A simple risk score to predict in-hospital mortality after pancreatic resection for cancer

Abstract

Background: Pancreatectomy for cancer continues to have substantial perioperative risk, and the factors affecting mortality are ill defined. An integer-based risk score based on national data might help clarify the risk of in-hospital mortality in patients undergoing pancreatic resection.

Methods: Records with the diagnosis of pancreatic cancer were queried from the Nationwide Inpatient Sample for 1998-2006. Procedures were categorized as proximal, distal, or nonspecified pancreatectomies on the basis of ICD-9 codes. Logistic regression and bootstrap methods were used to create an integer risk score for estimating the risk of in-hospital mortality using patient demographics, comorbidities (Charlson comorbidity score), procedure, and hospital type. A random sample of 80% of the cohort was used to create the risk score with a 20% internal validation set.

Results: A total of 5715 patient discharges were identified. Composite in-hospital mortality was 5.8%. Predictors used for the final model were age group, Charlson score, sex, type of pancreatectomy, and hospital volume status (low-, medium-, or high-volume center). Integer values were assigned to these characteristics and then used for calculating an additive score. Three clinically useful score groups were defined to stratify the risk of in-hospital mortality (mortality was 2.0, 6.2, and 13.9%, respectively; P < 0.0001), with a 6.95-fold difference between the low- and high-risk groups. There was sufficient discrimination of both the derivation set and the validation set, with c statistics of 0.71 and 0.72, respectively.

Conclusions: An integer-based risk score can be used to accurately predict in-hospital mortality after pancreatectomy and may be useful for preoperative risk stratification and patient counseling.