Funding
National Cancer Institute (NCI)
United States Department of Health and Human Services
Find out more...Clinical and Translational Science Institute, University of California, Los Angeles (CTSI)
ARTICLE ABSTRACT
Predicting an individual's risk of treatment discontinuation is critical for the implementation of precision chemoprevention. We developed partly conditional survival models to predict discontinuation of tamoxifen or anastrozole using patient-reported outcome (PRO) data from postmenopausal women with ductal carcinoma in situ enrolled in the NSABP B-35 clinical trial. In a secondary analysis of the NSABP B-35 clinical trial PRO data, we proposed two models for treatment discontinuation within each treatment arm (anastrozole or tamoxifen treated patients) using partly conditional Cox-type models with time-dependent covariates. A 70/30 split of the sample was used for the training and validation datasets. The predictive performance of the models was evaluated using calibration and discrimination measures based on the Brier score and AUC from time-dependent ROC curves. The predictive models stratified high-risk versus low-risk early discontinuation at a 6-month horizon. For anastrozole-treated patients, predictive factors included baseline body mass index (BMI) and longitudinal patient-reported symptoms such as insomnia, joint pain, hot flashes, headaches, gynecologic symptoms, and vaginal discharge, all collected up to 12 months [Brier score, 0.039; AUC, 0.76; 95% confidence interval (CI), 0.57–0.95]. As for tamoxifen-treated patients, predictive factors included baseline BMI, and time-dependent covariates: cognitive problems, feelings of happiness, calmness, weight problems, and pain (Brier score, 0.032; AUC, 0.78; 95% CI, 0.65–0.91). A real-time calculator based on these models was developed in Shiny to create a web-based application with a future goal to aid healthcare professionals in decision-making.
The dynamic prediction provided by partly conditional models offers valuable insights into the treatment discontinuation risks using PRO data collected over time from clinical trial participants. This tool may benefit healthcare professionals in identifying patients at high risk of premature treatment discontinuation and support interventions to prevent potential discontinuation.
