A multidomain early warning model for 12-month adverse outcomes in older adults with multimorbidity: a risk-stratified public health approach.

OBJECTIVE

To develop and internally validate a multidomain early warning model for 12-month adverse outcomes in older adults with multimorbidity, and to explore its potential for risk-stratified follow-up.

METHODS

This retrospective study included 690 older adults (≥60 years) with multimorbidity from a Chinese tertiary hospital, split into derivation (n = 483) and validation (n = 207) sets. The primary outcome was a composite of unplanned hospitalization, ≥2 emergency visits, new functional decline, or all-cause mortality within 12 months. LASSO and multivariable logistic regression were used for model development. A nomogram was constructed and assessed by discrimination, calibration, and decision curve analysis. Patients were stratified into low, intermediate, and high-risk groups. Multimorbidity network analysis and a retrospective scenario analysis of a model-driven stratified intervention were also performed.

RESULTS

The composite outcome occurred in 146 patients (21.2%). Seven predictors were retained: age ≥75 years, ≥4 chronic diseases, prior hospitalization, ADL impairment, malnutrition, depression, and chronic pain. The model showed good discrimination in the derivation set (AUC 0.847), random split validation set (AUC 0.826), and temporal validation set (AUC 0.819). Bootstrap internal validation yielded an optimism-corrected AUC of 0.831. Calibration was acceptable across all validation methods (calibration slopes: 0.94-1.02; Brier scores: 0.152-0.171). Using pragmatic thresholds, 41.2, 42.0, and 16.8% of patients were classified as low, intermediate, and high risk, respectively. The high-risk group accounted for 41.1% of all events. Network analysis revealed a prominent cardio-metabolic disease cluster. In a retrospective scenario analysis, a model-driven stratified intervention was projected to reduce composite adverse outcomes by 25.3%.

CONCLUSION

This internally validated model identifies a small high-risk subgroup that carries a large share of adverse events. External validation in community health centers, secondary hospitals, and other regions is required before implementation. The model provides a preliminary framework for targeted follow-up and risk-informed planning, but its generalizability remains unproven.