Utilizing Machine Learning for Pre- and Postoperative Assessment of Patients Undergoing Resection for BCLC-0, A and B Hepatocellular Carcinoma: Implications for Resection Beyond the BCLC Guidelines

Diamantis I. Tsilimigras,Rittal Mehta,Dimitrios Moris,Kota Sahara,Fabio Bagante,Anghela Z. Paredes,Ayesha Farooq,Francesca Ratti,Hugo P. Marques,Silvia Silva,Olivier Soubrane,Vincent Lam,George A. Poultsides,Irinel Popescu,Razvan Grigorie,Sorin Alexandrescu,Guillaume Martel,Aklile Workneh,Alfredo Guglielmi,Tom Hugh,Luca Aldrighetti,Itaru Endo,Timothy M. Pawlik
DOI: https://doi.org/10.1245/s10434-019-08025-z
IF: 4.339
2019-11-06
Annals of Surgical Oncology
Abstract:BackgroundThere is an ongoing debate about expanding the resection criteria for hepatocellular carcinoma (HCC) beyond the Barcelona Clinic Liver Cancer (BCLC) guidelines. We sought to determine the factors that held the most prognostic weight in the pre- and postoperative setting for each BCLC stage by applying a machine learning method.MethodsPatients who underwent resection for BCLC-0, A and B HCC between 2000 and 2017 were identified from an international multi-institutional database. A Classification and Regression Tree (CART) model was used to generate homogeneous groups of patients relative to overall survival (OS) based on pre- and postoperative factors.ResultsAmong 976 patients, 63 (6.5%) had BCLC-0, 745 (76.3%) had BCLC-A, and 168 (17.2%) had BCLC-B HCC. Five-year OS among BCLC-0/A and BCLC-B patients was 64.2% versus 50.2%, respectively (p = 0.011). The preoperative CART model selected α-fetoprotein (AFP) and Charlson comorbidity score (CCS) as the first and second most important preoperative factors of OS among BCLC-0/A patients, whereas radiologic tumor burden score (TBS) was the best predictor of OS among BCLC-B patients. The postoperative CART model revealed lymphovascular invasion as the best postoperative predictor of OS among BCLC-0/A patients, whereas TBS remained the best predictor of long-term outcomes among BCLC-B patients in the postoperative setting. On multivariable analysis, pathologic TBS independently predicted worse OS among BCLC-0/A (hazard ratio [HR] 1.04, 95% confidence interval [CI] 1.02–1.07) and BCLC-B patients (HR 1.13, 95% CI 1.06–1.19) undergoing resection.ConclusionPrognostic stratification of patients undergoing resection for HCC within and beyond the BCLC resection criteria should include assessment of AFP and comorbidities for BCLC-0/A patients, as well as tumor burden for BCLC-B patients.
oncology,surgery
What problem does this paper attempt to address?