Can registry-based auxiliary variables improve handling of missing outcome data in mental health trials? Evidence...

Missing outcome data are a pervasive challenge in randomized controlled trials (RCTs) of mental health interventions, potentially biasing effect estimates. Linking survey data with administrative registries offers new opportunities to include auxiliary variables in multiple imputation (MI), but evidence of their added value remains scarce.

We used data from the Norwegian RCT of Prompt Mental Health Care (PMHC; N = 681) which assessed self-report outcomes of depression (PHQ-9) and anxiety (GAD-7) at 6 and 12 months. Registry linkage provided information on prescriptions, consultations, sick leave, and benefits (N = 651). We examined whether registry variables were associated with missingness and assessed their correlations with depression and anxiety symptom levels. Further, we compared estimated treatment effects through a linear regression model with treatment group as a predictor across three strategies: (1) complete case analysis, (2) MI with trial data only, and (3) MI with all registry variables in addition to trial data.

Among registry indicators, baseline financial assistance was strongly associated with nonresponse at 12 months. Concurrent registry markers of mental-health consultations and prescriptions were modestly correlated with higher depression and anxiety symptoms at both 6 months and 12 months, whereas prescription use showed modest associations mainly with depressive symptoms (particularly at 12 months) and weaker or null associations with anxiety. The remaining registry variables were weakly associated with missingness and symptom levels, and none were substantially associated with both. Estimated treatment effects were consistent across missing data strategies. Adding registry auxiliaries in MI did not materially change standardized treatment effects of PMHC on depression or anxiety at either time point.

Registry variables showed theoretical and some empirical relevance, but most associations with outcomes and missingness were weak. Including them in MI models did not materially change treatment effect estimates, reinforcing the robustness of the trial's conclusions. These findings suggest that registry linkages may not always add value, highlighting the importance of prioritizing complete follow-up while identifying contexts where such data could be more impactful.

ClinicalTrials.gov NCT03238872. Registered on August 3, 2017 (retrospectively registered).