Predictors of poor medication adherence in patients undergoing peritoneal dialysis: a LASSO-based risk model from a cross-sectional study in Xinjiang, China

Medication nonadherence is a major challenge in peritoneal dialysis, yet predictive tools are available. This study aimed to identify key determinants of poor medication adherence and develop a clinically applicable prediction model. A cross-sectional survey was conducted among 401 patients undergoing maintenance peritoneal dialysis in Xinjiang, China. Data included sociodemographic characteristics, medication and medical history, and validated scales for self-management, self-efficacy, and helplessness. Least absolute shrinkage and selection operator (LASSO) logistic regression with 10‑fold cross-validation was used to select predictors and construct the model; model performance was assessed using the area under the curve (AUC), Brier score, and calibration curve. Overall, 286 participants (71.3%) had poor medication adherence. The LASSO model (λ = 0.054) retained three predictors: medication side effects, self-management level, and helplessness. Compared with the full‑variable model, the LASSO model achieved lower AIC (370.19 vs. 371.95) and BIC (386.17 vs. 399.91), indicating greater parsimony. The stepwise model had a slightly lower AIC, but its BIC was higher. Discrimination and calibration were comparable across models. Poor adherence is highly prevalent and primarily associated with medication side effects, inadequate self-management, and helplessness. The LASSO‑based model provides a concise, interpretable tool for identifying high‑risk patients to support targeted interventions.