Moderators of response to child-based and parent-based child anxiety treatment: a machine learning-based analysis

Abstract

Background: Identifying moderators of response to treatment for childhood anxiety can inform clinical decision-making and improve overall treatment efficacy. We examined moderators of response to child-based cognitive-behavioral therapy (CBT) and parent-based SPACE (Supportive Parenting for Anxious Childhood Emotions) in a recent randomized clinical trial.

Methods: We applied a machine learning approach to identify moderators of treatment response to CBT versus SPACE, in a clinical trial of 124 children with primary anxiety disorders. We tested the clinical benefit of prescribing treatment based on the identified moderators by comparing outcomes for children randomly assigned to their optimal and nonoptimal treatment conditions. We further applied machine learning to explore relations between moderators and shed light on how they interact to predict outcomes. Potential moderators included demographic, socioemotional, parenting, and biological variables. We examined moderation separately for child-reported, parent-reported, and independent-evaluator-reported outcomes.

Results: Parent-reported outcomes were moderated by parent negativity and child oxytocin levels. Child-reported outcomes were moderated by baseline anxiety, parent negativity, and parent oxytocin levels. Independent-evaluator-reported outcomes were moderated by baseline anxiety. Children assigned to their optimal treatment condition had significantly greater reduction in anxiety symptoms, compared with children assigned to their nonoptimal treatment. Significant interactions emerged between the identified moderators.

Conclusions: Our findings represent an important step toward optimizing treatment selection and increasing treatment efficacy.

Keywords: Anxiety; behavior therapy; machine learning; parent training.