Editorial: Prediction of Survival in Hepatocellular Cancer—Rise and Fall of the Machines

Survival from HCC depends on tumour factors (tumour size and number of nodules), the severity of the underlying liver disease, performance status and other comorbidities. In addition, prognosis is influenced by the treatment being administered. Prognostic classification and treatment allocation decisions have traditionally been made using the Barcelona Clinic Liver Cancer staging system. Since its original description [1] in 1999, the system is now on its 9th update in 27 years with the latest update published in 2026 [2]. To a great extent, the need for regular updates reflects the changes and the expanding availability of both locoregional and systemic treatment options [3, 4].

Within this evolving landscape of HCC management, Lim et al. conduct a retrospective study with the aim of developing an improved prognostic test for HCC using machine learning methodology (LCRN Index) [5]. They include participants diagnosed with HCC between 2010 and 2022 at 4 centres in Asia and one in the USA.

The characteristics of the cohort are worth considering to frame the study. Overall, only 39% of participants with HCC had cirrhosis and of those with cirrhosis about 50% had decompensated disease. This suggests that the cohort was more skewed towards milder liver disease. For comparison, in other HCC cohorts, 69%–76% had liver cirrhosis [6, 7]. The differences in the population evaluated here may explain to some degree the marginally higher performance of the LCRN Index when compared to BCLC. It suggests that the LCRN Index may be more appropriate for cohorts with lower proportions of cirrhosis. Indeed, in the external validation cohort where the proportion of people with cirrhosis is higher, at 52.7%, the superior performance observed in the derivation and internal validation cohorts is lost. This indicates the well‐established importance of external validation in order to understand the true potential of a biomarker or prognostic tests [8]. The need for further validation is correctly pointed out by the authors as something that would be needed before clinical adoption of the tests.

One other interesting finding is that in the external validation cohort, the performance of the LCRN index that was developed using machine learning methodology performs equally well to a score that includes the same predictors but is constructed using a traditional Cox regression model. This indicates that the performance of the model may be driven primarily by the number of predictors that are included and not so much by the machine learning methodology.

Furthermore, it is worth considering that out of the variables that were tested, only the presence of type 2 diabetes, AFP and albumin levels were found to be independent predictors of survival. This observation aligns with the established view that survival in this group of patients is largely determined by tumour factors (AFP), liver disease severity (albumin) and general fitness and comorbidities (type 2 diabetes). It also raises the question of whether a simpler prediction model could have been created, rather than the one that was eventually developed that included 7seven variables, some of which were not independent predictors of survival.

Author Contributions

Michael Pavlides: writing – original draft, writing – review and editing, investigation.

Funding

The author has nothing to report.

Conflicts of Interest

The author declares no conflicts of interest.

Linked Articles

This article is linked to Lim et al. papers. To view these articles, visit https://doi.org/10.1111/apt.70797 and https://doi.org/10.1111/apt.70894.

Pavlides M., “Editorial: Prediction of Survival in Hepatocellular Cancer—Rise and Fall of the Machines,” Alimentary Pharmacology & Therapeutics 64, no. 7 (2026): 1009–1010, 10.1111/apt.70863.

Data Availability Statement

Data sharing not applicable to this article as no datasets were generated or analysed during the current study.