Identification of effective combinations of behavior change techniques using meta-CART

Abstract

Background: In health psychology meta-analyses usually multiple moderators, such as behavior change techniques (BCTs), are available. In these cases, the question arises which combinations of moderators influence treatment effectiveness. However, traditional meta-analysis methods often lack sufficient power to investigate interaction effects between moderators, especially high-order interactions. To solve this problem, meta-CART was proposed by integrating Classification and Regression Trees (CART) into meta-analysis. This method appeared to be successful in detecting combinations of BCTs that result in a higher average treatment outcome. Method: The meta-CART method was improved upon two aspects: 1) the stepwise approach was changed into one integrated approach; 2) the fixed- or random-effects assumption was taken into account in the interaction detection procedure. The performance of the improved meta-CART was investigated via an extensive simulation study on different types of moderator variables (i.e., dichotomous, ordinal, and multinomial variables). Results: The method can achieve satisfactory performance (power > 0.80 and Type I error < 0.05) if the number of studies is large enough. The required minimum number of studies ranges from 40 to 120 depending on the complexity and strength of the interaction effects, and the residual heterogeneity. Discussion: The improved version of meta-CART applies the fixed- or random-effects assumption consistently in both detection and test procedure. This method is able to identify effective combinations of behavior change techniques. Knowledge about such combinations is useful for evaluating existing treatments and designing new treatments.