Outcomes after major cancer surgery: Can we predict them using parsimonious models?

Abstract

811 Background: Major cancer surgery is associated with significant risks of morbidity/mortality. Retrospective studies have demonstrated an association between Surgical Apgar Score (SAS) and postoperative risk of serious complications (SC). This study prospectively evaluated the predictive value of SAS to predict SC, as well as other adverse postoperative outcomes and length of stay (LOS), both singly and in combination with parsimonious measures of “fitness for surgery” (American Society of Anesthesiology [ASA] classification) and surgical complexity (work relative value units [wRVU]). Methods: Demographic, comorbidity, procedure, intraoperative, and outcome data was collected prospectively for 442 cancer patients undergoing elective major surgery between 2014-17. ASA and wRVU were assigned preoperatively; SAS was calculated postoperatively. Logistic regression was used to analyze association of ASA versus (vs) wRVU vs SAS vs ASA/wRVU/SAS (combined) with perioperative outcomes, including SC, return to the operating room (ROR), not being discharged to home, and unplanned readmission; areas under receiver operator characteristic curves were calculated to assess predictive accuracy. Accelerated failure time models were used to analyze associations with LOS and compared using Harrell’s concordance index. Results: Predictive accuracy of SAS for SC was modest (AUC0.655) and not improved when controlling for ASA and wRVU (both of which had poor predictive accuracy for SC). Both wRVU (AUC 0.634) and SAS (AUC 0.663) had modest predictive accuracies for ROR, whereas the predictive accuracy of ASA (AUC 0.749) surpassed that of wRVU (AUC 0.630) for not being discharged to home. All 3 measures were poor at predicting readmission. In contrast, the predictive accuracy of ASA, wRVU, and SAS for LOS was highest when combined (AUC 0.699). Conclusions: Commonly used, simple measures of comorbidity/functional status and surgical complexity can help predict risk of ROR and not being discharged to home (respectively), whereas only SAS has sufficient (albeit modest) discriminatory ability to predict risk of SC. All three measures are too coarse to predict unplanned readmission, but when used in combination, can help predict LOS.