Predicting In-Hospital Mortality in Patients Undergoing Complex Gastrointestinal Surgery Determining the Optimal Risk Adjustment Method

Objective: To compare the performance of Charlson/Deyo, Elixhauser, Disease Staging, and All Patient Refined Diagnosis-Related Groups (APR-DRGs) algorithms for predicting in-hospital mortality after 3 types of major abdominal surgeries: gastric, hepatic, and pancreatic resections. Design: Cross-sectional nationwide sample. Setting: Nationwide Inpatient Sample from 2002 to 2007. Patients: Adult patients (>= 18 years) hospitalized with a primary or secondary procedure of gastric, hepatic, or pancreatic resection between 2002 and 2007. Main Outcome Measures: Predicting in-hospital mortality using the 4 comorbidity algorithms. Logistic regression analyses were used and C statistics were calculated to assess the performance of the indexes. Risk adjustment methods were then compared. Results: In our study, we identified 46 395 gastric resections, 18 234 hepatic resections, and 15 443 pancreatic resections. Predicted in-hospital mortality rates according to the adjustment methods agreed for 43.8% to 74.6% of patients. In all types of resections, the APR-DRGs and Disease Staging algorithms predicted in-hospital mortality better than the Charlson/Deyo and Elixhauser indexes (P < .001). Compared with the Charlson/Deyo algorithm, the Elixhauser index was of higher accuracy in gastric resections (0.847 vs 0.792), hepatic resections (0.810 vs 0.757), and pancreatic resections (0.811 vs 0.741) (P < .001 for all comparisons). Higher accuracy of the Elixhauser algorithm compared with the Charlson/Deyo algorithm was not affected by diagnosis rank, multiple surgeries, or exclusion of transplant patients. Conclusions: Different comorbidity algorithms were validated in the surgical setting. The Disease Staging and APR-DRGs algorithms were highly accurate. For commonly used algorithms such as Charlson/Deyo and Elixhauser, the latter showed higher accuracy.