Which intervention works for whom: Identifying pre-treatment characteristics that predict who will benefit from a specific alcohol text message intervention from a randomized trial

Abstract

Introduction

Effective interventions show heterogeneity in treatment response. Addressing this heterogeneity involves identifying which intervention works best for whom. One method to address this heterogeneity identifies treatment-subgroup interactions to determine which of two interventions has greater effect for certain individuals based on their profile of pre-treatment characteristics. This secondary analysis of a randomized clinical trial (RCT) to address binge drinking examined whether two of the RCT's interventions, GOAL and COMBO, which produced similar reductions in drinking outcomes, might have involved treatment-subgroup interactions. Identifying treatment-subgroup interactions can inform efficient patient-treatment matching that optimizes individual outcomes.

Methods

These secondary analyses included young adults (n = 344; 68.6 % female, ages 18–25) randomized to GOAL or COMBO 12-week alcohol text message interventions and who completed 3-month follow-up (end of intervention). GOAL provided weekly support for drinking limit goals. COMBO included all GOAL features, in addition to pre-event feedback on drinking plans and post-event feedback on alcohol consumption. QUINT, a tree-based algorithm, aimed to identify treatment-subgroup interactions using 21 pre-treatment (baseline) characteristics (e.g., demographics, perceived risk of binge-drinking related harm, perceived number of peers drinking to intoxication) that predicted the primary outcome of binge drinking at follow-up.

Results

The algorithm used five pre-treatment characteristics (sex, race, perceived risk of binge drinking-related harm, perceived number of peers drinking to intoxication, and any cannabis use in the past 3 months) to identify 7 treatment-subgroup interactions. COMBO had greater effectiveness than GOAL, for example, for females who reported lower risk of binge-drinking related harm and no cannabis use in the past 3 months, whereas GOAL had greater effectiveness for females who reported higher risk of binge-drinking related harm and more peers who drank to intoxication. In comparison, GOAL had greater effectiveness than COMBO among White males, whereas males of other racial backgrounds benefitted more from COMBO than GOAL.

Conclusions

The identified treatment-subgroup interactions involving GOAL and COMBO indicated which intervention had greater effectiveness for which subgroups of individuals based on pre-treatment characteristics. These findings can help efficiently match individuals to effective interventions, bringing the field closer to personalized, precision care.

Clinical trials registration number: NCT02918565.