Integrating dynamic nomogram and machine learning for personalized disability prediction in elderly cardiometabolic multimorbidity: routine blood markers and mental health
Journal:
medRxiv
Published Date:
Jun 29, 2026
Abstract
Abstract Background: Disability prediction in elderly with cardiometabolic multimorbidity (CMM) is limited. We developed a dynamic nomogram and addressed three questions: predictive value of routine blood markers, depression vs. physical function, and plateau in CMM count.Methods: Using CHARLS data (46 predictors), disability defined as ADL/IADL impairment or self-report. LASSO and logistic regression built the nomogram, with mediation, RCS, trend tests, machine learning, and SHAP.Results: Six predictors (depression, cognition, stroke, CMM number, age, falls) formed a good-performing nomogram (https://xjbsashjtdx.shinyapps.io/DynamicNomogram/). Left-hand grip strength mediated 12.3% of strokes effect. Cognition showed an inverted U-shape (inflection point=12.043). CMM count plateaued after 3 diseases. Depression outranked grip strength and walking speed. SHAP identified HbA1c, creatinine, uric acid, hematocrit, fasting glucose, TyG, and CVAI as risk markers.Conclusions: The nomogram enables personalized risk stratification. Routine blood markers predict disability, depression dominates over physical function, and the CMM-disability relationship plateaus at CMM[≥]3.
