Improving Adjuvant Liver-Directed Treatment Recommendations for Unresectable Hepatocellular Carcinoma: An Artificial Intelligence-Based Decision-Making Tool

Abstract

<h3>Purpose/Objective(s)</h3> Liver-directed therapy (LDT) following transarterial chemoembolization (TACE) can lead to improvement in survival for selected patients with unresectable hepatocellular carcinoma (HCC). However, there is uncertainty in the appropriate application and modality of therapy in current clinical practice guidelines. The aim of this study was to develop a proof-of-concept, machine-learning (ML) tool for treatment recommendation in patients previously treated with TACE and identify patients who may benefit from additional treatment with combination stereotactic body radiotherapy (SBRT) or radiofrequency ablation (RFA). <h3>Materials/Methods</h3> This retrospective, observational study was based on data from an urban, academic hospital system selecting for patients diagnosed with stage I-III HCC from January 1, 2008 to December 31, 2018, treated with TACE, followed by adjuvant RFA, SBRT, or no additional liver-directed modality. The dataset was divided in half into training and testing groups. A feed-forward ML ensemble model based on the DeepSurv algorithm, provided a treatment recommendation using pair-wise assessments between each potential treatment option and an estimated benefit in survival. Final ensemble model treatment recommendation was selected based on the treatment yielding the best hazard rate among all pairwise models on an individual patient basis. <h3>Results</h3> Two hundred thirty-seven patients met inclusion criteria, of whom 54 (23%) and 49 (21%) received combination TACE and SBRT or TACE and RFA respectively. The ML model recommended a different consolidative modality in 33% of cases among patients who had previously received combination treatment. Patients treated in concordance with model recommendations had a numerical improvement in overall survival (HR 0.63; p=0.160) and significant improvement in progression free survival (HR 0.50; p=0.007). The most critical features for model treatment recommendation were cause of cirrhosis, stage of cancer, and ALBI grade (a measure of liver function). <h3>Conclusion</h3> In this proof-of-concept study, we developed an ensemble ML model which was able to provide treatment recommendations for HCC who had undergone prior TACE therapy. Our data suggest additional treatment in-line with model recommendations was associated with significant improvement in progression free survival suggesting a potential benefit for ML-guided medical decision making. In the future, this tool will provide the framework for a prospective study where treatment decisions for complex clinical cases may be enhanced through personalized, machine-learning recommendations.