Evaluating NSQIP Outcomes According to the Clavien–Dindo Classification: A Model to Estimate Global Outcome Measures Following Hepatopancreaticobiliary Surgery

Background: The National Surgical Quality Improvement Program (NSQIP) database provides one of the largest repositories of surgical outcome data—guiding local, national, and international quality improvement and research. We aim to describe a model to estimate Clavien–Dindo complication (CDC) rates from NSQIP data to enable comprehensive outcome measurement, allowing an NSQIP-based surrogate measure for longer-term outcomes. Methods: This is a validation study of a model to estimate CDCs from NSQIP data for pancreaticoduodenectomy (PD) and hepatic resection (HR). The primary objective of this study is to evaluate whether our method to estimate CDCs ≥ 3 outcomes from NSQIP data results in similar serious complication rates to large benchmark studies on outcomes following PD and HR. Secondary outcomes evaluate whether specific NSQIP outcomes provide adequate information to estimate CDC grades I-V following PD and HR. Results: We evaluated 20,575 patients undergoing PD, with 71.3% having pancreatic ductal adenocarcinoma. Comparing CDCs ≥ 3 complications for NSQIP and benchmark PD patients, we estimated a 23.2% rate with our model, which was significantly lower than the reported 27.6% in the benchmark study (p < 0.001). Additionally, the benchmark reported higher complication rates for every CDC grade compared to our estimates using NSQIP PD patients (p < 0.001). Further, we evaluated 29,809 patients within NSQIP undergoing HR, where most patients with a diagnosis listed had colorectal cancer metastases (30.8%). Compared to the benchmark HR study (n = 2159), the NSQIP patients were less likely to have hepatic resection for malignancy (57.7% vs. 84.0%; p < 0.001). Comparing CDCs ≥ 3 complications following HR demonstrated that rates were clinically similar (13.0% vs. 15.8%) but statistically different between the benchmark study and NSQIP data (p < 0.001). Additionally, the NSQIP patients had lower rates of estimated complications for nearly all CDC grades (p < 0.001). Conclusions: This is the first reported method to estimate aggregate morbidity from NSQIP data. Results demonstrate that despite differences in this and comparator cohorts, this model may underestimate CDC grade 1–2 complications but provide similar rates of CDCs ≥ 3 complications compared to benchmark studies. Future studies to validate or modify this estimation method are warranted and may allow extrapolation of short-term NSQIP measures to oncologic, quality of life, and long-term outcomes.