Lorenzo Lorenzo-Luaces1, Natalie Rodriguez-Quintana1, Tennisha N Riley1,2, John R Weisz3. 1. Department of Psychological and Brain Sciences, Indiana University-Bloomington, Bloomington, IN, USA. 2. Center for Research on Race and Ethnicity in Society (CRRES), Indiana University-Bloomington, Bloomington, IN, USA. 3. Department of Psychology, Harvard University, Cambridge, MA, USA.
Abstract
Introduction: Researchers have proposed that predicting who is a likely placebo responder may help guide treatment allocations to treatment regimens that differ in intensity. Methods: We used data from the Treatment of Adolescent Depression Study (TADS) in which adolescents (n = 439) were randomized 1:1:1:1 to placebo, cognitive-behavioral therapy (CBT), medications (MEDs), or their combination (COMB). We developed a prognostic index (PI) in the placebo group to predict self-reported (RADS) and observer-rated (CDRS) depression outcomes using elastic net regularization. We explored whether the PIs moderated outcomes in the treatment conditions. Results:PI-CDRS was predicted by multiple variables but it did not moderate outcomes. PI-RADS was predicted by baseline severity, age, sleep problems, expectations, maternal depression, and the action stage of change. It moderated outcomes such that there were treatment differences for less placebo-responsive patients. For participants prone to placebo response, type of treatment had no statistically significant impact on outcomes. Baseline depression severity accounted for this effect: treatment differences were small and non-significant for patients with milder depression but larger in more severely depressed patients. Discussion: Future work should investigate whether multiple variable explain outcomes beyond severity as well as complex interactions between severity and other variables.
RCT Entities:
Introduction: Researchers have proposed that predicting who is a likely placebo responder may help guide treatment allocations to treatment regimens that differ in intensity. Methods: We used data from the Treatment of Adolescent Depression Study (TADS) in which adolescents (n = 439) were randomized 1:1:1:1 to placebo, cognitive-behavioral therapy (CBT), medications (MEDs), or their combination (COMB). We developed a prognostic index (PI) in the placebo group to predict self-reported (RADS) and observer-rated (CDRS) depression outcomes using elastic net regularization. We explored whether the PIs moderated outcomes in the treatment conditions. Results:PI-CDRS was predicted by multiple variables but it did not moderate outcomes. PI-RADS was predicted by baseline severity, age, sleep problems, expectations, maternal depression, and the action stage of change. It moderated outcomes such that there were treatment differences for less placebo-responsive patients. For participants prone to placebo response, type of treatment had no statistically significant impact on outcomes. Baseline depression severity accounted for this effect: treatment differences were small and non-significant for patients with milder depression but larger in more severely depressedpatients. Discussion: Future work should investigate whether multiple variable explain outcomes beyond severity as well as complex interactions between severity and other variables.
Entities:
Keywords:
depression; machine learning; personalized medicine; risk stratification; stepped care
Authors: Erica A Smith; William P Horan; Dominique Demolle; Peter Schueler; Dong-Jing Fu; Ariana E Anderson; Joseph Geraci; Florence Butlen-Ducuing; Jasmine Link; Ni A Khin; Robert Morlock; Larry D Alphs Journal: Innov Clin Neurosci Date: 2022 Jan-Mar