Introduction: Knowing the risk of surgery before performing one is essential for informed consent and decisions making for the doctors, patients, and their families. Complexities of surgery and pancreatic cancer aggressiveness are two factors that contribute to high mortality rate in pancreaticoduodenectomy. There has been a few scoring systems developed to predict mortality after pancreaticoduodenectomy. However, external validation of these scoring systems failed to achieve good predictive value consistently. This could happen due to differences in patient demographics, variables and their cut-off, patient samples, and hospital or health care standard where these scoring systems were constructed. This study aims to develop a novel scoring system to predict postoperative mortality and to compare its performance with the existing scoring systems.
Methods: Data were collected retrospectively from all patients who underwent pancreaticoduodenectomy in Dr. Soetomo General Hospital, starting from January 2012 to October 2022. 75 patients were included based on the inclusion criteria. Multivariate logistic regression analysis was performed to select multiple risk factors correlated with in-house mortality. A novel scoring system was formed using these risk factors. We also assessed and compared the predictive value of the existing scores (Parikh, Nugroho, and WHipple-ABACUS) using the ROC curve. Chi square test with odds ratio calculation and logistic regression were used in this study.
Results: Post pancreaticoduodenectomy mortality rate was 28% (21 out of 75 patients) with 4 statistically significant risk factors. Those factors are age > 52 years. (OR 7.435; 95% CI 1.113-49.677, p = 0.038), comorbidities (OR 5.793; 95% CI 1.116-30.067, p = 0.037), American Society of Anaesthesiology (ASA) 3 (OR 8.932; 95% CI 1.329-60.022, p = 0.024), and total bilirubin > 16.6 mg/dL (OR 8.329; 95% CI 1.577-43.983, p = 0,013). These factors were used to form a novel scoring system. The predictive performance of the newly created scoring system was 0.911 (95% CI 0.823-0.965). While external validation of the existing scoring systems was as follows: Parikh 0.684 (95% CI 0.567-9.787), Nugroho 0.829 (95% CI 0.724-0.906), and WHipple-ABACUS 0.736 (95% CI 0.62-0.832).
Conclusion: The newly created scoring system uses 4 relatively easy and accessible preoperative variables to predict post-pancreaticoduodenectomy mortality. However, it still requires external validation in multicenter hospitals with large numbers of patients.