33.2 HARNESSING THE POWER OF BAYESIAN HIERARCHICAL MODELING: IMPLICATIONS FOR CLINICIANS

Abstract

This presentation aims to provide an overview of Bayesian hierarchical modeling (BHM) for clinicians and to demonstrate the application of this multilevel modeling approach to understanding variation in antidepressant response across large trials in children and adolescents. After a brief introduction to BHMs in clinically relevant terms, this presentation will examine how BHMs capture heterogeneity across groups of patients. The BHM will be compared to the standard mixed model with repeated measures (MMRM) approach. The presenter will then examine how BHMs can explicitly model the correlation in symptoms over time in an individual patient enrolled in a clinical trial. Finally, an application to TORDIA will be reviewed, in which the individual trajectory of treatment response was modeled using an individual logarithmic trend “random effects” coefficients BHM. From the model estimates, Bayesian 2-tailed p values were computed to evaluate the null hypothesis: no difference was found in efficacy between SSRIs and venlafaxine. Using BHMs in the TORDIA study (n = 334), we demonstrate that for patients who did not receive CBT (n = 168), SSRIs produced greater and faster improvement in depressive symptoms compared to venlafaxine (p = 0.015). For patients with anxiety symptoms in addition to MDD, no differences in response or trajectory of response were detected between SSRIs and venlafaxine (p = 0.168). For patients receiving CBT (n = 162), SSRIs and venlafaxine produced similar improvements in symptoms of anxiety (p = 0.491) and depression (p =0.077). The traditional MMRM is static, relies on large sample assumptions, and is more restrictive in modeling individual patient variation. The BHM addresses these shortcomings and can be applied to existing and new randomized controlled trials in child and adolescent psychiatry. Applying BHMs in TORDIA revealed that SSRIs produce greater and faster improvement in depressive symptoms compared to venlafaxine (p = 0.015). With this approach, clinicians can move beyond “average response” and consider trajectory of response based on individual patient characteristics. This can inform recommendations for treatment trial duration and help understand which patients benefit from what treatments.