Risk Prediction of 2-Year Incident Cognitive Impairment in Maintenance Hemodialysis Patients

Background: Cognitive impairment is common but often underrecognized in maintenance hemodialysis (MHD) patients and is associated with adverse outcomes. Short-term risk prediction tools in this population remain limited. We aimed to develop and internally validate a model for predicting 2-year incident cognitive impairment in MHD patients with normal baseline cognition. Methods: We conducted a multicenter prospective cohort study in five hospitals in Beijing and Tianjin, China. Adults with normal baseline cognition were enrolled between July and September 2022 and reassessed at 24 months (±3 months) using the Beijing version of the Montreal Cognitive Assessment. Among 625 eligible participants, 441 (71%) completed follow-up and were analyzed. Candidate predictors were prespecified based on prior evidence and clinical relevance. A prediction model was developed using least absolute shrinkage and selection operator logistic regression and internally validated with 1,000 bootstrap iterations. Results: At 2 years, 158 of 441 participants (36%) developed incident cognitive impairment. Six predictors were retained: age, years of education, hearing level, depression status, sleep quality, and hemoglobin level. The final model showed good discrimination and calibration, with an apparent area under the receiver operating characteristic curve (AUC) of 0.831 ( 95% CI 0.795–0.866) and a Brier score of 0.145 ( 95% CI 0.130–0.161). Compared with a baseline model including age and years of education, the final model showed better discrimination (AUC 0.831 vs. 0.797; DeLong test, P = 0.002) and improved overall fit (likelihood ratio test, P < 0.001). Bootstrap validation yielded an optimism-corrected AUC of 0.800 ( 95% CI 0.724–0.878) and a corrected Brier score of 0.151. A predicted probability of 0.364 identified higher-risk individuals. Conclusions: This prospectively developed model estimated 2-year risk of incident cognitive impairment in MHD patients using readily obtainable indicators. It may support risk stratification, prioritized cognitive screening, and targeted assessment of potentially modifiable factors.