Family and Couple Integrated Cognitive-Behavioural Therapy for Adults with OCD: A Meta-Analysis

A systematic review of the literature was conducted. PsychInfo, PubMed, and Scopus were searched. Search terms included: [“obsessive-compulsive disorder” OR “obsessive compulsive disorder” OR “OCD”] AND [family OR couple^* OR marital OR marriage OR “romantic relationship”] AND [“cognitive-behavio^*” OR “cognitive be-havio^*” OR cognitive OR behavio^*] AND [treatment OR intervention OR counsel^* OR therap^*]. The initial search produced 1456 papers. From this, 465 duplicates were removed, and 3 were identified as being websites, leaving 988 papers for title and abstract review. The papers were reviewed by two authors and disagreements were discussed (n = 13, or 1%). Papers were excluded if they: a) did not contain appropriate data for a meta-analysis (e.g., review or theoretical papers, descriptive outcomes, case-studies, n = 256), b) were not CBT treatment studies with family integration (e.g., spouse, child, sibling, parent; n = 222), c) did not include an OCD population (n = 175), or d) sampled an exclusively pediatric or adolescent population (n = 310). In total, 963 papers were excluded at this stage, leaving 25 papers for full-text review. During full-text review, 12 papers were excluded. Reasons for exclusion were not having appropriate data for meta-analysis (small sample case-studies, qualitive outcomes, n = 5), not being CBT treatment studies with family integration (n = 3), being a duplicate (thesis and publications with different titles, n = 2), authors being unable to provide necessary data when contacted (n = 1), not including an OCD population (n = 1), and not being written in English (n = 1). Previous reviews and references of relevant papers were also examined for studies that might have been missed in the search. Two additional papers were identified using this method, resulting in a total of 15 included papers. One of these papers included two independent samples (Himle et al., 2001), thus there were 16 samples included in the present analyses. A Preferred Reporting Items for Systematic Reviews and MetaAnalyses (PRISMA; Moher et al., 2009) flow diagram is included as Fig. 1. For study characteristics, see Appendix A and for meta-analysis reference list, see Appendix B.

Outcome measures were classified by two raters as either assessing: OCD symptoms, depression, anxiety, functional impairment, general relationship improvement, antagonism, accommodation, or family member mental health. See Appendix C for a list of all measure categorizations. Of note, two pairs of publications (Abramowitz et al., 2013 and (Belus et al., 2014; Remerswaal et al., 2017 and Remerswaal et al., 2019) utilized the same samples but included different outcome measures: patient-focused outcomes in the former of the pairs and family-focused outcomes in the latter of the pairs; therefore, there was no overlap in outcome measurement between the studies. For general relationship satisfaction, measures were not included if they exclusively pertained to romantic or sexual satisfaction.

Moderators were classified as follows: 1) time point (dichotomous: post-treatment/follow-up), 2) length of follow-up (continuous), 3) study design (dichotomous: within-subjects/pre-post controlled), 4) rater (dichotomous: clinician-rated/self-report, and when applicable, patient-rated/partner-rated), 5) explicit involvement of partner in ERP (dichotomous: yes/no), 6) explicit focus on reducing antagonism (dichotomous: yes/no), 7) explicit focus on reducing accommodation (dichotomous: yes/no), 8) format (dichotomous: individual/group), 9) sample analyzed (dichotomous: completer/intent-to-treat), 10) number of patient sessions (continuous), 11) dosage of family involvement (continuous: number of family sessions/number of total sessions), 12) patient gender (continuous: percentage female), and 13) year of publication (continuous).

Not all outcomes were classifiable by all moderator categories; in some instances there was not enough data to run the analyses (e.g., there were not enough clinician-rated anxiety measures to examine clinician/self-report as a moderator), or a moderator didn't make sense for a specific outcome (e.g., OCD symptoms were only rated by clinician or self-report, not by the partner, so patient/partner was not included as a moderator of OCD symptoms). Given that not all studies explicitly stated whether accommodation and antagonism were explicitly targeted, or whether family members were involved in ERP, two authors independently examined the full-text articles to classify whether they were explicitly mentioned as part of the protocol. The two raters agreed upon all classifications. In instances where it was not explicitly mentioned that the treatment targeted accommodation or antagonism, or integrated family into ERP, the authors were emailed to confirm. If the authors did not respond, then these papers were treated as not having explicitly targeted the variable of interest. Lastly, risk of bias was examined using the Cochrane RoB 2 (Sterne et al., 2019). Two raters independently assessed risk of bias. Three studies were not agreed upon following initial independent rating and were resolved through discussion.

Effect sizes were extracted from studies depending on the type of data given in the study, then converted to estimates of Hedge's g, which is a bias-corrected version of a standardized mean difference (Cohen's d). The formula for this calculation is listed below, where d is the original Cohen's d, df is the degrees of freedom, and V_d is the variance in the original Cohen's d. $g=(1-3/4(df)-1{)}^{*}d$

To facilitate these effect size extractions, suggestions from Feingold (2015), Lee (2005), Morris (2008), and Westfall (2016) were implemented. Extracted data were analyzed using the metafor R package (Viechtbauer, 2010). Because studies often have multiple measures that are intended to measure similar (or identical) outcomes, multiple effect sizes can be extracted, forcing the researcher to choose a method of dealing with their non-independence. The present strategy was to conduct a three-level meta-analysis, using strategies described by Assink and Wibbelink (2016). Though meta-analyses are inherently multi-level in nature, the conventional meta-analysis usually only operates on two levels: variance due to sampling error (Level 1) and variance due to between-study heterogeneity (Level 2). Three-level meta-analyses allow for multiple outcomes within each study, therefore allowing examination of sampling (Level 1), within-study (Level 2), and between-study (Level 3) heterogeneity of variance. For each level of analysis, the estimated proportion of variance that could be accounted was computed, indicating the relative heterogeneity of each. Likelihood ratio tests tested whether within- or between-study variance was nonzero in each model.

Results were broadly categorized into two sets of analyses: patient outcomes (symptom severity, depression, anxiety, and functional impairment) and family and relational outcomes (relational improvements, antagonism, accommodation, and family mental health). After reporting overall effect size and measuring heterogeneity, each subgroup was tested for differences between post-treatment and follow-up, to determine whether improvements were maintained over time. Moderators were examined for each outcome type when discrete moderators had at least three effect sizes across two studies per group and when continuous moderators had at least three measurement points.

Publication bias was measured in three ways. First, a trim-fill analysis was conducted on the two-level meta-analyses across outcome types. Because there is no current consensus on how to apply trim-fill to three-level meta-analyses, the funnel plots and estimates were inaccurate. Yet, despite the likely clustering of effect sizes due to within- study homogeneity, the distribution of effect sizes in a funnel plot still indicated whether other unreported clusters (rather than individual effect sizes) might be missing. Second, and similarly, a rank test was conducted to test funnel plot asymmetry. Third, publication year was used as a continuous moderator in the model, testing if newer studies were more likely to show higher/lower effect sizes than older studies.

Of the 16 samples (15 publications, with Himle et al., 2001 having two independent samples), ten included follow-up data (63%). Nine were within-subjects designs (56%) while seven included a control group (44%). Eight samples had partners take part in ERP (50%). Eight samples targeted antagonism explicitly (50%). Twelve samples targeted accommodation explicitly (75%). Eleven studies utilized an “individual” therapy format (i.e., dyads were treated alone, 69%), with 5 utilizing a “group” format (i.e., dyads were treated in groups, 31%). Ten samples used completer data (63%), four used intent-to-treat (25%), and two presented both (13%). Patient sessions ranged from 5 sessions to 24 sessions, with the average number being 11.88. Family involvement in sessions ranged from as little as 1 session to attending one more session than patients, with the average dosage being family attending 70% of sessions. Of the studies that reported gender, on average, 61.83% of the patients were female.

FIT was associated with improvements across all patient outcomes (Table 1): OCD symptom severity (κ = 41, g = −1.20, SE = 0.21, z = −5.68, p < .001), depression (κ = 25, g = −0.69, SE = 0.09, z = −7.7, p < .001), anxiety (κ = 11, g = −1.56, SE = 0.41, z = −3.60, p < .001), and functional impairment (κ= 14, g = −0.48, SE = 0.18, z = −2.66, p = .008). Patterns of effect sizes did not differ between post and follow-up for any patient outcome, regardless of if the comparison was discrete or continuous. There was significant between-study heterogeneity for symptom severity ($I_{\text{betweeen}}^2$ = 80.27%, LRT = 33.22, p < .001) and anxiety ($I_{\text{betweeen}}^2$ = 89.18%, LRT = 11.03, p < .001), whereas significant within-study heterogeneity was found for functional impairment ($I_{\text{within}}^2$ = 67.95%, LRT = 8.21, p = .004). See Table 1 for all outcome data.

Because the YBOCS was the measure of OCD symptom severity that was most prevalent among the studies in the meta-analysis and is considered the “gold-standard” measure of OCD symptom severity (Frost et al., 1995), we analyzed its results both with other OCD measures (see above) and alone. Consistent with the previous analysis, YBOCS scores decreased (κ = 23, g = −1.21, SE = 0.21, p < .001), with no detectable differences in effect sizes between post-treatment and follow-up ($\beta$ = −0.03, SE = 0.15, z = −0.21, p = .829).

There was significant heterogeneity of variance between studies that used the YBOCS ($I_{\text{betweeen}}^2$ = 89.18%, LRT = 15.82, p < .001), suggesting that moderator analyses may be appropriate. There were greater improvements in YBOCS scores for within-subject studies (κ = 12, g = −1.60, SE = 0.24, p < .001) than there were in pre-post controlled studies (κ =11, g = −0.78, SE = 0.24, p = .001; $\beta$ = −0.83, SE = 0.34, z = −2.44, p = .015). No other moderators reached the threshold of statistical significance ($\alpha$ = 0.05).

Results for all moderator analyses are in Table 2; only statistically significant results (at $\alpha$ = 0.05) are reported in-text. Studies that used within-subject designs (κ = 22, g = −1.71, SE = 0.22, p < .001) showed greater reductions in patient symptom severity than those that used pre-post control designs (κ= 19, g = −0.68, SE = 0.22, p = .002; $\beta$ = −1.03, SE = 0.31, z = − 3.34,p < .001). Patients reported greater reductions in depression when interventions explicitly addressed family accommodation of OCD behavior (κ = 16, g = −0.85, SE = 0.09, p < .001) than when they did not (κ = 9, g = −0.50, SE = 0.08, p < .001; $\beta$ = −0.35, SE = 0.12, z = −2.80, p = .005). FITs formated to treat dyads individually also resulted in greater reductions in patient depression (κ = 15, g = −0.85, SE = 0.10, p < .001) than group therapies (κ = 12, g = −0.54, SE = 0.08, p < .001; difference: $\beta$ = −0.31, SE = 0.13, z = −2.37, p = .018). Despite there being significant heterogeneity between studies for anxiety and within studies for functional impairment, none of the investigated moderators were significant.

All family and relational outcomes showed improvements: general improvement in relationships (k = 28, g = 0.35, SE = 0.074, p < .001), reductions in antagonism, indexed by patient and family reported reductions in hostility and criticism (k = 21, g = −0.42, SE = 0.14, p = .002), less familial accommodation of OCD behaviours (k = 11, g = −0.83, SE = 0.22, p < .001), and improved mental health of family members (k = 14, g = 0.54, SE = 0.16, p < .001). Patterns of effect sizes did not differ between post and follow-up for any patient outcome, regardless of whether the comparison was discrete or continuous. There was a small, but significant amount of heterogeneity in between-study effects for antagonism ($I_{\text{betweeen}}^2$ = 29.10%, LRT = 4.05, p = .044) but not across effects (Q(20) = 23.59, p = .261). On the other hand, for accommodation, there were significant sources of heterogeneity across levels of effects (Q(10) = 29.57, p = .001) but not at the within-study ($I_{\text{within}}^2$ = 1.12%, LRT = 0.004, p = .951) or between-study ($I_{\text{betweeen}}^2$ = 66.50%, LRT = 0.66, p = .417) levels individually.

Patients reported less relational improvements than their families (patient: κ = 15, g = 0.24, SE = 0.09, p = .008; family: κ = 13, g = 0.51, SE = 0.10, p < .001; difference: $\beta$ = −0.27, SE = 0.13, z = −2.05, p = .040). Surprisingly, the more sessions patients attended, the reduction in antagonism, as measured through family and patient self-report, was muted by g= 0.04 (SE = 0.01, z = 2.76, p = .006). Put another way, the improvement in antagonism was steeper for shorter patient treatment protocols. The more families were involved in patient therapy, the less improvement was seen in family members’ mental health ($\beta$ = −0.76, SE = 0.23, z = −3.28, p = .001). Specifically, there was a blunting in the improvement over time. Also, for each 1% increase in the ratio of men-to-women in the studies, the reduction in perceived antagonism was lessened by g= 0.008 (SE = 0.003, z = 3.12, p = .002). In other words, the reduction in antagonism was smaller when there were more men in the sample. No other moderators showed significant effects (see Table 2).

For publication bias, an investigation of funnel plot asymmetry (via rank test or trim-fill analysis) indicated that all outcome classes, except functional impairment, were likely biased (Table 2; see Appendix D for plots). Interestingly however, this bias was not consistent in its direction. Though the trim-fill analysis suggested that effects for symptom severity, depression, and anxiety were overstated, it also suggested that the effects for the YBOCS subgroup analysis and all family and relational outcomes were understated. Publication year was not a significant moderator across all outcome types (Table 2).

The uncontrolled studies all had high risk of bias. Of the controlled studies (seven), three were identified as having some concern of bias, and four were identified as having high risk of bias. Problematic areas were differences in drop-out between conditions, lack of reporting the reason for drop-out, the potential that knowledge of condition could influence outcome data, not-having pre-registered the trial, and it being unclear whether all outcome data was reported. Randomized controlled studies coded as having high risk of bias were compared to those coded as having some concerns of bias across OCD symptom severity, the YBOCS, and depression, as these were the only outcome types for which sufficient data were available. Only controlled studies were compared given that study design (within vs. between conditions) was already examined as moderator. Effect sizes for high risk studies did not differ from those that had some concern of bias, in terms of OCD symptom severity (B = 0.08, SE = 0.29, z = 0.28, p = 0.779) or the YBOCS (B = −0.09, SE = 0.47, z = 0.19, p = 0.848). However, contrary to expectations, studies coded as having some risk showed larger reductions in depression than those coded as high risk (B = − 0.57, SE = 0.27, z = –0.21, p = 0.034).

Five studies compared FIT to control conditions that included individual ERP, the gold-standard non-pharmacological treatment for OCD (see Table 3). Following FIT, OCD symptomology was reduced across OCD measures (k = 12, g = −0.77, SE = 0.15, z = –5.11, p < .001) and when exclusively examining the YBCOS (κ = 8, g = –0.70, SE = 0.14, z = –5.07, p < .001), compared to controls with ERP. FIT also resulted in greater reductions in depression (κ = 7, g = −0.60, SE = 0.17, z = –3.46, p < .001), functional impairments (κ = 12, g = –0.47, SE = 0.20, z = –2.23, p = .026), and accommodation (κ = 3, g = –0.53, SE = 0.27, z = –1.96, p = .050) than controls with ERP. However, there were no statistically significant reductions in anxiety compared to controls with ERP (κ = 4, g = –0.80, SE = 0.51, z = –2.23, p = .120). Other outcome variables did not have enough studies per outcome variable to be included, nor was there enough power to examine moderators in this sub-analysis. See Appendix E for forest and funnel plots.

All studies included individuals with a diagnosis of OCD or significant OCD symptoms. No study specified that participants had to have a sole diagnosis of OCD, suggesting that the treatments may be useful for individuals with comorbidity, which is likely to be the rule rather than the exception. This is a strength of the current body of literature. However, few studies assessed family member mental health; therefore, it is not clear whether FIT is effective for this population when family members themselves have significant mental health concerns.

This study aimed to investigate the effect of family integrated CBT specifically, given that it is the gold standard treatment for OCD (Hezel and Simpson, 2019). However, as a result, other FITs were excluded, and thus limitations of the present meta-analysis include its relatively narrow scope. There may be other FITs that did not use a cognitive-behavioural approach that were not included in the present analysis, despite the potential efficacy of other modalities. Another limitation of the meta-analysis was the categorization of explicit treatment targets for ERP, antagonism, and accommodation. Despite attempts made by the authors to confirm whether protocols included these factors, there was an element of subjectivity to this categorization. If authors did not respond regarding the inclusion of ERP, antagonism, and accommodation (n = 4, 4, and 3, respectively) and it was not explicit in their protocol, they were categorized as not including these treatment components. This was deemed justifiable as it would be unlikely that authors would mention variables they did not target in their manuscript. Another limitation was the relatively small sample size, which may explain the null findings of many moderator analyses. Finally, another limitation of the present meta-analysis was that the inclusion criteria were relatively lenient in order to maximize the number of studies and the power of our analysis. As more research is conducted in this area, future meta-analyses may want to consider restricting inclusion criteria to studies that only include patients diagnosed with validated diagnostic interviews, well-validated outcome measures, or studies with lower bias.

First, more research is needed that compares the efficacy of FIT to gold-standard OCD treatment (e.g., ERP, pharmacotherapy) on a variety of outcome measures. The present meta-analysis was only able to compare FIT to controls that included ERP on five outcome variables. Future studies should include relational variables such as antagonism and relationship satisfaction more broadly, as currently it is unknown whether improvements in these areas are due to family-integration specifically, or whether they are due to common-factors present in all treatments or other components of the treatment package. Of the outcomes that were compared to controls with individual ERP, it is also important to note that there was heterogeneity across the control conditions, as some controls included ERP and other strategies (e.g., pharmacotherapy), whereas others were highly matched, with the only difference being FIT (the latter design being more stringent). Further, FIT often included a focus on relational variables that may not be present in standard treatment. Thus, it remains unclear whether addressing relational variables, such as accommodation or satisfaction, in individual ERP would be adequate, or if the integration of family members is needed to enhance treatment outcomes. It is also important that trials are designed to minimize risk of bias. Many studies had a high risk of bias, and none were deemed as having a low risk of bias (although results were not impacted by higher degree of bias, as measured by the Cochrane guidelines). However, following Cochrane guidelines for clinical trials will be helpful for ensuring future studies are less biased.

Another area for future research should be to examine more predictors of treatment outcome. There was considerable heterogeneity between studies for anxiety and within studies for functional impairment despite no moderators being significant, suggesting that future research investigating potential moderators is needed to understand who is benefiting more from treatment. More information on moderating factors may also be useful for clinicians deciding on which patients should be targeted for FIT; this would require examination of both patient and family member characteristics.

Though evidence indicates that FIT leads to improvements in OCD symptoms, it is still unclear why change is occurring. Pinpointing these mediators may help to streamline FIT: increasing feasibility of including the family member in treatment while maximizing the benefits. One study included in the present review examined a potential mechanism of change in OCD symptoms. Thompson-Holland et al. (2015) found that change in accommodation predicted change in patient OCD symptoms whereas change in OCD symptoms did not predict change in accommodation. Given these findings along with those of the present meta-analysis, it is possible that FIT improves OCD symptoms via reductions in accommodation. Future research should examine accommodation as a mediator, in addition to other possible mechanisms of change, such as family burden and distress, family symptoms, and antagonism. In order to test these models, we suggest that researchers expand beyond exclusively taking measurements at pre and post and instead include measurement of outcomes and mechanisms at multiple time points during treatment.

With respect to protocols, the present analysis did not find that targeting accommodation or antagonism explicitly or involving family members in ERP moderated patient OCD symptom improvement; however, there were limitations in these analyses. Although most protocols were clear, some papers did not explicitly mention targeting these processes, which may have been because they were not targeted. In these cases, the authors were contacted to confirm; however, some did not reply and, as such, these papers were treated as not having targeted these processes. Thus, the present approach was cautious and although steps were taken to ensure the correct categorization, these findings should be understood in the context of this limitation. Further, only four studies in the present review included antagonism as an outcome variable, so future research may benefit from including measures of both accommodation and antagonism. In addition, the majority of these studies (n = 11) examined FIT in Western cultures. There may be significant differences in the effect of these treatments in other cultures, and more research in this area will be valuable. Some authors have argued that FIT may be especially important in cultures where close relationships with families are especially valued (e.g., Rosa-Alcázar and Inieta-Sepúlveda, 2018). Further, given we observed that men did not experience the same improvement in antagonism as women, it may be valuable to further explore gender differences in relationship improvement in treatment to refine treatment protocols.

In addition, future research should investigate the optimal dose of family involvement in treatment. Our findings suggest that family involvement ranging from participating in one session to all of the sessions is equally as effective for patient outcomes. Research investigating the minimal amount of family involvement needed and the necessary interventions to maximize outcomes would be helpful to inform clinical practice. It would also be beneficial to examine whether it is optimal to include the family member in the treatment sessions with the patient (e.g., both patient and family member attend together) or if the family member should be seen separately. Each format may offer specific benefits (allowing the patient/family to role-play vs. fostering an environment where the family may feel more able to express negative emotions such as frustration or burnout).

Lastly, we aim to highlight the potential benefit of integrating a family systems approach to treatment (e.g., Davidson et al., 2017), which challenges the “sick patient” and “healthy family member” dichotomy, into a CBT framework. Both patient and family members are involved in behaviours and cognitions that maintain and exacerbate the OCD. Thus, OCD may be conceptualized and treated as a disorder of the family, rather than a disorder within one person. It is promising that over the past few years there have been more publications integrating family into CBT for OCD (10 over the past decade), suggesting this may become an important area for further study. In the meantime, it seems remiss not to involve the family to some degree in treatment, if all parties are willing, given the preliminary support that FIT may be superior to individual ERP on some outcomes. It is possible that as the research in this area increases, the question may shift from “when should I include the family in treatment of OCD?” to “what is the optimal way to integrate the family into my patient's treatment?”