Development and Temporal Validation of a Discharge-Based Administrative Data Risk Model for 30-Day Unplanned...

Background: Postoperative complications after staging surgery for endometrial cancer often occur after discharge, and 30-day unplanned readmission is an important post-discharge outcome. Existing models lack same-day usability at discharge and sufficient validation, limiting discharge-time risk stratification and transitional care planning. This study aimed to develop and validate a discharge-time risk prediction model based on administrative data for 30-day unplanned hospital readmission among patients undergoing staging surgery for endometrial cancer. Methods: This retrospective cohort study was conducted in accordance with the TRIPOD reporting guidelines. We constructed a nationally representative adult cohort using the Healthcare Cost and Utilization Project Nationwide Readmissions Database (HCUP-NRD) from 2016 to 2022, including index hospitalizations discharged between January and November to ensure complete 30-day follow-up. Variables available at discharge (demographics, payer/income, presentation, inpatient course, discharge disposition, hospital characteristics, Elixhauser index) were included; age and length of stay were modeled with restricted cubic splines. Complex survey-weighted multivariable logistic regression was used for model development with 1000 bootstrap internal validations. Temporal validation within the NRD was performed using 2020-2021 and 2022 data. Decision curve analysis, subgroup, and sensitivity analyses were conducted. Results: A total of 89,627 hospitalizations were included; development and validation cohorts had balanced baselines. Internal validation showed moderate discrimination (AUC 0.719) and good calibration (Brier 0.045; intercept 0.014; slope 0.955). Discharge to home health or institutional care, non-elective or emergency admission, and a higher Elixhauser readmission index (a measure of comorbidity burden) were associated with increased risk. Age and length of stay showed nonlinear associations. Temporal validation yielded a similar performance (AUC 0.708-0.713; Brier 0.043-0.046), and decision curves indicated positive net clinical benefit. Conclusions: The model demonstrated stable discrimination and maintained good calibration after temporal validation and recalibration. It represents a candidate discharge-time decision-support tool, based on administrative data, for identifying patients at higher risk of 30-day unplanned readmission, and may support nurse-led transitional care planning. External validation and local recalibration are required before application in other healthcare settings.