27 to 3.06; p = 0.020).

TABLE 28 - Unadjusted SF-12 PCS descriptive statistics

max., maximum; min., minimum.

Time point	Collaborative care (n = 344)	Usual care (n = 361)	Total (n = 705)
Baseline, n (%)	337 (98.0)	356 (98.6)	693 (98.3)
Mean (SD)	38.0 (13.37)	36.5 (13.02)	37.2 (13.20)
Median (min., max.)	37.5 (4.6, 69.9)	35.1 (5.7, 66.6)	36.4 (4.6, 69.9)
4 months, n (%)	254 (73.8)	315 (87.3)	569 (80.7)
Mean (SD)	40.0 (12.39)	35.4 (12.96)	37.5 (12.90)
Median (min., max.)	40.6 (10.5, 66.0)	34.9 (10.0, 69.7)	38.2 (10.0, 69.7)
12 months, n (%)	226 (65.7)	276 (76.5)	502 (71.2)
Mean (SD)	38.8 (13.11)	35.4 (12.73)	36.9 (13.00)
Median (min., max.)	39.2 (4.6, 69.6)	34.9 (7.1, 64.5)	36.5 (4.6, 69.6)

FIGURE 5.

Unadjusted mean SF-12 PCS scores (with 95% CIs).

TABLE 29 - Group difference in mean SF-12 PCS scores

Mixed-effects model adjusted for treatment group, time (4 and 12 months), group × time interaction, PHQ-9 score at randomisation, baseline SF-12 PCS score, prescribed antidepressants at baseline (yes/no), baseline GAD-7 score and baseline PHQ-15 score, including random residual variances and covariance over time.

Time point (months)	Collaborative care			Usual care			Group difference
	n	Mean	95% CI	n	Mean	95% CI	Mean	95% CI	p-value
4	263	38.8	37.7 to 39.9	316	36.0	35.0 to 37.0	–2.83	–4.03 to –1.62	< 0.001
12	263	37.8	36.6 to 39.0	316	36.1	35.0 to 37.2	–1.67	–3.06 to –0.27	0.020

Short Form questionnaire-12 items mental component summary score

The SF-12 MCS score ranges from 0 to 100, with higher scores indicating better functioning. Unadjusted means for psychological functioning are presented in Table 30 and Figure 6 and the results of the formal statistical analysis by mixed modelling are given in Table 31.

TABLE 30 - Unadjusted SF-12 MCS descriptive statistics

max., maximum; min., minimum.

Time point	Collaborative care (n = 344)	Usual care (n = 361)	Total (n = 705)
Baseline, n (%)	337 (98.0)	356 (98.6)	693 (98.3)
Mean (SD)	44.3 (10.96)	45.1 (10.02)	44.7 (10.49)
Median (min., max.)	44.9 (12.5, 66.0)	46.3 (9.6, 67.0)	45.6 (9.6, 67.0)
4 months, n (%)	254 (73.8)	315 (87.3)	569 (80.7)
Mean (SD)	48.7 (10.89)	46.6 (9.82)	47.5 (10.36)
Median (min., max.)	50.3 (15.1, 68.8)	47.3 (7.5, 66.9)	48.6 (7.5, 68.8)
12 months, n (%)	226 (65.7)	276 (76.5)	502 (71.2)
Mean (SD)	48.2 (10.11)	45.6 (10.55)	46.7 (10.43)
Median (min., max.)	48.5 (17.5, 72.8)	46.6 (13.9, 66.9)	47.2 (13.9, 72.8)

FIGURE 6.

Unadjusted mean SF-12 MCS scores (with 95% CIs).

TABLE 31 - Group difference in mean SF-12 MCS scores

Mixed-effects model adjusted for treatment group, time (4 and 12 months), group × time interaction, PHQ-9 score at randomisation, baseline SF-12 MCS score, baseline SF-12 PCS score, prescribed antidepressants at baseline (yes/no), baseline GAD-7 score and baseline PHQ-15 score, including random residual variances and covariance over time.

Time	Collaborative care			Usual care			Group difference
	n	Mean	95% CI	n	Mean	95% CI	Mean	95% CI	p-value
4 months	263	47.6	46.3 to 48.9	316	45.7	44.6 to 46.9	–1.88	–3.29 to –0.47	0.009
12 months	263	46.8	45.4 to 48.1	316	44.6	43.4 to 45.9	–2.15	–3.70 to –0.59	0.007

The figures indicate that mental functioning was just below the average adult mental health status (score of < 50) for participants throughout the trial period. Patients in the usual-care arm maintained physical functioning scores at approximately 45–46 score points over follow-up, whereas patients in the collaborative-care arm improved on average to up to 48 points over the follow-up. The differences in mental functioning were statistically significant based on the mixed-effects analysis at 4 months (mean score difference 1.88, 95% CI 0.47 to 3.29; p = 0.009) and 12 months (mean score difference 2.15, 95% CI 0.59 to 3.70; p = 0.007).

European Quality of Life-5 Dimensions

The EQ-5D measures quality of life on five dimensions – mobility, self-care, usual activities, pain/discomfort and anxiety/depression – with participants given three response options to indicate their level of problems for each dimension. The weighted summary index derived from these dimensions is summarised and formally analysed as part of the CASPER cost–utility analysis in the economic evaluation (see Chapter 6). For the purpose of exploring differences in quality of life between treatment arms, the frequencies of responses for each category in each dimension are presented descriptively in Table 32 and illustrated in Figures 7–11.

TABLE 32 - European Quality of Life-5 Dimensions descriptive statistics

Level 1, no problems; level 2, some/moderate problems; level 3, severe problems.

EQ-5D dimension	Time	Severitya	Collaborative care			Usual care
			Total (N)	n	%	Total (N)	n	%
Mobility	Baseline	Level 1	339	129	38.1	357	115	32.2
		Level 2		210	61.9		241	67.5
		Level 3		0	0.0		1	0.3
	4 months	Level 1	254	117	46.1	316	98	31.0
		Level 2		137	53.9		218	69.0
		Level 3		0	0.0		0	0.0
	12 months	Level 1	231	98	42.4	279	83	29.7
		Level 2		133	57.6		194	69.5
		Level 3		0	0.0		2	0.7
Self-care	Baseline	Level 1	338	287	84.9	356	292	82.0
		Level 2		48	14.2		62	17.4
		Level 3		3	0.9		2	0.6
	4 months	Level 1	250	215	86.0	315	253	80.3
		Level 2		34	13.6		61	19.4
		Level 3		1	0.4		1	0.3
	12 months	Level 1	231	191	82.7	279	225	80.6
		Level 2		39	16.9		51	18.3
		Level 3		1	0.4		3	1.1
Usual activities	Baseline	Level 1	338	136	40.2	356	124	34.8
		Level 2		189	55.9		209	58.7
		Level 3		13	3.8		23	6.5
	4 months	Level 1	253	123	48.6	314	108	34.4
		Level 2		116	45.8		187	59.6
		Level 3		14	5.5		19	6.1
	12 months	Level 1	231	112	48.5	280	89	31.8
		Level 2		108	46.8		165	58.9
		Level 3		11	4.8		26	9.3
Pain/discomfort	Baseline	Level 1	339	76	22.4	356	54	15.2
		Level 2		224	66.1		262	73.6
		Level 3		39	11.5		40	11.2
	4 months	Level 1	253	69	27.3	317	45	14.2
		Level 2		166	65.6		239	75.4
		Level 3		18	7.1		33	10.4
	12 months	Level 1	230	63	27.4	280	43	15.4
		Level 2		147	63.9		200	71.4
		Level 3		20	8.7		37	13.2
Anxiety/depression	Baseline	Level 1	337	132	39.2	355	133	37.5
		Level 2		193	57.3		211	59.4
		Level 3		12	3.6		11	3.1
	4 months	Level 1	251	151	60.2	313	141	45.0
		Level 2		95	37.8		163	52.1
		Level 3		5	2.0		9	2.9
	12 months	Level 1	231	123	53.2	280	117	41.8
		Level 2		104	45.0		152	54.3
		Level 3		4	1.7		11	3.9

FIGURE 7.

European Quality of Life-5 Dimensions: mobility – percentage of patients in each severity category.

FIGURE 8.

European Quality of Life-5 Dimensions: self-care – percentage of patients in each severity category.

FIGURE 9.

European Quality of Life-5 Dimensions: usual activities – percentage of patients in each severity category.

FIGURE 10.

European Quality of Life-5 Dimensions: pain/discomfort – percentage of patients in each severity category.

FIGURE 11.

European Quality of Life-5 Dimensions: anxiety/depression – percentage of patients in each severity category.

The majority of CASPER participants indicated no problems or some problems in each of the EQ-5D dimensions, with few patients having severe difficulties. The most frequent use of the severe category was in the pain/discomfort dimension. The greatest treatment group differences were seen for anxiety/depression, with the number of non-depressed patients in the collaborative-care arm increasing from 39% to 60% at 4 months’ follow-up compared with a lower recovery rate from 37% to 45% in the usual-care arm. Relatively greater improvements in favour of the intervention arm were also seen for mobility, usual activities and pain/discomfort, although in each case the number of patients with no difficulties at baseline was already 5–7% higher in the collaborative-care group. Any differences between treatment arms were generally maintained between 4 and 12 months. There were no evident group differences in the self-care dimension.

Psychological anxiety

The GAD-7 is a brief measure of symptoms of anxiety with a score range of 0–21, with higher scores indicating more severe anxiety. Unadjusted means for psychological anxiety based on the GAD-7 are presented in Table 33 and Figure 12 and the results of the formal statistical analysis by mixed modelling are given in Table 34.

FIGURE 12.

Unadjusted mean GAD-7 scores (with 95% CIs).

The figures indicate that anxiety was, on average, at a mild level (scores between 5 and 9) for all participants at baseline. Both treatment groups improved anxiety levels at 4 months’ follow-up, which was maintained up to 12 months; however, patients in the usual-care arm remained just above the cut-off for mild anxiety (a score of 5) whereas collaborative-care patients improved to be below that level, with an average score of around 4. The differences in psychological anxiety were statistically significant based on the mixed-effects analysis at 4 months (mean score difference 1.08, 95% CI 0.52 to 1.64; p < 0.001) and 12 months (mean score difference 1.01, 95% CI 0.42 to 1.61; p = 0.001).

Antidepressants

Patients indicated on questionnaires whether or not they were currently prescribed any of a list of 10 antidepressants. Table 35 shows the frequencies of prescriptions by trial arm. Overall, patients were prescribed between none and three of the selected drugs (only one patient indicated three and this was at a single time point). Citalopram was the most commonly prescribed antidepressant.

A binary variable was created to indicate whether patients had been prescribed any of the listed antidepressants or not. The variable was assumed to be missing if the participant indicated ‘Don’t know’ and none of the drugs had been selected.

Table 36 presents the numbers and percentages of patients on antidepressants at baseline and follow-up by treatment arm. This is illustrated in Figure 13. Approximately 10% of patients in the collaborative-care arm were prescribed antidepressants at baseline compared with 14% in the usual-care arm. This baseline imbalance (although CIs overlapped) was maintained at each of the two follow-up time points, with little change over time evident for either patient group.

FIGURE 13.

Unadjusted percentage of patients (with 95% CIs) who were prescribed antidepressants.

It was planned to analysed the data using a logistic mixed-effects model including prescribed medication (yes or no) at 4 and 12 months as the outcome, predicted by treatment group, time (4 and 12 months), group × time interaction, prescribed antidepressants at baseline, depression severity at randomisation (PHQ-9 score), baseline anxiety (GAD-7 score) and baseline physical functioning (SF-12 PCS score). The correlation of outcomes by the same patient was to be modelled by a suitable covariance structure.

Such a model did not converge in Stata, however, and data were instead analysed by individual logistic regressions at each follow-up time point using the same predictors as above except for the variable for time. The resulting treatment effect estimates are presented in Table 37.

The relative odds of being prescribed antidepressants were higher in the usual-care arm than in the collaborative-care arm; however the differences were not statistically significant at 4 months’ follow-up (OR 1.98, 95% CI 0.76 to 5.19; p = 0.165) or 12 months’ follow-up (OR 1.39, 95% CI 0.57 to 3.41; p = 0.469).

Mortality

A total of 23 participants died within their 12-month follow-up period, five in the collaborative-care arm (1.5% of randomised patients) and 18 in the usual-care arm (5.0% of randomised patients). Causes of death are summarised in Table 38. A chi-square test revealed that the difference in mortality rate between treatment arms was statistically significant (χ² = 6.97, degrees of freedom = 1, p = 0.008).

All deaths were further recorded as SAEs (see Adverse events) and their potential relatedness to the trial treatment was assessed as part of the processing of adverse events. In total, 81% of deaths were categorised as being unrelated to treatment and 18% as unlikely to be related to treatment. The DMEC noted the marked imbalance in mortality rate between treatment arms and assessed that causes of death could not be reasonably attributed to either the intervention or the control treatment. The difference was therefore treated as a chance result.

Patient Health Questionnaire-15 items

The PHQ-15 is a measure of physical health problems. In this study it had a score range of 0–28 (usual maximum is 30), as a question regarding menstrual problems was removed for the elderly CASPER patient population.

The data are described by summary statistics in Table 39 and are illustrated in Figure 14. Physical health problems as measured by the PHQ-15 did not change over time for usual-care patients, with a score of around 9 (medium symptom severity) at baseline and 4 and 12 months’ follow-up; however, symptoms improved for collaborative-care patients over follow-up, especially at 4 months. Symptoms for which the greatest improvements were observed were pain in arms, legs or joints; dizziness; shortness of breath; constipation, loose bowels or diarrhoea; and trouble sleeping.

FIGURE 14.

Unadjusted mean PHQ-15 scores (with 95% CIs).

Connor–Davidson Resilience Scale two-item version

The two-item CD-RISC 2 resilience measure has a score range of 0–8, with a higher score indicating greater psychological resilience. The CD-RISC 2 data are described by summary statistics in Table 40 and are illustrated in Figure 15. Average resilience at baseline was just under 6 score points and remained constant over the 12 months of follow-up for patients in the usual-care group. Patients in the collaborative-care group improved their average resilience to > 6 score points and the difference between the treatment groups was most pronounced at 4 months’ follow-up.

FIGURE 15.

Unadjusted mean CD-RISC 2 scores (with 95% CIs).

Collaborative care: required resources and associated costs

Case managers were psychological wellbeing practitioners (PWPs) employed at NHS Band 5 level. After recruitment to provide collaborative care as part of the CASPER trial, case managers all received training. In total, there were three training events covering four regions of the study (i.e. Durham, Leeds, Newcastle upon Tyne and York), each consisting of 2 consecutive days of training. The number of attendees per training event varied and efforts were made to provide training in a manner which ensured that the overall costs of travel and accommodation were minimal.

During the training, PWPs were orientated to the case managers’ manual, outlining the overall principles of collaborative care and a ‘session by session overview’ of what case managers aimed to achieve with patients. The training courses for case managers were predominantly provided by two trainers; these trainers subsequently also supervised case managers during implementation of the collaborative-care programme.

The manual stipulated that the programme of treatment should consist of ‘8–10 mainly telephone contacts with occasional face-to-face contacts over a period of 12 weeks’. In terms of the expectation for each session, it further stated that ‘contacts last 45 minutes for session one and 20–30 minutes for each subsequent contact’. The first session was generally held face to face and took place at a participant’s home, a GP surgery or another community venue.

Case managers also received weekly supervision from a designated supervisor. The schedule of supervision followed a standardised agenda whereby for each patient there was a weekly discussion, with case managers preparing feedback on each case to discuss with their supervisor. Supervisors were responsible for providing support to case managers on the process of collaborative care, medication management and specific psychological interventions. On average, each patient contact was discussed between the case manager and the supervisor for approximately 5 minutes.

Case managers provided participant-specific feedback to GPs. In the first instance, case managers worked with and advised participants’ GPs on their care. During treatment case managers would provide a letter to update GPs on participants’ progress and, when appropriate, whether or not GPs might consider further treatment. At the end of the programme, case managers also sent a participant-specific summary report to the GPs. Supervisors were available to advise case managers on next steps and consultations with GPs. Three letters were prepared and sent over the 12 weeks, requiring approximately 30 minutes’ administration per letter.

Case managers were also charged with a duty of care to engage outside agencies (e.g. social services or in response to safeguarding issues) in situations where they became aware of abuse. However, the client group had a generally low level of need in this respect and this additional responsibility was not generally required.

To estimate personnel costs required to provide collaborative care (as intended within the manual), estimates of NHS unit costs were derived from national reference costs53 (Table 44).

Table 45 summarises the resources required over the 12-week programme of collaborative care and indicates our estimate of the direct cost for the base-case analysis. Without further information from implementation, the direct cost of collaborative care (based on the previous estimation within the manual) was estimated as £494.73. The base-case estimate is naive of the actual levels of collaborative care required by individuals and observed variation was explored in sensitivity analyses.

Consequences for health care by treatment group

Table 46 presents information collected from primary care practices on each patient, comparing summary statistics of those who accessed collaborative care and those who underwent usual care. In all cases (except GP telephone consultations), we observed that contact rates were lower in the collaborative-care group than in the usual-care group. Using a simple t-test indicated that the mean contact rates for GP home visits, practice nurse appointments and nurse telephone consultations displayed significant levels of difference.

This resource utilisation data have highly skewed distributions (similar to most count data) and also have the feature of a significant number of zeros (representing the large proportion of people who do not use any services). In robustly estimating the marginal effect of treatment assignment on each resource, specific analytical procedures are required. 57 Zero-inflated negative binomial regression58 was utilised to model the count variables while accommodating the excessive number of zeros. This approach found that treatment assignment was most strongly associated with differences in the number of nurse appointments and nurse telephone contacts (see Appendix 8 for full details). The results suggest that, relative to usual care, the rate of nurse appointments was 15% less (p = 0.072) and that of nurse telephone consultations was 47% less (p = 0.062) in the collaborative-care group.

Cost–consequences and total costs

By assigning unit costs (as presented in Table 44) to resource use, the associated individual-level costs of health-care contacts were estimated and means calculated by treatment group (Table 47). Excluding the cost of collaborative care, the average cost of health-care utilisation was £51.01 less in the collaborative-care group than in the usual-care group, suggesting an average cost offsetting of 10% off the base-case cost of collaborative care (£494.73).

Overall, the mean total cost in the collaborative-care group was £893.96 (95% CI £847.15 to £940.76) and the mean total cost in the usual-care group was £450.24 (95% CI £408.84 to £491.64). The implication is that the unadjusted incremental cost of collaborative care was £443.72 (95% bootstrapped CI £381.321 to £506.1115; n = 620) or the cost specifically for the sample available for the base-case cost-effectiveness analysis was £420.93 (95% bootstrapped CI £347.73 to £494.14; n = 448).

Health state utility by time point

The utility score for each participant was estimated from the responses to the EQ-5D at baseline, 4 months and 12 months. Table 48 presents a summary of the unadjusted utility scores by time point and treatment group.

Comparing utility scores at baseline, the treatment groups display a difference which suggests that the estimate of cost-effectiveness should control for baseline utility.

Quality-adjusted life-years

Quality-adjusted life-years were estimated by summing the time-weighted averages of the utility scores between each time point. Table 49 compares the summary statistics of unadjusted QALYs by treatment group.

This suggests that the difference between group means (i.e. unadjusted incremental QALYs) is 0.096 gained from collaborative care compared with usual care. However, this difference does not account for the between-group difference in baseline utility scores. To adjust for baseline utility, we applied ordinary least squares regression to explain QALYs, controlling for treatment group and baseline utility. Table 50 presents the outputs of the ordinary least squares regression.

Examining regression coefficients (and their respective level of significance), collaborative care significantly increases QALYs by 0.044 (p = 0.003) when controlling for baseline utility.

Ex post adjustments of the cost of collaborative care

The base-case analysis of the cost of collaborative care assumed that all participants received the full course of treatment (i.e. 8–10 sessions) whereas for usual care it was assumed that participants received no extra services beyond contacts within primary care. To provide a more realistic insight into how collaborative care might be implemented in real-world NHS settings and how demand varies with regard to individuals’ specific levels of need, all case managers were asked to log their activities with patients on PC-MIS, which has been designed for IAPT.

The number and duration of participant contacts with their case manager were contemporaneously logged on PC-MIS. It was noted whether these contacts occurred face to face or by telephone. Figure 19 summarises the distribution in the number of contacts. The total number of registered sessions was highly variable, with a modal number of eight sessions. Furthermore, a bimodal distribution seems apparent, suggesting that there may exist a certain subgroup of participants who were more likely to withdraw during the early stage of care. Information on referrals to IAPT services was not available within the control group.

FIGURE 19.

Number of sessions of collaborative care.

Accounting for variation in the numbers of contacts, the duration of contacts and how the contacts took place (i.e. face to face or by telephone), Table 51 provides adjusted ex post costs of collaborative care.

Examining real-time resource use for collaborative care as delivered in the context of this trial demonstrated an average cost of £223.70 (95% CI £221.55 to £225.87).

The cost-effectiveness analysis was repeated to account for the costs implied by the PC-MIS data and to summarise the findings. Figure 20 reproduces the CEAC, adjusting for the ex post costs of collaborative care. The updated ICER of £3328 per QALY suggests that collaborative care requires fewer resources than expected and may prove significantly more cost-effective if implemented.

FIGURE 20.

Cost-effectiveness acceptability curve using ex post costs of collaborative care.

This sensitivity analysis incorporates information that would be known only once a programme had been implemented but demonstrates that, should collaborative care be recommended by the NHS, the programme has the scope to become efficient through adjusting the resources required within any given context. However, the reason why participants’ service needs varied has so far not been established and requires further investigation.

Design

A semistructured interview study was used to gather in-depth information on the trial participants’ experiences of receiving the collaborative-care intervention alongside the case managers’ experiences of delivering the intervention. Interviews were conducted with participants in the trial (at the end of the intervention period) and case managers delivering the intervention during the pilot phase of the RCT.

Sampling

The sampling followed a pragmatic and purposive strategy. Initially, because of slow recruitment into the pilot phase of the trial, all eligible trial participants were approached for participation in the qualitative study. The second half of the sample was recruited by using a purposive sampling strategy to gain a varied sample of pilot trial participants representing the wider population of older people in terms of age, sex, education, mental health scores, trial arm and those who had withdrawn from the intervention. This was with the aim of achieving maximum variation59 to represent the views of the whole range of individuals who could potentially be offered the intervention, rather than to consider any similarities or differences between any subgroups of the sample by broad characteristics (e.g. age and sex). Previous studies have indicated that a sample of approximately 40 trial participants would be sufficient to address these aims from the point of view of service users. 60 All case managers working in York and Leeds at the time of the pilot phase were recruited for interview by invitation through the local trial co-ordinators.

Sample

For the participant interviews, 39 older people were recruited during the pilot phase of the trial (Table 52). Of the 39 interviewees, five were in the usual-care group and another five had been selected for interview specifically because they had withdrawn from the intervention but had consented to remain in the trial. These interviews were to provide information on participants’ motivations for participation and reasons for withdrawal.

Case managers in York were based at the university whereas case managers in Leeds were based in the community mental health team. All 18 case managers were female and aged between 27 and 50 years. All case managers had been trained as NHS PWPs as part of the IAPT initiative. They each had several years’ experience of delivering low-intensity psychological interventions. In addition, two of the case managers had been involved in training PWPs. They were also involved in training case managers for the CASPER trial and in their supervision.

Consent

All CASPER trial participants were eligible for participation in the qualitative study. Following sampling, study participants were approached by letter, which contained an information sheet and two consent forms. In accordance with ethical guidelines (see Chapter 3, Approvals obtained), informed consent was gained by the researcher before the commencement of the interview and after the aim of the interview was explained and the participant had an opportunity to ask questions about the study. Anonymity and confidentiality of participants’ personal information were assured by the researcher.

Case managers were invited to take part in the interviews. They received an invitation, information sheet and consent forms in the post. This was followed up by a telephone call. Case managers were given an opportunity to ask questions about the study and were assured anonymity and confidentiality. Similar to the process for trial participants, consent was subsequently obtained at the time of interview.

Data collection

The semistructured interviews with participants were carried out in the home between January and September 2012 and on average lasted for 90–120 minutes; they were conducted using a topic guide (see Appendix 10). All interviews were audio recorded, transcribed and anonymised before data analysis.

The semistructured interviews with case managers were carried out in 2011 and 2012 in a private room on university premises (York) or on premises where the community mental health team was based (Leeds). The interviews lasted between 60 and 90 minutes and were conducted using a topic guide (see Appendix 9).

Data analysis

Initial thematic analysis61 was carried out using the qualitative data analysis software package ATLAS.ti 6 (Scientific Software Development GmbH, Berlin). 62 An initial coding framework was developed based on a priori themes relating to issues included in the topic guide while allowing for emergent themes. Descriptive coding was conducted, following familiarisation with the data, by the main qualitative researcher on the project (FZ), informed by regular discussion with the qualitative supervisors (KS and JA). Subsequently, initial codes were refined to address the aims of the qualitative study outlined earlier, following the analysis of the main trial. A constant comparison method63 was used to check and compare across the data set and to establish appropriate analytical categories. This also ensured that any additional categories were added to reflect as many of the nuances or outlier views in the data as possible, taking into consideration the participants’ wider contexts. Anonymised participant IDs are used for the reporting of results; those starting ‘CM’ indicate a quote from a case manager.

Our method of analysis was particularly suited to the project as specific questions were being addressed through the qualitative data. The main researcher (FZ) was immersed in the data by listening to the recordings and reading transcripts to list key ideas and recurrent themes. Memos were used during the initial stages of analysis to provide a visible ‘audit trail’ as the analysis moved from ‘raw’ data to the production of findings. Although the main qualitative researcher was responsible for the descriptive coding of the data, analysis was discussed with the wider qualitative team (KS, JA and PK).

To promote quality, the following strategies were used: description of the participants to provide context (credibility and transferability), transparency of the research process (transferability), evidence of consistency using multiple examples from data (dependability) and engagement of the wider research team with interim findings (confirmability). 64 In addition, a reflexive approach was taken to data analysis, with the main interviewer (FZ) being an academic research fellow from a nursing background with research experience in the topic of ageing but with no previous knowledge or experience of collaborative care or RCTs. The other members of the qualitative team (KS, PK and JA) all have academic research backgrounds (KS also has a nursing background) with no previous knowledge or experience of collaborative care. This placed the qualitative research team in a very neutral position with regard to any prior expectations relating to the study intervention.

Lack of engagement with the intervention

The perceived utility of the intervention, in terms of mental well-being, was directly related to the level of engagement with the intervention. This was observed for the patients themselves and articulated by the case managers:

I suppose it depends really how much they want to sort of engage with it. You know, that’s, that’s the thing that I’ve noticed. I mean if somebody, if, if you’ve got a participant that really wasn’t, you know, motivated and they didn’t really sort of know what else they could do, I could see that, you know, helping them, you know, with suggestions and things would be helpful, but then, there again it’s, it’s how much they want to engage in doing these things, isn’t it. You know, each individual would be different.

CMQ2

A significant proportion of participants felt that they could not relate to the intervention being tested. There were a number of possible reasons for this, as discussed in the following sections.

Positive aspects of the intervention

Many participants had really positive experiences of the intervention and could see direct benefits to their mental well-being. According to individual participants, these came in several forms. As a case manager commented:

I think for some people it’s been an eye-opener, and for some it’s been a confirmation, that what they’re doing is right, and that they’re OK. And some have been like, you know, a light bulb has turned on.

CMQ2

Intervention delivery

In this section we present the findings relating to aspects of intervention delivery and highlight any issues that would facilitate the implementation of the intervention on a larger scale.

Strengths and weaknesses

The qualitative sampling approach succeeded in achieving a broad range of views and experiences of the collaborative-care intervention. We achieved the aims of our purposive sampling by incorporating men and women of different ages, from different social backgrounds and with a range of depression scores. It was particularly important to include the views of those who withdrew from the intervention to provide insight into the possible reasons for this. By including all case managers, based across two different trial centres, we were able to obtain a range of views on the delivery of the intervention. However, the qualitative sample was obtained from the pilot phase of the CASPER trial only and therefore the views and experiences expressed obviously reflected the start of the process; these views may have changed during the main trial, particularly for case managers with greater experience in the role.

Limitations

The results of the CASPER trial need to be considered in the light of limitations that emerged during the study. First, regarding trial design, blinding was not feasible, which means that there was potential for contamination at the GP level as well as at an individual level. Each participating GP practice treated an average of 22 participants (minimum 2, maximum 83) and it is possible that GPs, who needed to be aware if a patient was currently receiving collaborative care, managed cases differently based on their knowledge. Furthermore, many participants would be living geographically close to one another in the same catchment area. Within a population of that age, it is reasonable to assume that some would know each other and share their trial experiences. In either case, we anticipate that contamination would result in additional benefits to control-arm participants, thereby reducing any group differences during follow-up and rendering our result a conservative estimate of the treatment effect. Also relating to study design, participants were recruited by means of postal screening from age-based GP practice lists and so participants identified with subthreshold depression had not necessarily presented with this problem. Therefore, the results of the CASPER trial might not automatically apply to older people who screen positive for depression in the context of primary care attendance or physical health checks for older people.

Retention and differential attrition between the trial arms was a further limitation. Although follow-up rates were high overall (83% at 4 months) and exceeded the anticipated trial retention on which the trial was powered, there was a much higher rate of attrition in the collaborative-care arm than in the usual-care arm (24% in the collaborative-care arm and 10% in the usual-care arm). This was in part accounted for by a number of participants who disengaged from the collaborative-care intervention and fully withdrew from the trial at the same time. It remains possible, however, that the patients who withdrew from the trial and who did not provide outcome data may have presented a very different outcome profile from those who continued, which may have biased the treatment effect. Based on the very similar baseline characteristics between randomised patients and those available for the primary analysis as well as our exploration of the impact of missingness, such bias appears less likely, however. We did adapt our withdrawal procedure when it came to light that participants disengaging from the intervention were actually tending to withdraw in full. In the early stages of the study, participants wishing to withdraw were able to do so through their case manager. Once we had identified the problem we changed the procedure so that case managers could no longer withdraw participants. The case managers would inform the research team if a participant wished to withdraw and the research team would make contact with those participants to discuss which level of withdrawal they required and to encourage them to remain in the study for follow-up. This proved successful and rates of retention to follow-up improved following this procedure change.

There are a couple of limitations relating to the cost-effectiveness analysis. First, it was not possible to include antidepressant use in the cost calculations. This was because of the high levels of polypharmacy in older adults, which impacted on the level of medication data obtained from GP practices. Some participants were prescribed large quantities of products during the 12-month follow-up period and any who exceeded 20 products were subjectively censored at the practice level, meaning that the level of detail to cost pharmaceutical use (e.g. product name, dose, duration of treatment, schedule of use) was not available. It was therefore not possible to obtain a reliable overall cost related to the use of prescribed medicines, or specifically antidepressants, from the available data. Additional learning from the process of collecting objective data relates to the mechanism of obtaining the data. Our experience was time-consuming and labour intensive, requiring data on the number of practice contacts for each patient to be manually copied and pasted into a document. In the future it would be helpful if the process could be made more efficient and effective by working with practices to develop ways of exporting anonymous objective data on a patient’s resource use directly from GP practice database systems (such as SystmOne, EMIS Web and Vision). This would make the task less onerous for time-deprived GP practices and provide a richer set of objective data, allowing for more accurate future cost-effectiveness analysis. It should also be recognised that the results presented are of the within-trial cost-effective analysis with an explicitly stated 12-month time horizon and without introducing external assumptions required to extend our results to a lifetime horizon. However, we have aimed to provide sufficient detail to inform a future state transition decision model. A further limitation of the cost-effectiveness analysis is that the reported economic evaluation considers only primary care NHS costs. Future cost-effectiveness estimates would benefit from having a wider perspective, factoring in secondary and tertiary care costs, along with unpaid costs such as family carers and volunteering.

The lowering of the age limit from the pilot phase to the main trial, from ≥ 75 years to ≥ 65 years, is likely to have affected the results as the proportion of ‘older’ older adults will have been reduced. It is also recognised that the qualitative findings are not representative of the entire trial. The data were collected during the pilot phase when the intervention was in its early stages of development. Had interviews been conducted with participants in the main trial, the findings may have been different. It is likely that, as case managers’ confidence grew through becoming more accomplished at delivering the intervention, participants would have experienced a more established intervention, which could have impacted on them more positively.

A final limitation is the truncation of follow-up at 12 months. Although we now know collaborative care to be effective in the medium term, the longer-term benefits beyond 12 months remain unknown.

Implications for health care

Collaborative care was acceptable for many of the older adults with low subthreshold depression and could readily be delivered over the telephone following a first face-to-face meeting. However, the experience of using IAPT workers to deliver the intervention demonstrated that older adults were not the priority of IAPT services. As a result, it may be worth exploring other methods of delivering the intervention such as through nurses who conduct comorbidity checks or healthy living workers. Certainly, the provision of care for older people with subthreshold depression would require substantial expansion in the scope of IAPT services if this intervention were to be delivered through that mechanism.

Collaborative care proved clinically effective at improving depression scores and preventing the onset of case-level depression for older people with subthreshold depression. The small to moderate effect size of 0.3 may represent limited change at the individual level but it has a substantial impact at the population level. 48 Moreover, the robust cost-effectiveness estimates for the use of collaborative care to treat subthreshold depression were cost-effective under conventional willingness-to-pay thresholds. This study has shown that collaborative care represents a new way of treating subthreshold depression in primary care. Depression is a relatively common condition, affecting about 5% of older adults. Given that subthreshold depression affects a larger proportion of the population, around 15–20%, this trial’s findings demonstrate the scope for collaborative care at the population/epidemiological level if it were to be rolled out into primary care. The CASPER trial evidence could be used by policy-makers and primary care to improve services and reduce the disease burden of our ageing population.

A final implication for health care relates to the relatively high dropout rate from the collaborative-care arm and what this would mean for take-up of the intervention in the real world. Some participants found the intervention intrusive and felt that talking and thinking about their symptoms made them feel uncomfortable. This may signal a potential problem if collaborative care were to be offered as part of NHS services. As with all psychological services, this type of intervention will not necessarily suit everyone and care should be taken to ascertain the likelihood of this being the case before any referral to such a service.

Recommendations for research

• Investigate the longer-term effect of collaborative care on relapse and the prevention of future case-level depression. Extended follow-up of CASPER participants, potentially at 60 months, represents a possible method to examine the longer-term effect of collaborative care, to see if the clinical benefits identified at 12 months were maintained over a longer duration.

• Investigate the longer-term effect of collaborative care on multimorbidities. A large proportion of participants in the CASPER trial had at least one long-term physical health condition. Although there were some improvements in function and quality of life among participants, there is little evidence on the effectiveness of collaborative care at treating comorbidities. Evidence from a US trial71 that tested collaborative care for the treatment of comorbid depression or subthreshold depression and diabetes mellitus showed that it helped improve depression care and outcomes but did not result in improved glycaemic control. Given that the population is ageing and the rate of long-term conditions is rising, there is demand for future trials to investigate the effectiveness of collaborative care at improving physical and mental health outcomes in older adults with multimorbidities.

• Many patients in the collaborative-care arm discontinued treatment or dropped out of the trial. Further qualitative and quantitative work should explore the reasons for this and identify the most appropriate target population for the intervention.

• Translating the research findings into clinical practice will be challenging and would benefit from further research. This relates to both enabling capacity to deliver the intervention to patients, and being able to target it at those most likely to complete the process and make use of the resource.

• There are no trials of collaborative care for people of working age with subthreshold depression. It would be useful to decision-makers to know whether or not the results of the CASPER trial can be replicated in this population.

Papers

Mitchell N, Hewitt C, Adamson J, Parrott S, Torgerson D, Ekers D, et al. A randomised evaluation of CollAborative care and active surveillance for Screen-Positive EldeRs with sub-threshold depression (CASPER): study protocol for a randomised controlled trial. Trials 2011;12:225.

Lewis H, Hems D, Bosanquet K, Overend K. Editorial: is enough being done to treat depression in the elderly? Age Health 2013;9:243–5.

Pasterfield M, Bailey D, Hems D, McMillan D, Richards D, Gilbody S. Adapting manualized behavioural activation treatment for older adults with depression. Cogn Behav Ther 2014;7:e5.

Bosanquet K, Mitchell N, Gabe R, Lewis H, McMillan D, Ekers D, et al. Diagnostic accuracy of the Whooley depression tool in older adults based in primary care. J Affect Disord 2015;182:39–43.

Gilbody S, Lewis H, Adamson J, Atherton K, Bailey D, et al. Effect of collaborative care vs usual care on depression symptoms in older adults with subthreshold depression. The CASPER randomized clinical trial. JAMA 2017;317:728–37.

Posters

Lewis H, Mitchell N, McMillan D, Bosanquet K, Hall R, Lilley A, et al. Collaborative Care in Screen Positive Elders – the CASPER trial: a randomised controlled trial pilot study. Presented at the UK Primary Care Mental Health Research Conference, Bristol, 2012.

Bosanquet K, Mitchell N, Lewis H, Bailey D, Gabe R, McMillan D, Gilbody S. Diagnostic accuracy of Whooley depression tool in older adults based in primary care in the UK. Presented at the UK Primary Care Mental Health Research Conference, Manchester, 2013. Won best academic poster award.

Traviss G, Holmes J, Lewis H, Mitchell N, McMillan D, Hems D, et al. CASPER – CollAborative care for Screen Positive EldeRs. Presented at the British Psychological Society Annual Conference, Harrogate, 2013.

Radio broadcasts

BBC1 ‘Inside out’ North East focused on care for older adults and interviewed Simon Gilbody, Deborah Jane Foster and Della Bailey regarding the CASPER and CASPER Plus trials; BBC Radio York interviewed Simon Gilbody regarding mental health and older adults in connection with the CASPER study; BBC1 ‘Look North’ focused on depression in older adults, interviewing a GP whose practice participated in the CASPER study, 15 October 2012.

Classifying level of risk

Case managers determine a participant’s level of suicide ideation by asking the risk questions provided in the CASPER case manager manual. This will identify whether the suicide ideation was in the past or is in the present (current) and whether it is active or passive. As such, there are four types of risk: past passive, past active, current passive and current active. The procedure for reporting these risks differs – past passive, past active and current passive risks follow a similar procedure; current active risk needs reporting urgently and follows a different procedure.

Reporting risk: non-urgent (past active, past passive and current passive suicide ideation)

In cases where suicide ideation was in the past tense, whether or not it was passive or active, case managers need to document it in the participant’s notes on PC-MIS. There is no need to communicate past risk immediately to either the participant’s GP or their clinical supervisor. Rather, case managers should inform the GP in writing in the next letter they have to send as part of collaborative care progress updates. Similarly, case managers can wait until their next supervision session to raise it with their supervisor; there is no need to contact the supervisor directly. This procedure also applies to current passive risk where no level of current intent exists. However, if in any doubt, always ‘err on the side of caution’ and classify risk as greater than it may actually be. Remember to explain to the participant that you will inform their GP next time you send a progress update letter to them. If the participant objects to this, discuss with your clinical supervisor at the next supervision session who will advise you; in some cases it may not be necessary to inform the GP of past suicide ideation. However, any current passive suicide ideation needs to be communicated to the GP in the next letter to the GP.

Reporting risk: urgent (current active suicide ideation)

If a participant presents current active suicide ideation, the case manager is to explain that they are concerned about the participant’s well-being and will have to inform their (participant’s) GP. First, they must immediately report risk to their clinical supervisor. If no answer to telephone call, text the clinical supervisor’s mobile, alerting them to suicide risk. If unable to contact clinical supervisor, contact GP directly without clinical advice. If participant’s named GP cannot be reached, speak to another GP from the practice. Inform the GP of the participant’s suicide ideation, explaining that it is now their decision as how to proceed with the participant’s care. Remember that even if the participant has objected to you informing their GP, in cases where there is current active suicide ideation the participant’s wish must be overridden. Document the procedure in the participant’s notes on PC-MIS. Fill in the suicide ideation form (overleaf) to be signed off by the clinical supervisor. E-mail the clinical supervisor, chief investigator and trial manager to confirm that participant ‘x’ has presented suicidal ideation, which has been dealt with following the reporting risk procedure. In addition, case manager to send a letter to the GP as confirmation of reporting risk telephone call.

Exploring other risk areas

• Risk from others includes abuse of elders and can take many forms including physical, psychological, emotional, sexual or financial abuse. If risk from others is detected follow Procedure 4, Safeguarding Older Adults.

• Self-neglect does occur sometimes in older adults. Examples include not eating or caring for one’s physical needs. If concerned about self-neglect contact local site clinical supervisor/principal investigator.

• Cognitive impairment does occur in older adults and typically presents as loss of memory. If concerned about cognitive impairment contact local site clinical supervisor.

(PDF download)

Introduction and background

• Background information on participant (e.g. age, general health, health problems that they are currently experiencing and how these affect their everyday life).

Understanding mental well-being

• Study participant’s understanding of mental well-being, where they have gained this information from (e.g. family, friends, professionals, media), how much of a problem they perceive this to be.

• Their thoughts on the relationship between mental well-being and any current health problems they are currently experiencing.

• Extent to which mental well-being is an important issue to them.

• Views about how they manage their mental well-being (i.e. keeping their spirits up, feeling fed-up) including medication management.

Views of their personal mental well-being

• Understanding of ‘risk factors’: what kind of things do they think might cause older adults, friends or themselves to have poor mental well-being (e.g. whether they mention poor social support, loss of friends/family, death of partner, poor health).

• Experience of being identified as experiencing subthreshold or above-threshold symptoms (e.g. how did this compare with what they were expecting – whether they expected not to be experiencing symptoms, thought their feelings were part of normal ageing).

• Thoughts on collaborative care:

  • advantages of collaborative care (ease of delivery, method of delivery, meeting with case manager, adherence to collaborative care, maintenance over time, usefulness of collaborative care for mental well-being/future behaviour, practical issues)

  • disadvantages of collaborative care (ease of delivery, method of delivery, meeting with case manager, adherence to collaborative care, maintenance over time, usefulness of collaborative care for mental well-being/future behaviour, practical issues)

  • thoughts on mental well-being and health-care professionals having had collaborative care [any change in beliefs about mental well-being (positive/negative/no change) having been involved in collaborative care].

• Views and feelings about future mental well-being.

• Views about how well they feel that they cope with difficulties in their life and in comparison to their friends.

Participating in the study

• Views of their experience of being involved in a study like CASPER.

• Views about being randomised to the intervention group (e.g. experiences of randomisation process and understanding of this, feelings about completing questionnaires for the study).

• Views about value of the trial.

Any other issues

• Any other issue the participant would like to raise.

• Experiences of taking part in the interview.

• Clarify what next in terms of CASPER study.

• Thank them for their time.

Introduction and background

• Background information on participant.

Understanding mental well-being

• Study participant’s understanding of mental well-being.

• Their thoughts on the relationship between mental well-being and health problems.

• Extent to which mental well-being is an important issue to them.

• Views about how they manage mental well-being, including medication management.

• Understanding of ‘risk factors’: what kind of things do they think might cause older adults, friends or themselves to have poor mental well-being (e.g. whether they mention poor social support, loss of friends/family, death of partner, poor health).

• Thoughts on collaborative care:

  • advantages of collaborative care (ease of delivery, method of delivery, adherence to collaborative care, maintenance over time, usefulness of collaborative care for mental well-being/future behaviour, practical issues)

  • disadvantages of collaborative care (ease of delivery, method of delivery, adherence to collaborative care, maintenance over time, usefulness of collaborative care for mental well-being/future behaviour, practical issues).

• Views and feelings about future mental well-being.

Participating in the study

• Views of their experience of being involved in a study like CASPER.

• Views about value of the trial.

Any other issues

• Any other issue the participant would like to raise.

• Experiences of taking part in the interview.

• Clarify what next in terms of CASPER study.

• Thank them for their time.