# Bridging the gap between hepatocellular carcinoma management guidelines and personalised medicine: a Bayesian network study

## TL;DR

This study uses Bayesian networks to help personalize hepatocellular carcinoma treatment decisions by bridging the gap between standard guidelines and individual patient data.

## Contribution

The paper introduces a Bayesian network model to evaluate treatment outcomes for hepatocellular carcinoma patients outside standard guidelines.

## Key findings

- The Bayesian network model was used to assess treatment effects for hepatocellular carcinoma patients not treated in compliance with BCLC guidelines.
- Detailed scenarios for ten cases showed differences in survival time based on treatment compliance and type.
- The model can serve as an AI-based clinical decision support system for patient stratification.

## Abstract

There are numerous treatment options available for patients with confirmed hepatocellular carcinoma (HCC). Guidelines such as Barcelona Clinic Liver Cancer (BCLC) support treatment decisions by way of a flow diagram that is organized around groups of patients. Though such guidelines continue to make a major contribution to standardization of treatment, in clinical reality, cases are often more nuanced than is captured in any flow diagram, even one as comprehensive as BCLC. A fundamental challenge for a clinician is to combine such a population-wide guideline with specific information about the individual patient. Bayesian networks (BNs) offer a way to “bridge this gap” and combine standardized care and precision medicine. They do this by enabling answers to detailed “what-if” questions from the clinician.

We use real-world data of HCC patients who received treatments between 2019 and 2020 to construct a BN to assess the potential treatment effect for cases that were
not
treated in compliance with BCLC.

We report detailed scenarios for ten randomly selected cases and summarise the difference in survival time for each scenario. For each case, the counterfactual treatment scenarios are made based on whether or not the case is in compliance with BCLC guidelines, the type of treatment received and the waiting time to receive treatment.

We consider two cases with similar clinical characteristics (but received different treatments) and discuss whether or not they are treated in compliance to the guidelines resulting in better outcomes than the actual clinical decision. We include a detailed discussion about the assumptions made in constructing the BN and we highlight why such a BN can serve as an AI-based clinical decision support system particularly when there is need for further patient stratification.