Predicting Thirty-Day Hospital Readmissions Among Adults With Multiple Chronic Conditions: A Critical Methodological Review

Abstract

Adults with multiple chronic conditions account for a disproportionate share of thirty-day hospital readmissions, yet most readmission risk models were developed and judged in general or single-disease populations. This critical methodological review examines what is required for such a model to be trustworthy in multimorbid adults. Literature was identified purposively and by citation chaining, anchored on published systematic reviews of readmission models and on the reporting and appraisal standards for prognostic research, and was appraised against CHARMS, PROBAST, TRIPOD+AI and published guidance on evaluating prediction model performance. The review does not pool performance estimates and reports no new quantitative synthesis. Prior syntheses consistently describe modest discrimination, sparse calibration reporting, and little external validation, and direct evidence in older adults shows widely used scores discriminating poorly. The review argues that the ceiling on performance lies more in what routine data measure than in how models are estimated: functional status, frailty, falls, medication access and social circumstances are repeatedly associated with readmission but seldom captured in administrative data. It sets out an appraisal framework, a predictor-domain map, a transportability checklist and a worked illustration of why modest discrimination yields low positive predictive value. The Multimorbidity Readmission Prediction Readiness framework links infrastructure, measurement, model governance and care-coordination capacity, and generates four testable propositions. Local recalibration, calibration reporting, decision-analytic evaluation and equity auditing should precede deployment.