The machine learning (ML) decision support model developed by Korean researchers may be a useful tool to guide treatment decisions for hepatocellular carcinoma (HCC) by distinguishing between patients' risk of death from liver transplant (LT) or surgical resection (SR) by deriving that it is beneficial to optimize survival by optimizing the introduction of liver transplant (LT) or clinical treatments, and thus to guide treatment decisions for hepatocellular carcinoma (HCC). Survival results.1
“A key contribution of this study is the potential identification of clear patient subgroups by integrating comprehensive clinical variables into the ML algorithm,” said research investigator Kim and colleagues in this paper.1
The selected model initially stratified a large retrospective cohort of HCC patients who were elicited from a national cancer registry that received LT or SR between 2008 and 2018 by risk of death. Following risk stratification, Kaplan-Meier analysis was performed to estimate survival outcomes.
Among patients who received LT (n = 296), the model classified 119 patients as high risk and 177 patients as low risk. The low-risk group showed significantly better survival rates compared to the high-risk group, with a hazard ratio (HR) of 0.25 (95% CI, 0.15-0.42). p <.001). Median overall survival (OS) was not reached in either group.
Of the patients who received SR (n = 3619), 1028 patients were classified as high risk and 2591 were classified as low risk. Additionally, the low-risk group had a HR of 0.17 (95% CI, 0.15 – .019), which showed significantly better survival rates compared to the high-risk group. p <.001). The median OS in the high-risk group was 32.5 months (95% CI, 30.2–34.7). For low-risk groups, the median OS was not reached.
Based on the findings of risk stratification, the model determined the subgroups of patients who would clearly benefit from LT (LT-Favorable), which patients had comparable results to SR (adverse to LT) in the context of survival, and individual treatment recommendations were based. Specifically, the model recommended treatment for patients, resulting in a lower estimated risk of death. This model recommended SR in 74.7% (lt-nonfavorable) of LT recipients. Conversely, LT was recommended for 19.4% of SR recipients (LT-Favorable).
Finally, in counterfactual analysis, the model estimated and compared survival outcomes under ML-based treatment recommendations and actual recommendations. In particular, ML-based treatment was significantly associated with improved survival rates of 0.46 (95% CI, 0.42–0.50 in HR. p <.001), which corresponds to a 54% reduction in the risk of death compared to actual treatment.
What are the limits of research?
To ensure the actual accuracy and generalizability of the model, the model was validated against an external validation cohort of patients with HCC who underwent either LT or SR between 2009 and 2020, derived from hospital data sets (n = 614). However, as a retrospective study, investigators note that designs are affected by selection bias and residual confounding from uncomputed variables.1
“Despite the obvious strengths of the study, we did not address some variables, including previous liver-directed therapy, surgical resectability, patient comorbidities, tumor biology assessment, cohort allogeneity, and limited follow-up, so results should be examined with caution,” commented Varvara A. Professor of Surgery and Internal Medicine at the University of Minnesota.2
Furthermore, this study did not consider the surgical resectability and anatomical distribution of tumors in SR, as well as the cardiovascular compatibility of patients to undergo LT.
What is the clinical meaning?
The current literature establishes that LT produces better long-term survival outcomes compared to SR.3 Nevertheless, treatment choice remains a challenge due to a lack of organ donors and patient heterogeneity.1 This study suggests that ML could be utilized to address this challenge, providing practical tools for increasing clinical benefits, optimizing organ allocation, improving patient provider communication, and sharing decision-making.
“While future ML-based models can become an integral aspect of individuality medicine due to the evolution of technology and data management, even with full predictions, decisions to advance treatment should always be respected for patient autonomy and informed consent for treatment,” concluded Kirchner and Pruett.2
