Usefulness of the ACS-NSQIP risk calculator in liver resection for primary liver malignancy in an Australian cohort

Abstract

The ACS-NSQIP risk calculator predicts 30-day post-operative outcomes, but its performance in liver resection for primary malignancy is not clear in local populations. This study evaluated the predictive accuracy of this tool in patients who underwent resection for primary liver cancer. A retrospective review of patients who underwent liver resection for hepatocellular carcinoma or cholangiocarcinoma between 1998 and 2023 by a single surgeon was conducted. ACS-NSQIP predicted risks were compared with observed outcomes using a one-sample t-test. Calibration was assessed using Brier scores, and discrimination with the C-statistic. For a consecutive series of 158 patients, the mean predicted risk for serious complications, sepsis, return to the operating room (ROT), and mortality by the ACS-NSQIP calculator did not differ significantly from observed event rates, however, discrimination was poor for most outcomes. It underestimated any complications (30.4% vs. 18.9%), cardiac events (7.6% vs. 2.0%), venous thromboembolism (7.0% vs. 2.5%), and length of stay (3.5 days) but overestimated readmissions (5.1% vs. 10.1%) and pneumonia (0.6% vs. 3.6%). Calibration was poorest for any complications (Brier 0.249), with acceptable discrimination for ROT (AUC 0.73). Patients with a predicted serious complication risk < 13.65% were more likely to achieve textbook oncological outcomes (TOO) (71.7% vs. 50.9%, p  = 0.016), however, this was not an independent predictor of achieving TOO on multivariable regression (OR 0.68, 95% CI = 0.28–1.67, p  = 0.396). The ACS-NSQIP calculator showed reasonable agreement between predicted and observed event rates for major events, but poor discrimination for most individual outcomes. It underestimated minor morbidity possibly because disease/procedure-specific factors are not adequately captured.