A machine learning personalized treatment rule to optimize assignment to psychotherapies for grief among veterans

Abstract

BackgroundComplex grief patterns are associated with significant suffering, functional impairments, health and mental health problems, and increased healthcare use. This burden may be even more pronounced among veterans. Behavioral Activation and Therapeutic Exposure (BATE-G) and Cognitive Therapy for Grief (CT-G) are two evidence-based interventions for grief. The goal of this study was to use a precision medicine approach to develop a personalized treatment rule to optimize assignment among these psychotherapies. MethodsWe analyzed data (N = 155) from a randomized clinical trial comparing BATE-G and CT-G. Outcome weighted learning was used to estimate an optimal personalized treatment rule. Baseline characteristics including demographics, social support, variables related to the death, and psychopathology dimensions were used as prescriptive factors of treatment assignment. ResultsThe estimated rule assigned 72 veterans to CT-G and 56 to BATE-G. Assigning participants according to this rule was estimated to lead to markedly lower mean grief level following 6 months from treatment compared to assigning everyone to either BATE-G (Vdopt – VBATE-G = −18.57 [95 % CI: (−29.41, −7.72)]) or CT-G (Vdopt – VBATE-G = −20.89 [95 % CI: (−30.7, −11.07)]) regardless of their characteristics. LimitationsParticipants were primarily male veterans, and identified with Black or White race. The estimated rule was not externally validated. ConclusionThe estimated rule used relatively simple, easily accessible, client characteristics to personalize assignment to treatment using a precision medicine approach based on machine learning and causal inference. Upon further validation, such a rule can be easily implemented in clinical practice to prescriptively maximize treatment benefits.