Conventional and machine learning-based risk scores for patients with early-stage hepatocellular carcinoma

Abstract

Background/aims: The performance of machine learning (ML) in predicting the outcomes of patients with hepatocellular carcinoma (HCC) remains uncertain. We aimed to develop risk scores using conventional methods and ML to categorize early-stage HCC patients into distinct prognostic groups.

Methods: The study retrospectively enrolled 1,411 consecutive treatment-naïve patients with the Barcelona Clinic Liver Cancer (BCLC) stage 0 to A HCC from 2012 to 2021. The patients were randomly divided into a training cohort (n=988) and validation cohort (n=423). Two risk scores (CATS-IF and CATS-INF) were developed to predict overall survival (OS) in the training cohort using the conventional methods (Cox proportional hazards model) and ML-based methods (LASSO Cox regression), respectively. They were then validated and compared in the validation cohort.

Results: In the training cohort, factors for the CATS-IF score were selected by the conventional method, including age, curative treatment, single large HCC, serum creatinine and alpha-fetoprotein levels, fibrosis-4 score, lymphocyte-tomonocyte ratio, and albumin-bilirubin grade. The CATS-INF score, determined by ML-based methods, included the above factors and two additional ones (aspartate aminotransferase and prognostic nutritional index). In the validation cohort, both CATS-IF score and CATS-INF score outperformed other modern prognostic scores in predicting OS, with the CATSINF score having the lowest Akaike information criterion value. A calibration plot exhibited good correlation between predicted and observed outcomes for both scores.

Conclusion: Both the conventional Cox-based CATS-IF score and ML-based CATS-INF score effectively stratified patients with early-stage HCC into distinct prognostic groups, with the CATS-INF score showing slightly superior performance.

Keywords: Fibrosis; Hepatocellular carcinoma; Inflammation; Machine learning; Prognosis.

Figure 1.

Study flow chart. HCC, hepatocellular carcinoma; TPEVGH, Taipei Veterans General Hospital; BCLC, Barcelona Clinic Liver Cancer classification.

Figure 2.

(A) LASSO coefficient profiles of the 17 variants (1: age, 2: sex, 3: curative treatment or not, 4: multiple tumors or not, 5: size >3, 6: single large HCC, 7: platelet<100,000, 8: albumin<3.5, 9: creatinine>1.2, 10: bilirubin>1, 11: ALT>40, 12: AST>45, 13: FIB-4>3.25, 14: LMR<3.62 15: PNI<45, 16: ALBI grade 2 or 3, 17: AFP>20). (B) 10 risk factors selected using LASSO Cox regression analysis. The two vertical dotted lines were drawn at the optimal scores according to minimum criteria and 1-s.e. LASSO, least absolute shrinkage and selection operator; HCC, hepatocellular carcinoma; ALT, alanine aminotransferase; AST, aspartate aminotransferase; FIB-4, fibrosis-4 index; LMR, lymphocyte-to-monocyte ratio; PNI, prognostic nutritional index; ALBI, albumin-bilirubin; AFP, alpha fetoprotein.

Figure 3.

(A) Kaplan–Meier survival analysis of training cohort according to the conventional Cox-based CATS-IF score (low risk: 0–102, intermediate risk: 103–211, high risk: 212+). (B) Kaplan–Meier survival analysis of training cohort according to the ML-based CATS-INF score (low risk: 0–114, intermediate risk: 115–223, high risk: 224+). (C) Kaplan–Meier survival analysis of the validation cohort according to the conventional Cox-based CATS-IF score (low risk: 0–102, intermediate risk: 103–211, high risk: 212+). (D) Kaplan–Meier survival analysis of the validation cohort according to the ML-based CATS-INF score (low risk: 0–102, intermediate risk: 103–211, high risk: 212+). ML, machine learning.

Figure 4.

(A) Time-dependent ROC curve of Cox-based CATS-IF score. (B) Time-dependent ROC curve of ML-based CATS-INF score. (C) Consecutive comparison of time-dependent area under the ROC curve (AUROC) of the Cox-based CATS-IF score and ML-based CATS-INF score. Both scores showed good predictability and were equally competitive, while ML-based CATS-INF score had slightly greater AUC in the long term (≥2 years). ML, machine learning.