Predicting Days at Home after Elective Surgery for Gastrointestinal Cancer using HOMEDAYS: Prediction Model Development and Internal Validation.

Abstract

Patient-centered outcomes after gastrointestinal cancer surgery are poorly captured by traditional metrics, limiting effective preoperative counseling. Days at home within 90 days after surgery (DAH-90) integrates survival and healthcare utilization but lacks a clinically applicable prediction tool. Adults undergoing elective gastrointestinal cancer resection in Ontario, Canada (2003-2021) were identified using population-based administrative data. A multivariable prediction model (HOMEDAYS) was developed using quantile regression to estimate median DAH-90 based on preoperative patient, cancer, and treatment factors. Model performance was assessed using mean absolute error (MAE), calibration (slope, intercept, decile plots), and discrimination (g-index), with internal validation via 500-bootstrap resampling and prespecified sensitivity analyses. Among 91,270 patients, median DAH-90 was 82 days (IQR 77-85). The final model incorporated 23 predictors with 3 interaction terms and modeled age using restricted cubic splines. Performance demonstrated strong accuracy (MAE 8.67), good calibration (slope 1.0, intercept 0.29), and discrimination (g-index 3.27). Optimism-corrected metrics remained stable (MAE 8.68; slope 1.0; intercept 0.29; g-index 3.26). Calibration across deciles showed minimal deviation between predicted and observed DAH-90 (0.1-0.6 days). Model performance was robust across multiple sensitivity analyses, including alternative outcome definitions and additional predictors. HOMEDAYS provides accurate, internally validated predictions of DAH-90 using preoperative variables, enabling individualized, patient-centered risk communication for elective gastrointestinal cancer surgery. This tool may enhance shared decision-making and perioperative preparedness.