Developing and validating a center-specific preoperative prediction calculator for risk of pancreaticoduodenectomy.

BACKGROUND: The American College of Surgeons (ACS) Surgical Risk Calculator predicts postoperative risk based on preoperative variables. The ACS model was compared to an institution-specific risk calculator for pancreaticoduodenectomy (PD).

METHODS: Observed outcomes were compared with those predicted by the ACS and institutional models. Receiver operating characteristic (ROC) analysis evaluated the models' predictive ability. Institutional models were evaluated with retrospective and prospective internal validation.

RESULTS: Brier scores indicate equivalent aggregate predictive ability. ROC values for the institutional model (ROC: 0.675-0.881, P < 0.01) indicate superior individual event occurrence prediction (ACS ROC: 0.404-0.749, P < 0.01-0.860). Institutional models' accuracy was upheld in retrospective (ROC: 0.765-0.912) and prospective (ROC: 0.882-0.974) internal validation.

CONCLUSIONS: Identifying higher-risk patients allows for individualized care. While ACS and institutional models accurately predict average complication occurrence, the institutional models are superior at predicting individualized outcomes. Predictive metrics specific to PD center volume may more accurately predict outcomes.