A systematic review of the clinical effectiveness and cost-effectiveness of pharmacological and psychological interventions for the management of obsessive compulsive disorder in children/adolescents and adults

Diagnostic criteria: disease classification

The first two versions of the Diagnostic and Statistical Manual of Mental Disorders (DSM) of the American Psychiatric Association were heavily influenced by the psychodynamic concepts of mental illness and defined obsessive and compulsive phenomena accordingly. In the first edition (DSM-I), the term ‘Obsessive Compulsive Reaction’ was used; the term ‘reaction’ referred to the way in which a person reacts to unconscious intrapsychic conflicts using defence mechanisms. 3 This was classified under the broader category of ‘Psychoneurotic Disorders’. In the second edition (DSM-II), the term ‘Obsessive Compulsive Neurosis’ was used. 4 Next editions of the manual abandoned the effort of classifying mental disorders in accordance with aetiology and adopted an atheoretical model based on descriptive phenomenology and empirical research. This was mainly done to increase the reliability of psychiatric diagnosis. The World Health Organization (WHO) followed along the same path and published clinical descriptions and specific diagnostic criteria for research. 5 From 1980 (DSM-III) to 2000, when the fourth edition of DSM was published (DSM-IV), there were few differences in the conceptualisation of OCD. 6 The disorder is classified under the broad category of ‘Anxiety Disorders’ and the two main characteristics are the presence of either obsessions or compulsions. Obsessions are defined as recurring and persistent thoughts, images or impulses that are intrusive and inappropriate and cause much distress and anxiety. Owing to their content, the patient tries to resist and control these thoughts or to suppress the resulting anxiety with compulsions. These are repetitive behaviours or mental acts that may aim to reduce the anxiety brought on by the obsession or that the person feels driven to perform in accordance with a rigid sequence or idiosyncratic rules. Quite often, these behaviours are not connected in a realistic way with what they aim to neutralise or prevent, or they are clearly excessive. In order for these symptoms to be considered clinically significant, they should have a considerable impact on the everyday functioning of the individual.

In the latest edition of the DSM (DSM-V),7 there have been some slight changes to the definition of the disorder, some of the most important of which are the following:

1. OCD has been separated from the broader category of ‘Anxiety Disorders’ and it is now described under the category of ‘Obsessive–Compulsive and Related Disorders’ which also includes body dysmorphic disorder, hoarding disorder, trichotillomania and excoriation disorder.

2. It is acknowledged that there is a spectrum of insight and that some patients may have absent insight or ‘delusional’ beliefs. Therefore, in DSM-V insight is coded as ‘good/fair’ (obsessions are recognised as excessive and abnormal ideas), ‘poor’ (obsessions take the form of overvalued ideas) or ‘absent’ (obsessions share some characteristics with delusions). An absent insight does not preclude the diagnosis of OCD. As Leckman et al. 8 point out, it is assumed that patients with currently absent insight have shown some insight in the past during the course of their disorder.

3. Hoarding disorder is now a separate disorder and not a subtype of OCD.

4. A form of OCD related to chronic tics is now included as a new subtype, as there is evidence that this subtype has a younger age at onset and different treatment response. 8

5. The definition of OCD according to the International Classification of Diseases, Tenth Edition (ICD-10), is very similar to that of the DSM, with slight and not essential differences. 8

6. The ICD-10 does not include separate definitions of obsessions and compulsions but the emphasis is placed on their shared characteristics.

7. The DSM implies that obsessions and compulsions have a functional relationship (in the sense that compulsions are behaviours that aim to reduce the distress caused by the obsessions), whereas the ICD does not suggest such a connection.

Form and content of obsessions and compulsions

Previous studies of the phenomenology of OCD have described both the form and the content of obsessions and compulsions. Khanna et al. 9 described the following forms for obsessions (in order of frequency): fears, thoughts, doubts, urges, convictions and images; and, for compulsion: repeating, rituals, checking and avoiding.

The thematic content of obsessions and compulsions has been described in detail by Rasmussen and Eisen10 and Foa et al. 11 Rasmussen and Eisen10 have used data from their large cohort of 560 OCD patients (diagnosed in accordance with DSM-III or DSM-III-R criteria), whereas Foa et al. 11 reported data from 425 patients with OCD (in accordance with DSM-IV criteria). Although there are some differences between these two studies, it is generally accepted that common themes of obsessions are (all figures from Foa et al. 11) worries about dirt/contamination (40%), aggressive obsessions (25%), content related to sexual or religious themes (12%), need for symmetry (10%), somatic/hypochondriac concerns (7%) and unacceptable urges (4%). In the Rasmussen and Eisen10 cohort, pathological doubt (regarding responsibility for a terrible event) was the second most common content, reported in 40% of patients (this theme is related to aggressive ideas or harm-related content in the Foa et al. 11 study). Regarding compulsions, common themes include checking (28%), cleaning/washing (27%), repeating/counting (13%), mental rituals (11%), ordering/symmetry compulsions (6%) and hoarding-related compulsions (4%).

Given that there may be a functional relationship between obsessions and compulsions, some studies have investigated the structure of symptoms using the statistical techniques of factor or cluster analysis. Recently, Bloch et al. 12 carried out a meta-analysis of all studies that used factor-analytic methods to investigate the symptom structure of the most commonly used symptom scale, the Yale–Brown Obsessive–Compulsive Scale (YBOCS) (n = 21 studies involving 5124 participants). They concluded that four distinct factors explained 79% of the variance in the total sample: (1) a symmetry factor, which included symmetry obsessions and ordering, repeating and counting compulsions; (2) a factor associated with ‘forbidden’ thoughts, which included aggressive, sexual and religious obsessions; (3) a cleaning factor, including dirt/contamination obsessions and cleaning compulsions; and (4) a hoarding factor. The results were quite similar in both the children and adolescents and adults subsamples.

Although most patients have a main/primary theme for their obsession or compulsion, it is not uncommon to report other themes of milder intensity or frequency. 9 Mataix-Cols et al. 13 investigated the longitudinal stability of symptoms in 117 adult patients and concluded that symptoms were quite stable at 2 years’ follow-up and that shifts between symptom dimensions were relatively rare.

Phenomenological differences between the two genders

The phenomenological differences between the two genders have been recently reviewed by de Mathis et al. 14 Men are younger at onset, and this is sometimes associated with the presence of chronic tics and a worse prognosis. Most phenomenological studies conclude that men are more likely to develop obsessions with a sexual/religious theme, whereas women more often develop dirt/contamination obsessions and cleaning compulsions. 15,16 Some studies also report that symmetry/ordering obsessions are more common in men,15,17 whereas in women the presence of obsessions (either fears or impulses) with an aggressive content may be more common. 15,18

Phenomenology in children and adolescents

Obsessive–compulsive disorder may start very early in childhood,19 and it is interesting to investigate differences in the presentation of symptoms between children and adults. Geller et al. 20 compared symptom dimensions in a sample of 101 patients aged < 18 years, including a subsample of children (n = 46) and a subsample of adolescents (n = 55), and compared this with a reference adult patient sample previously reported in Rasmussen and Eisen. 21 Regarding obsessions, they reported significant differences between both the children and adolescents samples and the adults sample as regards the presence of aggressive/catastrophic obsessions (less common in the adult sample) and sexual/religious obsessions (which were more common in the adolescent sample). In addition, they reported that confessing/asking compulsions were more common in children. Similar studies have also confirmed that contamination obsessions and washing/cleaning compulsions are very common in children and adolescents. 22,23

Measurement of disease severity

Several instruments have been developed to assess symptom severity in OCD. These include both clinician-administered interviews and self-report questionnaires (or parent report in the case of the paediatric population). Grabill et al. 24 included four clinician-administered instruments and 10 self-report questionnaires in their review of this issue.

Of the clinician-administered instruments, the YBOCS25 is the most widely used instrument to assess symptom severity and is considered the gold standard in OCD literature, especially to assess change in symptoms after treatment. 26 This is a semistructured clinician-administered instrument assessing the severity and frequency of obsessions and compulsions. It yields three scores, an obsessions severity score, a compulsions severity score and a total score (ranging from 0 to 40 for the total). Good psychometric properties have been reported for both clinical and non-clinical samples of patients. 24 A cut-off score of 16 is often used in clinical trials for patients to be eligible for inclusion in a study. This score distinguishes patients with moderate-to-severe symptoms from patients with mild or subclinical symptoms and has demonstrated good sensitivity. 26 Other versions of the YBOCS include a self-report version26 and a modified version (YBOCS-II)27 to take into account more recent research on the phenomenology of OCD. An adaptation of the same scale for children and adolescents has been also developed, the Children’s Yale–Brown Obsessive–Compulsive Scale (CYBOCS),28 which has been used extensively in paediatric OCD trials. Studies of the factor structure of the YBOCS have generally confirmed the existence of two factors (severity of obsessions and severity of compulsions), although a third factor of resistance has been replicated in some studies. 24,29

Aetiology

Obsessive–compulsive disorder is a complex neuropsychiatric disorder, and several genetic, biological and psychological factors may have an important role in the aetiology of the condition. Although aetiological research in the OCD field is very active, the clinical heterogeneity and complexity of the disorder have resulted in the limited translational capacity of basic research into clinical practice. 30

Epidemiology

Comorbidity

Several studies with clinical samples have confirmed that OCD is often comorbid with other psychiatric disorders. 66–68 In most clinical studies, the most common comorbidity is mood disorders, a finding that is compatible with the view that patients with OCD will often seek help from a mental health specialist when they develop depression or some other psychiatric disorder. The reported prevalence of comorbid disorders differs from study to study depending on the methodology and the time frame used (e.g. 1 month, 1 year, lifetime). In a Dutch sample of 420 outpatients with OCD,67 24% had comorbid current depression/dysthymia and 13% had any anxiety disorder [most often social phobia (3.6%) and panic disorder (2.6%)]. These figures are much higher than the reported prevalence in the general population. One study reported that alcohol use disorders in patients with OCD were less common than in the general population. 67 Another study from the USA, which included 334 outpatients from the adult OCD clinic at the National Institute of Mental Health,66 assessed the lifetime prevalence of comorbid disorders. Approximately 66% of patients had experienced major depression at some time in their lives, whereas one in four had experienced social phobia, panic disorder or dysthymia. It is worth noting that in this cohort, lifetime alcohol dependence was high, at 25%. In women, the prevalence of eating disorders was increased (26% of the sample). Data from the Brown cohort68 showed that a minority of patients (< 10%) had not experienced any other disorder in their life. As reported elsewhere, depression was the most common comorbid condition (67% lifetime prevalence and 15% current episode). Other common diagnoses in this cohort were social phobia (28% lifetime, 19% current), panic disorder (18% lifetime, 7% current) and alcohol dependence (23% lifetime, 4.5% current). Eating disorders were also common in this cohort (10% lifetime prevalence for the entire sample).

Studies conducted in the general population have confirmed that these patterns of comorbidity are not the result of help-seeking bias. In the British Psychiatric Morbidity Survey,52 37% of participants with OCD also met criteria for current depression. In addition, comorbidity with anxiety disorders, such as panic disorder (22%) and social phobia (17%), and with alcohol dependence (33% in men, 11% in women) was particularly high. In the replication study of the National Comorbidity Survey in the USA,69 participants who met criteria for OCD had an increased lifetime prevalence for other mood and anxiety disorders (40% for depression, 44% for social phobia, 20% for panic disorder, 38% for alcohol dependence). These figures are very similar to those reported from the first large epidemiological survey of OCD in the US general population, the ECA study. 46

The association of OCD with bipolar disorder has attracted research interest over the past decade. Several studies have reported that OCD patients have a lifetime history of bipolar disorder, mainly type 2 bipolar disorder (bipolar II), with prevalence rates that are much higher than in the general population (up to 15% in some samples70). Conversely, an OCD history is often reported in patients with bipolar disorder (up to 35% in a German study71). In a recent review of this issue,72 it is pointed out that there are disagreements between studies regarding the extent of this comorbidity. For example, in a French study that included mainly type 1 bipolar disorder (bipolar I) patients,73 history of OCD was quite low (at 3%), in contrast to histories of panic disorder and phobic disorders, which were higher (16% and 11%, respectively).

The relationship between obsessive–compulsive symptoms and psychotic disorders in general, or schizophrenia in particular, has been noted since the early twentieth century,74,75 but only recently has this association been studied more systematically. 72 A recent meta-analysis of this topic identified 37 studies that aimed to estimate the prevalence of OCD in patients with schizophrenia and related disorders. 76 This analysis reported a mean OCD prevalence of 12.3% (95% CI 9.7% to 15.4%), which is much higher than that of the general population. Obsessive–compulsive symptoms were even more common. Eisen et al. ,77 in a study that used a very careful methodology, reported that the prevalence of OCD in 77 patients with psychotic disorders (schizophrenia, n = 52; schizoaffective disorder, n = 25) was 7.8% (6/77 patients). It is worth noting, however, that five of these six patients with OCD had schizoaffective disorder (5/25, 20%) and only one had schizophrenia (1/52 patients, 1.9%). Another interesting study from the Netherlands, among patients with first-episode psychosis or at ultra-high risk for developing psychosis,75 reported that the prevalence of OCD was 1.5%, whereas that of obsessive–compulsive symptoms not meeting full diagnostic criteria for OCD was 9.3%. The authors note that these figures are very similar to those reported from general population samples. The prevalence of both the disorder and the symptoms was lower in those patients who met criteria for schizophrenia rather than schizophreniform or schizoaffective disorders. No significant differences were found between the time of onset of obsessive–compulsive symptoms prior to or after the onset of the first episode of psychosis. OCD did not precede the onset of psychosis in patients with both disorders. These findings are compatible with the view that obsessive–compulsive symptoms may be either prodromal symptoms of first-episode psychosis or a secondary side effect of antipsychotic medications. 75 Studies that have investigated the presence of psychotic symptoms among patients with OCD are few. In the Brown cohort,78 6% of OCD patients had a comorbid psychotic disorder (4% schizophrenia; 2% delusional disorder). In NEMESIS,79 the presence of obsessive–compulsive symptoms at baseline predicted the onset of psychotic symptoms at follow-up and vice versa. From these studies, it is concluded that although there seems to be an association between symptoms of OCD and psychotic disorders, this association is bidirectional and complex.

Suicidality

In the past, OCD was considered a relatively rare condition,19 with a low risk for suicide, at least compared with other mental disorders. 80,81 More recent studies, however, have changed this view. In Brazil, a large clinical study of outpatients with OCD (n = 582) found that 11% of the sample had attempted suicide at least once in their lifetime. 81 Studies in unselected samples of the general population have confirmed these findings. In the UK, a history of suicide attempt was reported by 26% of the participants who met criteria for OCD, compared with 14.5% of those with other common mental disorders, a statistically significant difference. 52 These studies were cross-sectional and assessed suicidality retrospectively. There are few longitudinal studies that have reported suicidal behaviour. In a prospective clinical study in Spain,82 218 outpatients with OCD were followed up for a mean duration of 4 years. Two patients (0.9%) committed suicide and 11 (5%) attempted suicide. Risk factors for suicidal behaviour were the presence of symmetry/ordering obsessions and the initial severity of depressive symptoms. In NEMESIS,83 the cumulative incidence of suicide attempts in participants with OCD, after 3 years’ follow-up, was 0.4%, a very low figure compared with other common mental disorders and the lowest among the anxiety disorders. Incidence of suicidal behaviour was also low in a meta-analysis that used data from patients who had participated in randomised controlled trials (RCTs) submitted to the US Food and Drug Administration (FDA). 84 In that analysis, the annual incidence of suicide attempts in OCD patients was approximately 1.47% (1468/100,000 per year) and the suicide rate was 0.11% (105/100,000 per year). It is worth noting that these figures were comparable with the other anxiety disorders covered in this analysis and significantly lower compared with the figures for depression (2.9% and 0.8%, respectively) that have been reported in a separate paper with the same methodology. 85

In conclusion, the results of these studies show that suicide risk in OCD is higher than in the general population by a factor of 10 or more, but is comparable to risk in popuations with other anxiety disorders. It is lower than the risk associated with depression, but it should be pointed out that, because OCD is often comorbid with depression, the incidence in real clinical practice might be higher. In a secondary analysis of the NEMESIS sample, for example, the suicide risk (either attempt or ideation) was higher in participants who met criteria for both depression and anxiety disorders. 86

The study of suicidal behaviour in child/adolescent OCD samples is not so extensive. A recent study of 54 patients aged 7–17 years from a tertiary centre in the USA87 reported that 13% of the sample (n = 7) had clinically significant suicidal ideation during the past month. Significant associations were found with the presence of symmetry/ordering obsessions and obsessions of sexual or religious content, with increasing age and with the presence of depressive symptoms.

Natural history: prognosis

Historically, OCD was considered a disorder with a poor prognosis. This view was challenged by the seminal study of Pollitt. 88 Pollitt’s study used a very strict methodology in a period in which there was no specific form of treatment other than leucotomy (with uncertain effects). To avoid any possible treatment effects of this procedure, Pollitt presents his results on the course of OCD separately for patients with or without leucotomy. According to this study, the longitudinal course of the illness was good: complete remission was observed in 24% of the patients after a mean duration of follow-up of 3.4 years (range 0.5–15 years), whereas 36% had a mild illness (i.e. 60% of the patients had a benign course). A study conducted in Sweden by Skoog and Skoog89 included 144 patients with OCD who were examined by one of the authors between 1954 and 1956 by means of a semistructured clinical interview. The patients were re-examined by the same researcher after 40 years using the same methodology. Rates of complete remission (20%) were comparable to those in the Pollitt study,88 whereas 28% of patients had mild symptoms at follow-up. Therefore, in this study almost half of the patients had a good course.

With the advent of new and effective treatments in the 1980s [BT/CBT, clomipramine and selective serotonin reuptake inhibitors (SSRIs)], it is interesting to review longitudinal studies of the prognosis of OCD in patients who received such treatments. Two such studies have been published recently. Bloch et al. 90 investigated the longitudinal course of illness (10–20 years) in 83 patients with OCD who had participated in clinical trials in their centre (Yale). The authors reported that 20% of the sample experienced complete remission, whereas another 30% experienced partial remission. Almost half of the patients still had symptoms that would make them eligible for inclusion in a new clinical trial (a score on the YBOCS of ≥ 16). It is worth noting that 70% of these patients were receiving medication at follow-up and approximately half had received BT or CBT at some point in their lives (after the baseline assessment). Similar results were reported in the study of the Brown cohort that included 213 patients with OCD. 91 Complete remission at 5 years’ mean follow-up was observed in 17% of the sample, whereas partial remission was observed in 22% of patients. This study also assessed the rates of relapse after partial or complete remission, which were quite high (59%). In another study from Italy, which included 55 outpatients with OCD treated with SSRIs, the rates for complete and partial remission at 3 years’ follow-up were 22% and 34%, respectively. 92

The main conclusion that can be drawn from the above discussion is that remission rates in the modern era have not improved compared with those reported in earlier studies88,89 despite the wide availability of effective treatments. It is difficult, however, to interpret this finding, as changes in diagnostic preferences or criteria may have resulted in non-comparable groups of patients being selected for inclusion in these studies.

Regarding the factors that are associated with a poor prognosis, several studies report that an early onset, more severe initial symptoms, a longer duration of illness and comorbidity with depression are all associated with a poor prognosis. 89,92–95 In the Brown cohort,91 patients with primary hoarding obsessions/compulsions had a worse prognosis with very low remission rates, whereas patients with primary obsessions regarding an inflated sense of responsibility for harm had a better prognosis. In the Yale cohort,90 an initial good response to SSRIs was a good prognostic factor. In other studies, the presence of schizotypal92 or obsessive–compulsive personality disorder91 was associated with a poor prognosis.

The long-term prognosis of OCD in children and adolescents has been reviewed by Stewart et al. ,96 who included 16 studies from various settings. Stewart et al. 96 report a mean remission of 40% after a mean duration of 5.7 years’ follow-up. When including partial remission, this rate is increased to 59%. Focusing on the studies that have used non-clinical samples, the remission rate is even higher, at 74%. Some more recent studies from the USA97,98 and the UK99 have also been published. The results of more recent studies are similar to those seen in the Stewart et al. review,96 despite the use of selected samples from tertiary centres. In the Yale cohort,98 58% of the patients had complete remission after a mean follow-up duration of 9 years. In the Maudsley cohort,99 approximately 60% of the patients had at least a partial remission after 5 years’ mean follow-up. From these findings it can be concluded that remission rates in children and adolescents may be higher than those in adults. Regarding factors associated with a poor prognosis, the following have been reported: duration of illness,96,99 early onset96 and presence of hoarding obsessions/compulsions. 98 A better prognosis has been reported in patients with chronic tics98 and in patients who showed a good initial treatment response. 96

Impact on quality of life and functioning

As a chronic disorder, OCD can have a severe impact on everyday functioning and quality of life. Two systematic reviews have recently investigated the published literature on this issue. 100,101 Most studies have used clinical samples and compared several dimensions of quality of life in OCD and other psychiatric disorders, chronic physical disorders and the general population. Fewer studies have used non-selected samples in the community,101,102 but these have confirmed that OCD, even in individuals living in the community who have not made contact with services, can have a detrimental effect on quality of life compared with the healthy population. In a study in Asia, OCD was associated with a worse quality of life than in other common mental disorders. 103 Studies have reported that contact104 or relationships with family members103 may be more severely affected in patients with OCD than in patients with other mental disorders.

Most studies in clinical samples have compared quality of life in patients with OCD with that in patients with other mental disorders, chronic physical disorders or population norms. One of the most cited studies that compared the quality of life in patients with depression and anxiety disorders used data from patients who took part in several multicentre RCTs of sertraline. 105 According to this study, patients with OCD had a better overall quality of life compared with patients with depression and comparable to other anxiety disorders, with the exception of post-traumatic stress disorder. Olatunji et al. 106 carried out a meta-analysis of 33 studies that examined quality of life in patients with anxiety disorders. Six of these studies focused on OCD and their findings show that OCD is associated with a worse quality of life for patients than for the general population, but other anxiety disorders may have a more harmful effect (e.g. social phobia or post-traumatic stress disorder). In the clinical samples, the dimension of quality of life more severely affected in OCD is the one associated with social relationships. 101

Some studies have investigated quality of life as a long-term outcome in RCTs of psychopharmacological or psychosocial interventions. 100 These studies have concluded that changes in quality of life can be quite delayed and certainly are not expected in the short term. In the psychopharmacology trials, these changes may become evident after 1 year of continuous treatment. 107 It should be pointed out that because OCD is often comorbid with other disorders, in particular depression, this may lead to further worsening of quality of life and, generally, is an important factor in determining levels of functional impairment. 100

Studies conducted in child and adolescent samples are limited and assess functional impairment more often than quality of life. It should be noted that these are related but not identical concepts. 108 Two recent studies have investigated quality of life in children and adolescents with OCD. 108,109 Both confirmed the negative effect of the disorder on the quality of life of patients compared with healthy children. An important factor that predicted a worse quality of life was the presence of comorbid internalising or externalising disorders.

Treatment options

Current guidelines

Table 1 presents the most recent published guidelines for the management of OCD. The most recent clinical practice guidelines for the pharmacological treatment of OCD have been published by the British Association for Psychopharmacology. 111 The Canadian Anxiety Disorders Association has also published guidelines for both pharmacological and psychological interventions for all anxiety disorders including a separate section for OCD. 115 The American Psychiatric Association had recently updated its previous detailed clinical practice guideline (see http://psychiatryonline.org). 116 The World Federation of Societies of Biological Psychiatry published its guidelines for all anxiety disorders, including OCD, in 2008. 117 In addition, NICE published a very comprehensive clinical practice guideline for OCD and body dysmorphic disorder in 2005 (NICE CG31). 118 NICE has recently placed the 2005 OCD guideline on a ‘static’ list of guidelines that will not be reviewed again within the next 5 years unless there is important new evidence of either efficacy or safety. NICE has also published a quick reference guide, which includes a detailed version of their stepped care model for treating OCD in all age groups (see www.nice.org.uk/nicemedia/live/10976/29945/29945.pdf).

Medications

Pharmacotherapy with the tricyclic antidepressant clomipramine or a SSRI (paroxetine, fluvoxamine, fluoxetine, citalopram, escitalopram and sertraline) has shown efficacy in OCD. 119 Meta-analyses of seven RCTs of clomipramine120 and 17 randomised, double-blind placebo-controlled trials of various SSRIs121 have been performed. The trials were generally short term (i.e. of 4–12 weeks’ duration) and showed that all these compounds were superior to placebo. Patients were roughly twice as likely to respond to a SSRI as to placebo. Data on comparisons between different SSRIs and between SSRIs and clomipramine are limited but have shown no significant differences in efficacy. The SSRIs are recommended as the first-line pharmacological treatment for OCD, with clomipramine reserved for those who do not respond to or tolerate SSRIs, owing to the more favourable adverse event profile. 111,116,118 SSRIs tend to take longer to be effective (between 4 and 12 weeks) when used for OCD than when used for other disorders, such as depression and anxiety. A positive dose–response relationship has been observed with several SSRIs (paroxetine, fluoxetine and escitalopram), and higher doses of SSRI are often required. 116,118,119 Approximately 40–70% of patients respond to a SSRI, but the long-term improvement in total symptom severity is relatively low, averaging 20–40%,122,123 as is the remission rate, with full remission ranging from approximately 10%123 to 40%. 124 A long duration of untreated illness,94 coexisting tic116 and hoarding symptoms125 have all been associated with a poorer treatment response to clomipramine and the SSRIs.

The findings of acute treatment studies indicate that the proportion of responding patients increases steadily over time. Long-term (up to 12 months) double-blind RCTs demonstrate an advantage for continuing with medication in patients who have responded to acute treatment. 126,127 A randomised placebo-controlled trial with paroxetine as an active comparator found that a low dosage of escitalopram became efficacious only in the second half of a 24-week study. 124 Most (but not all) placebo-controlled relapse prevention studies in patients who have responded to previous acute treatment reveal a significant advantage for staying on active medication (escitalopram, fluoxetine at higher daily doses, paroxetine and sertraline), compared with switching to placebo, for up to 12 months,128 but the optimal duration of continuation treatment is uncertain. 129 For these reasons, it is recommended that clinicians continue drug treatment for at least 12 months in patients who have responded to treatment. 111 As approximately 50% of patients with OCD relapse if they discontinue medication after up to 1 year of successful treatment, it is advisable to counsel patients about the risk of relapse prior to drug discontinuation and, if relapse occurs, medication is usually reinstated and continued indefinitely. 130

Adverse events with medications

Selective serotonin reuptake inhibitors are generally safe and well tolerated, according to the placebo-referenced treatment trials that reported adverse event-related withdrawal rates of approximately 5–15%. As a group, however, SSRIs may cause unwanted nausea, insomnia, somnolence, dizziness and diarrhoea. Sexual side effects include reduced libido and delayed orgasm, and can affect up to 30% of individuals. 131 Fluoxetine has a long half-life and fewer discontinuation effects, which can be advantageous for patients who forget to take their tablets. It has also been extensively used in pregnancy and generally shown to be safe. 132 The recent demonstration of prolongation of the electrocardiogram QT-interval associated with higher dose levels of citalopram (and, to a lesser extent, escitalopram)133 argues for a degree of caution in using higher doses of these compounds in OCD, especially if individuals are taking other medications that increase the QT interval. However, a recent large study found no elevated risks of ventricular arrhythmia or all-cause, cardiac or non-cardiac mortality associated with citalopram doses exceeding 40 mg per day. 133

Clomipramine can also be associated with potentially dangerous side effects. Cardiotoxicity and cognitive impairment occur much more often with clomipramine than with SSRIs. In addition, there is an increased risk of convulsions in patients taking clomipramine (up to 2%). Overdose on clomipramine can prove fatal, and this needs to be borne in mind when prescribing for OCD, in view of the elevated suicide risk associated with the illness. Clomipramine is also associated with greater impairment of sexual performance (up to 80% of patients) compared with SSRIs, with weight gain and with troublesome anticholinergic effects. 134

Psychotherapy

The general principles of the psychological model of OCD have been described (see Aetiology). Based on this model, a number of psychological treatments have been developed. A comprehensive historical review of these treatments is given by Abramowitz. 137 Two main treatments have been developed, a behaviourally oriented treatment (ERP) and a treatment based on the cognitive model of OCD. 40

Exposure and response prevention was first described in a clinical setting by Meyer,138 and the relative success of this method soon replaced other behaviourally oriented methods such as Wolpe’s systematic desensitisation. 137 According to Abramowitz,137 in ERP, first, the patient undergoes prolonged exposure to situations or stimuli that provoke obsessional fears and, second, the patient is advised to refrain from performing the compulsive behaviour (response prevention). Response prevention helps the patient learn that anxiety will eventually decrease on its own over time and also that obsessions are not really dangerous or do have catastrophic consequences. 137 The intensity of the treatment differs, but typical forms of therapy include at least 16 sessions over 8 weeks. Some of the sessions are supervised by the therapist and the patient also practises self-exposure between sessions. 137 Owing to the nature of the treatment, some patients may not tolerate the distress associated with the exposure or they may not be willing to refrain totally from the ritualistic behaviour. Despite the difficulties in applying this treatment, ERP has established its effectiveness both in research and practical settings. 118

The cognitive model of obsessions is primarily based on the work of Salkovskis,40 who suggested that, although disturbing, intrusive and unacceptable thoughts (normal obsessions) are experienced by all,139 patients with abnormal obsessions appraise the intrusions in a pathological way. Salkovskis suggested that such appraisals ‘appear to relate specifically to ideas of being responsible for damage or harm coming to oneself or to others’. 40 Compulsions are viewed as efforts by the individual to prevent any harmful consequences or to reduce the unwanted intrusions. Apart from Salkovskis’ ‘inflated responsibility’, other faulty appraisals of intrusions have been described by the Obsessive Compulsive Cognitions Working Group,140 such as the overimportance of thoughts, the excessive concern about the importance of controlling one’s thoughts, the overestimation of threat and the intolerance of uncertainty. Based on the cognitive model, cognitive therapy (CT) that does not require the use of ERP techniques has been developed, although behavioural experiments are used to help patients modify their views about the risks associated with obsessions. 137 CBT for OCD combines both ERP techniques and cognitive restructuring. 137 There is now evidence that both therapies are effective in the management of OCD and have comparable efficacy to ERP with a slightly improved tolerability. 118

Population

Children and adolescents, and adults with OCD.

Intervention and relevant comparators

Any antidepressant medication with some serotonergic properties used in the management of OCD (including amitriptyline, imipramine, clomipramine, all SSRIs, all serotonin–noradrenaline reuptake inhibitors, mirtazapine and hypericum), BT (therapy that includes some kind of exposure and/or response prevention), CT or CBT and any drug/psychotherapy combination of these interventions. Comparators included drug placebo, psychological placebo, waitlist and any other comparator from the list of interventions that would allow an indirect comparison with network meta-analysis (NMA).

Outcomes

The primary outcome for effectiveness was the reduction in symptoms of OCD as measured at the end of the study period by the YBOCS scale (or the CYBOCS). The secondary outcome was acceptability, as measured by the total number of dropouts in each intervention arm.

Subgroup analyses

Regarding the preplanned subgroup analyses, where sufficient data were available, metaregression/subgroup analyses were conducted to explore the impact of:

1. publication date

2. length of trial

3. inclusion of patients with comorbid depression

4. pharmaceutical sponsorship of drug trials.

Identification of trials: search strategy

Study selection: inclusion and exclusion criteria

Studies were included (or excluded) in accordance with the following criteria:

1. Study design: RCT. Trials with a crossover design were not excluded and we tried to extract all available data up to the point of the crossover. Quasi-randomised trials (such as those allocating by using alternate days of week) were excluded. Owing to the aim of the review, we included trials irrespective of blinding (because otherwise a lot of psychotherapy trials might not be eligible for inclusion). Sensitivity analyses examined the possible effect of unblinded or single-blinded trials.

2. Patient population.

   • Age: all patients aged ≤ 74 years (if patients aged ≥ 75 years were included, mean age should be within the range).

   • Diagnosis: a primary diagnosis of OCD in accordance with standardised diagnostic criteria (ICD, DSM, Feighner or research diagnostic criteria). Studies that specifically focused on treatment-resistant OCD were excluded. Treatment resistance should have been defined within the study using specific criteria. Most often, studies will have used a first, uncontrolled, treatment phase, in which all patients received the same intervention and the non-responders (usually showing < 25% reduction in the YBOCS scale) were eligible for the second randomised phase. Studies that had included patients that could be considered refractory to treatment outside the context of the particular study (e.g. because they might have tried medications or other interventions in the past unsuccessfully) were not excluded. It is worth noting that most psychotherapy trials have included patients who were symptomatic despite being stable on medications before entering the study.

   • Comorbidities: these will be accepted if OCD was the primary disorder. However, studies that included patients with schizophrenia or bipolar disorder were excluded.

   • Diagnostic criteria: the authors should have used established diagnostic criteria to diagnose OCD (either ICD, or any version of DSM, or Research Diagnostic Criteria or Feigner criteria). The method of assessment of these criteria (either through typical clinical examination or use of more formal diagnostic interviews) was not a reason for exclusion.

3. Experimental intervention.

   • For pharmacological interventions: any antidepressant medication with some serotonergic properties (including: amitriptyline, imipramine, clomipramine, all SSRIs, all serotonin-noradrenaline reuptake inhibitors, mirtazapine). Studies that have used hypericum were included, whereas other non-standard approaches were excluded (e.g. studies that have used folic acid, herbal medicines other than hypericum, vitamins or omega-3 supplements). Studies that had used a mainly noradrenergic medication as the experimental intervention of interest [e.g. reboxetine (Edronax^®, Pfizer) or nortriptyline] were excluded.

   • For psychological interventions: we included trials that have used as their main intervention (1) BT (therapy that included some kind of exposure and/or response prevention); (2) CBT; or (3) CT (therapy that included some kind of cognitive restructuring intervention). We excluded studies that used therapies based on psychodynamic principles (including interpersonal psychotherapy or other insight-oriented therapies exploring unconscious mental processes), Gestalt therapy, systemic therapy and family therapy. We also excluded studies that used behavioural-type therapies with no exposure component (e.g. behavioural activation, social skills training) and biofeedback as their experimental intervention.

4. Comparator intervention.

   • For pharmacological interventions: drug placebo or any other antidepressant with some serotonergic properties, or any other psychotherapy from those eligible (BT, CT, CBT), or other inactive type of therapy considered as ‘control’.

   • For psychological interventions: any type of psychological placebo (including attention placebo) or non-specific therapy (including supportive therapy), or waitlist/no treatment, or any other BT/CT/CBT type of therapy, or drug placebo, or any other antidepressant with some serotonergic properties.

5. Focus of analysis: Between-group comparison of treatments should be reported. Studies that report only additional secondary analyses (e.g. predictors of treatment outcome) were excluded. Studies that did not report continuous outcome were not included in the quantitative synthesis.

Data extraction

We used a standardised data extraction form to extract detailed information on included studies. This form also included a section on inclusion and exclusion criteria that was used for all studies that passed through the abstract screening. We originally developed this form in a Microsoft Word^® 2013 (Microsoft Corporation, Redmond, WA, USA) document format, but to facilitate data extraction, we transferred the various sections of this form into several Excel spreadsheets. All information extracted was directly recorded onto these spreadsheets on a computer.

Initial data extraction was carried out independently by two reviewers (HB and PS). As the agreement between the reviewers was high, one reviewer independently extracted the remaining papers (HB) and a second reviewer (PS) validated the extraction. Potential discrepancies were discussed by the reviewers and, if necessary, by all other collaborators during the meetings.

Data extracted from papers included the following information:

1. inclusion and exclusion criteria (study design, experimental intervention, control intervention, age range, primary diagnosis, diagnostic criteria, focus of analysis)

2. details of participants (country, treatment setting, age, diagnostic classification, primary severity scale used, comorbidities)

3. details of experimental and control interventions

4. details of continuous outcome (primary scale used, end of treatment follow-up time)

5. risk-of-bias assessment

6. results [baseline, end of treatment continuous measures for YBOCS or other primary scale, change from baseline, mean difference (MD) between arms, completers analysis or use of methods for handling missing data such as last observation carried forward]

7. dropouts (total dropouts per arm).

Risk-of-bias assessment: quality assessment strategy

To assess the methodological quality of included trials we used the criteria for quality assessment recommended by the Cochrane Handbook for Systematic Reviews of Interventions. 142 Two reviewers independently assessed and a second reviewer validated these criteria, which mainly focus on descriptions of sequence generation, allocation concealment, blinding, completeness of outcome data and selective outcome reporting, and other potential sources of bias (such as attrition rates). Studies were given a quality rating of ‘low’, ‘unclear’ or ‘high’ risk of bias in accordance with these criteria. If there was disagreement on quality assessment, the final rating was made by consensus with the involvement (if necessary) of another member of the review group. Studies with a high risk of bias were included in the main analysis, but we also examined in a sensitivity analysis the effect of excluding them.

Primary outcome

Pairwise analysis and NMA were conducted for the primary outcomes of reduction in OCD symptoms (as measured by the YBOCS in adults or the CYBOCS in children and adolescents) and for acceptability (as measured by total number of dropouts per study). A NMA is the simultaneous comparison of multiple competing treatments in a single statistical model. 143,144 In exploiting both direct and indirect evidence, a NMA produces estimates of the relative effects of each treatment compared with all others in the network, even if treatments have not been directly compared. It is then possible to calculate the probability of a treatment being better, or worse, for a specific outcome.

Derivation of primary outcome and handling of missing data

For the primary outcome, data are continuous and were reported as either (1) mean scores at baseline and at follow-up for treatment and control groups; or (2) mean change from baseline scores in each group. If both formats were reported, we chose the mean change from baseline score as our preferred summary, which captures correlations in measures within individuals. 145 If mean change from baseline was not reported, we used mean score at follow-up for each group, as this gives an unbiased estimate of treatment effect if randomisation is adequate. 145 If data were missing on either the total number of patients randomised or mean YBOCS/CYBOCS scores, the study was excluded from the quantitative analysis. Where possible, missing standard deviations (SDs) were derived from reported statistics following guidance in the Cochrane Handbook for Systematic Reviews of Interventions. 142 If derivation of missing SDs was not possible, they were estimated based on a prediction from a hierarchical model for SDs in those studies that did report them. Here, we assumed that any missing SD is exchangeable (i.e. broadly similar) with the reported SDs. (Further detail and code is available in Appendix 8.) For the secondary outcome, data are dichotomous and were extracted on intention-to-treat principles. Any participant dropout that occurred after the point of randomisation was included in our analysis.

Assessment of transitivity

The assumption of transitivity is the crucial starting point for a NMA. 146 Transitivity suggests that intervention A is similar when it appears in A versus B and A versus C studies. 147 It can be examined by comparing the distribution of potential effect modifiers across the different comparisons148 because if there is an imbalance in the presence of effect modifiers across the A versus B and A versus C comparisons, the conclusions about B versus C may be in doubt.

Study-level characteristics that were considered potential effect modifiers were mean baseline symptom severity, gender, participant age, length of trial follow-up, proportion of participants with concurrent mental illness (depression) and year of trial publication. We examined the study characteristics tables (see Tables 10 and 11) and concluded that, with the exception of participant age, the assumption of transitivity was likely to hold across the trials and comparisons. We considered that the assumption may be breached on the basis of participant age, because it would appear that children and adolescents were more likely than adults to be randomised to a psychological therapy. Indeed, of pharmacological treatments, only sertraline and fluvoxamine are licensed for use in patients aged < 18 years in the UK, and NICE also mentions fluoxetine in cases with significant comorbid depression. 118 Therefore, we decided that, contrary to our protocol specification, we would analyse children and adolescents separately to adults.

Pairwise and network meta-analysis

Network diagrams were drawn using Stata version 13 (StataCorp LP, College Station, TX, USA) to ensure that treatments formed a connected network for both outcomes and populations considered. 149 Pharmacological placebo was considered as the reference treatment throughout all analyses.

All analyses were conducted in a Bayesian framework and were undertaken using OpenBUGS version 3.2.3 (members of OpenBUGS Project Management Group; see www.openbugs.net). Pairwise meta-analyses were conducted in a single model assuming independent treatment effects and a shared heterogeneity parameter. 150 We used the NMA programme code given by Dias et al. 145 and modified to incorporate an additional class hierarchy,151 such that interventions with a similar mechanism of action were grouped together in a class in which pooled effects might be assumed to be ‘similar’. This approach allows both the relative effectiveness of the individual treatments and that of the classes to be estimated. On the basis of the systematic review and clinical expertise, it was assumed that only the SSRIs citalopram, escitalopram, fluoxetine, fluvoxamine, paroxetine and sertraline could be considered a ‘class’ on this criterion. Psychological therapies were analysed as individual treatments. Non-specific psychological therapies (‘psychological placebo’) were distinguished from waitlist controls. In psychological therapies we did not distinguish between individual or group delivery format. In addition, we did not distinguish treatments based on drug dose or intensity of psychological treatment. (All OpenBUGS code is available in Appendix 8. )

Random-effects models were used, accounting for the correlation between trial-specific effects in multiarm studies. 145 Vague priors were used for all parameters, including the prior for within-class variability. Convergence was checked based on two chains using the Brooks–Gelman–Rubin diagnostic plots and visually using history plots available in OpenBUGS. In all cases, the first 50,000 iterations were discarded as ‘burn-ins’. Reported estimates are based in the subsequent 100,000 iterations. We report the relative effectiveness of each treatment versus every other treatment and also the probability that each treatment is the most effective on each outcome.

Model fit and assessment of statistical inconsistency

Heterogeneity was assessed by examining the posterior median of the between-studies heterogeneity parameter from the random-effects model. Goodness of fit was measured by the posterior mean of the residual deviance. In a well-fitting model, the residual deviance should be close to the number of data points. Model comparison was based on the deviance information criterion (DIC). 152 A difference of 3 or more points was considered meaningful. 145 A key assumption of NMA is that of consistency between the direct and indirect evidence. To assess inconsistency, we compared the fit of a model assuming consistency with that of a model assuming independent treatment effects. 153 In addition, we also compared the results of the pairwise meta-analysis with the NMA. As a further proxy measure, where the NMA effect estimate did not fall within the 95% credible intervals (CrIs) from the pairwise analysis, we defined these as inconsistent.

Sensitivity analysis and meta-regression

Sensitivity analyses were conducted, excluding studies at high risk of bias on the following domains as defined by the Cochrane Collaboration’s risk-of-bias assessment tool:142

1. allocation concealment

2. outcome assessor blinding

3. incomplete outcome data

4. studies with high levels of attrition (overall attrition > 25% or differential attrition > 15%).

Separate meta-regressions were also conducted assuming a common interaction term for the following study-level characteristics:

1. length of trial (including follow-up)

2. year of publication of trial

3. pharmaceutical sponsorship of drug trials.

Studies excluded

Sixty-eight papers were excluded from the analysis. A detailed table of the excluded studies and the reasons for exclusion can be found in Appendix 2. Table 2 shows the main reasons for exclusion.

General summary characteristics of the included studies

We included 86 unique studies reported in 85 publications. It can be seen from Table 3 that 64 studies (74%) were conducted in adult patients, whereas 22 studies (26%) were conducted in child and adolescent samples. The majority of the studies (84%) had only two arms. Approximately half of the studies were conducted after the 2000s and 14% were conducted before the 1990s.

In total, 8611 patients were randomised (7306 adults and 1305 children and adolescents) into 194 arms including 23 different interventions or combinations of interventions. Table 4 summarises data on the number of randomised patients in arms/studies.

Country of publication

Figure 2 presents summary data on the country of publication of included studies. The majority of the studies (52% in total; 48% in adults vs. 66% in children/adolescents) were conducted in North America (33 studies in the USA, eight in Canada and four in both for the total sample). Five studies (all in adults) were multinational, that is, defined as having recruited patients from three or more countries (one fluoxetine vs. placebo; one citalopram vs. placebo; one paroxetine vs. placebo and clomipramine; one escitalopram vs. placebo and paroxetine; and one fluvoxamine vs. clomipramine study). Countries with more than three studies were the Netherlands (five studies), and Australia, Brazil and UK (three studies each). A total of seven studies (six in adults and one in children/adolescents) were conducted in Asia (three in Japan, three in Iran and one in China). There were no studies from Africa.

FIGURE 2.

Country of publication [n (%) of included studies].

Types of interventions

Fifty-six of the included arms (29%) involved a supposedly inactive intervention, either drug placebo (18.5%) or psychological placebo (4.5%), or a waitlist control (6%). Table 5 shows the number of arms/number of patients per type of intervention in the total sample and Figure 3 presents a relevant bar diagram.

FIGURE 3.

Number of arms per type of intervention: total sample. SRI, serotonin reuptake inhibitor.

It can be seen that in adults (Table 6), the most used active intervention was clomipramine (12%), followed by fluvoxamine (10%). Paroxetine, however, had the second largest sample of randomised patients after clomipramine. Overall, 48% of the arms in the adult set involved an active drug intervention, 22% involved an active psychological intervention and 4% involved a combination treatment. In children and adolescents (Table 7), the most used active intervention was CBT (18% of the arms), followed by sertraline (11.5%, either alone or in combination with CBT) and fluoxetine (9% of the arms). Approximately 30% of the arms in children and adolescent samples included a medication, 25% of the arms included a psychological intervention and 7% included a combination of both types of treatments.

Specific characteristics of individual studies

Tables 8 and 9 present specific characteristics of individual studies for the adult (n = 64 studies) and child and adolescent (n = 22 studies) subsets of data, respectively. The following data are presented for each individual study: study ID (including year of publication), total sample size (original number of randomised patients), number of arms, type of included interventions (grouped into three categories: medication arms only, psychological therapy arms only and arms with a combination of treatment interventions), specific intervention used in each arm, duration of the trial in weeks (primary end point), mean age of the total sample of randomised patients, percentage of female patients, primary scale used for the assessment of obsessive–compulsive symptoms, percentage of patients with depression comorbidity (grouped into six categories: none, < 25%, 25–50%, > 50%, unspecified and unclear), sponsorship of the study from drug companies (grouped into three categories: yes, no and unclear. This is not applicable for studies with psychological arms only. It should be noted that specific details of the interventions (mean dose, range of dose, number of psychotherapeutic sessions and mean duration of each session) are given for all studies in Appendix 6.

For the adult subset, the median number of randomised patients per study was 66 (range 16–466); 60% of the studies included drug arms only and 12% included combined arms; median duration of follow-up per study was 12 weeks (range 3–24 weeks); median percentage of female patients per study was 52.5% (range 0–94%); 85% of the studies used the YBOCS as their primary symptom scale, 54% of the studies excluded patients with major depression, and 60% of the studies that used at least one drug or combined arm were sponsored by the pharmaceutical industry. Of the 12 psychotherapy studies that used an inactive control condition, seven (58%) used a waitlist control (five CBT trials, one BT and one CT trial).

For the children and adolescents subset, the median number of randomised patients per study was 42 (range 9–207); 43% of the studies included drug arms only and 24% included combined arms; median duration per study was 12 weeks (range 5–43 weeks); median percentage of female patients per study was 42% (range 30–63%); 95% of the studies used the CYBOCS as their primary symptom scale; 24% of the studies excluded patients with major depression; and 50% of the studies that used at least one drug or combined arm were sponsored by the pharmaceutical industry. Of the seven psychotherapy studies that used an inactive control condition, five (71%) used a waitlist control (four CBT trials and one BT trial).

Individual studies per included active intervention

For ease of reference, we also present separate tables with the included studies for each active intervention (Tables 10 and 11 for adults and children and adolescent subsets, respectively). We present the following information: study ID (including year of publication); total sample size (original number of randomised patients); specific intervention used in each arm; duration of the trial in weeks (primary end point); and primary scale used for the assessment of obsessive–compulsive symptoms. It should be noted that some studies appear more than once because they compared an active drug with another active drug.

Description of the data set

Table 16 presents the raw data used for the YBOCS analysis in the adult subset of the data (a complete copy of the full data extraction is available in Appendix 4). Of the 64 studies eligible for inclusion in the NMA, 9 were excluded because they had not used the YBOCS scale. 156,165,168,169,186,194,205,208,210 This decision was made in light of the well-documented methodological and interpretational difficulties associated with the standardised MD. 142 The excluded studies are summarised here for completeness: these studies involved a total of 288 randomised patients (4% of the total randomised patients in the adult subset) and four clomipramine arms, three BT arms, three CT arms, two fluvoxamine arms, one amitriptyline arm, one imipramine arm and three placebo arms. One additional study was excluded because it was not part of the connected network. 179 Therefore, 54 studies were included in this analysis (see Table 16).

Table 17 presents summary data per type of intervention for the studies included in the NMA (number of arms and number of randomised patients per intervention).

Network meta-analysis: results

Network geometry

Figure 6 shows the network geometry of the YBOCS outcome in the adult subset, and Table 18 presents summary data per type of intervention (number of patients randomised, total number of links with other treatments, number of unique treatments compared). Overall, of the 136 pairwise comparisons that can be made among the 17 treatment conditions, only 37 (27%) were studied directly by head-to-head comparison in 54 studies involving 6652 randomised patients. It should be noted, however, that 24 of these 37 direct comparisons are made in single trials. Each circle (node) represents an intervention and is proportional to the number of participants randomised to each treatment (i.e. the larger the node, the greater the number of participants randomised to each intervention). Placebo (n = 1515), paroxetine (n = 902), clomipramine (n = 831), fluoxetine (n = 633), sertraline (n = 565), fluvoxamine (n = 521), citalopram (n = 311), BT (n = 287), CBT (n = 231), escitalopram (n = 226), psychological placebo (n = 196), CT (n = 172), venlafaxine (n = 98) and waitlist (n = 97) had a sample approximately ≥ 100 (Table 18). Lines represent the available direct evidence and are proportional to the number of trials making a randomised comparison of each pair of treatments. Figure 6 includes 79 randomised pairwise comparisons and the most common comparisons are those between placebo versus paroxetine (n = 7), placebo versus fluvoxamine (n = 6), placebo versus fluoxetine, placebo versus clomipramine, CBT versus waitlist and BT versus psychological placebo (n = 5 each), placebo versus sertraline and fluvoxamine versus clomipramine (n = 4 each). Nodes with the most connections (links) in the network (see Table 24) are drug placebo (n = 35 links with 10 different interventions), clomipramine (n = 17 links with eight different interventions), fluvoxamine (n = 16 links with seven different interventions), paroxetine (n = 15 links with six different interventions), BT (n = 14 links with seven different interventions), CBT (n = 9 links with six different interventions), fluoxetine (n = 8 links with four different interventions), sertraline (n = 7 links with four different interventions), psychological placebo (n = 6 links with four different interventions), waitlist (n = 6 links with two different interventions), citalopram (n = 5 links with two different interventions) and CT (n = 5 links with two different interventions).

FIGURE 6.

Network diagram for YBOCS analysis representing individual treatments (adult subset). BTCLO, BT + clomipramine; CBTFLV, CBT + fluvoxamine; CIT, citalopram; CLO, clomipramine; ESCIT, escitalopram; FLV, fluvoxamine; FLX, fluoxetine; HYP, hypericum; PAR, paroxetine; PL, placebo; PSYPL, psychological placebo; SER, sertraline; VEN, venlafaxine; WL, waitlist. Reproduced from Skapinakis P, Caldwell DM, Hollingworth W, Bryden P, Fineberg N, Salkovskis P, et al. Pharmacological and psychotherapeutic interventions for management of obsessive-compulsive disorder in adults: a systematic review and network meta-analysis [published online ahead of print June 15 2016]. Lancet Psychiatry 2016. © Skapinakis et al. Open Access article distributed under the terms of CC BY. 214

TABLE 18 - Summary data per type of intervention for the YBOCS analysis, sorted by number of randomised patients (adult subset)

Intervention	Number of patients randomised	Number of pairwise comparisons (links)	Number of unique treatment comparisons
Placebo	1515	35	10
Clomipramine	831	17	8
Fluvoxamine	521	16	7
Paroxetine	902	15	6
BT	287	14	7
CBT	231	9	6
Fluoxetine	633	8	4
Sertraline	565	7	4
Psychological placebo	196	7	4
Waitlist	97	6	2
Citalopram	311	5	3
CT	172	5	2
Escitalopram	226	4	2
Venlafaxine	98	2	2
CBT + fluvoxamine	6	2	2
BT + clomipramine	31	2	3
Hypericum	30	1	1

Description of the data set

Table 24 presents the raw data used for the dropout analysis in the adult subset of the data. From the 64 studies eligible for inclusion in the NMA, 11 were excluded: eight studies either did not report dropout data or did not report dropout data separately for each arm;179,182,187,188,194,201,204,208 and three studies were excluded because there were no dropouts (zero dropouts) in all arms. 183,192,199 Therefore, 53 studies were included in this analysis. 124,126,127,154–178,180,181,184–186,189–191,193,195–198,200,202,203,205–207,209–213