USING ROUTINE LABORATORY RESULTS TO PREDICT 30-DAY ALL-CAUSE READMISSION: A RETROSPECTIVE COHORT STUDY AT A TERTIARY HOSPITAL IN RIYADH, SAUDI ARABIA

Abstract

Background Thirty-day unplanned readmission is a key transition-of-care metric. We developed and internally validated a pragmatic prediction approach using routinely available discharge-time information, including laboratories from the prior 24–48 hours.

Methods Retrospective cohort of consecutive adult discharges from medical and surgical wards. Eligible index stays ended in live discharge; deaths, direct acute transfers, discharges against medical advice without window labs, and planned returns were excluded. Outcome was all-cause, unplanned readmission within 30 days to the same hospital. Two logistic models were fit: a core model (age, sex, length of stay, prior 90-day discharge, prior 12-month admissions, HOSPITAL points) and an extended model adding discharge-proximal labs (e.g., CRP, eGFR, troponin). Model fit, calibration, discrimination, classification, and collinearity were assessed.

Results Among 913 patients (mean age 55.8 years; 54.8% male), 30-day readmission was 12.3% (n=112). The core model retained age, length of stay, prior 90-day discharge, and HOSPITAL points, with good calibration (Hosmer–Lemeshow p=0.239), high specificity, and low sensitivity at the default threshold. The extended model (N=339) added CRP and eGFR, modestly increased Nagelkerke R² (0.10→0.14) and sensitivity (to 8.8%); AUC=0.683; collinearity acceptable (VIF 1.01–1.41).

Conclusion Recent utilization and near-discharge biological instability jointly shaped 30-day readmission. A universal core model is deployable for all patients; routine laboratories provide incremental signal where available.