Estimating Effect Variability Using Targeted Superlearning🔗

Article 6 is entitled “Psychological Intervention in Patients with Subthreshold Depression: Individual Participant Data Meta-Analysis of Treatment Outcomes and Effect Heterogeneity”. It presents an IPD-MA in which we synthesize the benefits of psychological intervention in patients with subthreshold depression (sD) up to two years. Analyzed outcomes included effects on symptom burden, remission, response and reliable symptom deterioration.

In a second step, targeted superlearning was used to estimate the distribution of true treatment effects across all patients. Based on these distributions, the variability of individual benefits was estimated, as well as the proportion of patients who do not fare better under treatment than under control. Lastly, univariate moderator analyses were conducted to determine relevant predictors of differential treatment effects.

Our findings strongly support the provision of psychological intervention in sD, at least when patients already present with moderate symptoms. Nevertheless, results of superlearner analyses indicate that benefits differ substantially across individuals, with up to 40% predicted to show no better outcomes when receiving an intervention.

The study combines analyses prespecified in two separate statistical analysis plans (SAPs); one focusing on the IPD-MA of treatment outcomes and effect modifiers (SAP-1; osf.io/vba7f), the other on targeted superlearning (SAP-2; osf.io/28xnc). A modified version of the paper was published in the “British Journal of Psychiatry”.