Predicting 365-Day Lower-Limb Amputation at Discharge After Diabetes-Related Foot Ulcer Hospitalization: Development and Temporal Validation

Abstract Background Patients discharged after diabetes-related foot ulcer hospitalization remain at risk of limb loss. We developed and temporally validated discharge-time models for 365-day amputation, comparing clinical-plus-discharge predictors with a fuller set including contextual information. Methods Epic Cosmos provided 431,493 development patients discharged in 2016–2024 and 36,823 temporal-validation patients discharged in January–June 2025. Cause-specific Cox models estimated any and major amputation with death as a competing event. Model C included clinical, contextual, and discharge variables; Model D omitted contextual variables. Models were compared using discrimination, calibration, and decision curves. Results Observed 365-day cumulative incidences were 19.3% for any amputation and 6.4% for major amputation. Model C areas under the receiver operating characteristic curve were 0.710 and 0.737, compared with 0.711 and 0.736 for Model D. Corresponding calibration slopes were 1.003 and 1.026 for Model C and 1.010 and 1.025 for Model D. Brier scores were similar. Gangrene and peripheral artery disease had the largest joint Wald statistics in Model C. Its highest-risk quintile contained 40.6% of any-amputation and 49.6% of major-amputation events. Conclusions Models C and D showed similar performance, with moderate discrimination and risk enrichment. These findings do not establish superiority of the fuller model, irrelevance of contextual factors, or improved patient outcomes. Independent validation, local recalibration, and prospective impact evaluation are required before clinical use.