Perioperative calculator to predict complications for patients with hepatocellular carcinoma resection.
446 Background: Hepatocellular carcinoma (HCC) is a common cancer worldwide, but few patients are optimal surgical candidates due to liver disease. Current prediction models (Childs-Pugh and MELD) poorly estimate serious morbidity risks for those patients considering resection. Methods: Using ACS-NSQIP database, patients with HCC were selected from 2006-2013. Patients included had a curative resection (partial hepatectomy or formal lobectomy) and were excluded if they had emergency surgery or disseminated cancer. Outcomes included 30 day morbidity and serious morbidity. Data was randomly divided into testing (n=1764) and validation (n=441) cohorts. Regression analyses of the testing cohorts were used to construct prediction models, then optimized using the validation cohort. Results: We identified 2,205 patients. Major morbidity or serious morbidity occurred in 34% (n=745) and 21% (n=456). Patient demographics are shown in Table 1. Factors significant for morbidity include age, surgical procedure, BMI, ASA class, hypertension, alkaline phosphatase (AP), bilirubin (Bili), BUN, AST, albumin and MELD score. Factors included in risk score for morbidity included MELD, ASA class, surgical procedure, platelet count, smoking and AP. The model predicts morbidity AUC 0.616 vs 0.597 for model and validation cohort respectively. Factors significant for serious morbidity included age, race, procedure, BMI, diabetes, ASA class, hypertension, AP, Bili, AST and albumin. Serious morbidity score included MELD, surgical procedure, platelet count, diabetes and AST. The model predicts serious morbidity AUC 0.566 vs 0.592 for model and validation cohort respectively. Conclusions: Our calculator predicts morbidity based on surgical procedure for patients undergoing HCC curative intent surgery. Risk prediction can assist in appropriately selecting patients for surgical vs medial therapy. [Table: see text]