Background: Pancreatoduodenectomy (PD) is often performed in frail patients and is associated with significant morbidity. The five-factor modified frailty index (mFI-5) has been utilized to predict adverse postoperative outcomes, but has not been tested in PD. We aimed to develop risk tools to generate and predict 30-day outcomes after PD and compare their performance with the mFI-5. Risk tools were then used to generate a PD-specific calculator.
Methods: Elective PDs from the 2014-2016 ACS NSQIP® Procedure Targeted Pancreatectomy PUFs were identified. Multivariable logistic regression models were constructed to predict postoperative mortality, any complication, serious complication, clinically relevant postoperative pancreatic fistula (CR-POPF), and discharge not-to-home. Predictive accuracy was evaluated through repeated stratified tenfold cross-validation and compared to the mFI-5.
Results: Nine thousand eight hundred sixty-seven PDs were captured. Nine risk factors were retained: sex, age, BMI, DM, HTN, ASA classification, pancreatic duct size, gland texture, and adenocarcinoma. Cross-validated C-indices ranged from 0.49 to 0.61 for the mFI-5 and 0.63 to 0.75 for our risk models. The best-performing model was for discharge not-to-home (C = 0.75), and the model delivering the largest increase in predictive accuracy was for CR-POPF (CmFI-5/Model = 0.49/0.70). A user-friendly risk calculator was created predicting the five outcomes of interest.
Conclusion: We have created a PD-specific risk calculator that outperforms the mFI-5. This calculator may serve as a useful adjunct in shared decision-making for patients and surgeons.
Keywords: Calculator; Frailty; Pancreatic surgery; Pancreatoduodenectomy; Whipple.