Poster
151
(#151) Time-Varying Predictors of Relapse in Patients Treated with TV-46000 or Placebo in the RISE Study: A Machine Learning Analysis
Psych Congress 2025
Abstract: Background: Relapse prevention is a key treatment goal of schizophrenia. Previous analyses identified baseline characteristics associated with relapse with TV-46000 or placebo in the RISE phase 3 trial. This analysis aimed to identify relapse variables at baseline and during follow-up in the randomized maintenance stage in RISE.
Methods: In RISE, adults with schizophrenia stabilized on oral risperidone were randomized to TV-46000 or placebo. 175 and 171 baseline/follow-up variables were analyzed for TV-46000 and placebo patients, respectively. Machine learning using LASSO was fitted and validated with 200 bootstrapping samples to identify time-varying relapse predictors. Univariate Cox regression was used to calculate relapse risk (hazard ratio [HR]) for top predictors.
Results: Relapse data for 540 patients (361 TV-46000; 179 placebo) were analyzed. Mean (95% CI) internal validated C-index for the model was 0.70 (0.56-0.73) and 0.71 (0.63-0.76) for TV-46000 and placebo patients, respectively. For TV-46000, top predictors associated with higher risk of relapse were Personal and Social Performance (PSP) subscale item 4 (disturbing/aggressive behaviors) increase from baseline (HR, 95% CI: 7.25, 4.41-11.93; P.001) and PSP total score decrease from baseline (1.37, 1.27-1.49; P.001). For placebo, top predictors associated with lower relapse risk were PSP subscale item 4 decrease from baseline (0.19, 0.13-0.29; P.001), PSP total score increase from baseline (0.71, 0.65-0.78; P.001), and EuroQol 5-Dimension total score increase from baseline (0.88, 0.80-0.97; P=.01).
Conclusions: Changes in psychosocial functioning and quality-of-life scores were identified as key time-varying predictors of relapse in TV-46000- and placebo-treated patients.
Short Description: This post hoc analysis of the RISE study used machine learning to identify time-varying predictors of relapse in patients receiving TV-46000 or placebo. For TV-46000, relapse was more likely with Personal and Social Performance (PSP) subscale 4 score (disturbing/aggressive behaviors) increase from baseline and PSP total score decrease from baseline. For placebo, relapse was less likely with PSP subscale 4 score decrease, PSP total score increase, and EuroQol 5-Dimension score increase during follow-up.
Name of Sponsoring Organization(s): Teva Branded Pharmaceutical Products R&D LLC
Methods: In RISE, adults with schizophrenia stabilized on oral risperidone were randomized to TV-46000 or placebo. 175 and 171 baseline/follow-up variables were analyzed for TV-46000 and placebo patients, respectively. Machine learning using LASSO was fitted and validated with 200 bootstrapping samples to identify time-varying relapse predictors. Univariate Cox regression was used to calculate relapse risk (hazard ratio [HR]) for top predictors.
Results: Relapse data for 540 patients (361 TV-46000; 179 placebo) were analyzed. Mean (95% CI) internal validated C-index for the model was 0.70 (0.56-0.73) and 0.71 (0.63-0.76) for TV-46000 and placebo patients, respectively. For TV-46000, top predictors associated with higher risk of relapse were Personal and Social Performance (PSP) subscale item 4 (disturbing/aggressive behaviors) increase from baseline (HR, 95% CI: 7.25, 4.41-11.93; P.001) and PSP total score decrease from baseline (1.37, 1.27-1.49; P.001). For placebo, top predictors associated with lower relapse risk were PSP subscale item 4 decrease from baseline (0.19, 0.13-0.29; P.001), PSP total score increase from baseline (0.71, 0.65-0.78; P.001), and EuroQol 5-Dimension total score increase from baseline (0.88, 0.80-0.97; P=.01).
Conclusions: Changes in psychosocial functioning and quality-of-life scores were identified as key time-varying predictors of relapse in TV-46000- and placebo-treated patients.
Short Description: This post hoc analysis of the RISE study used machine learning to identify time-varying predictors of relapse in patients receiving TV-46000 or placebo. For TV-46000, relapse was more likely with Personal and Social Performance (PSP) subscale 4 score (disturbing/aggressive behaviors) increase from baseline and PSP total score decrease from baseline. For placebo, relapse was less likely with PSP subscale 4 score decrease, PSP total score increase, and EuroQol 5-Dimension score increase during follow-up.
Name of Sponsoring Organization(s): Teva Branded Pharmaceutical Products R&D LLC
