A Machine Learning-Based Nomogram for Predicting Overall Survival in BCLC Stage 0-B Hepatocellular Carcinoma Patients
Yao Liu1, Ke Shi1, Xiaojing Wang1
1Center of Integrative Medicine, Beijing Ditan Hospital, Capital Medical University, Beijing, 100015, People's Republic of China.
Purpose:
This study aimed to investigate the prognostic value of the hemoglobin, albumin, lymphocyte, and platelet (HALP) score in patients with Barcelona Clinic Liver Cancer (BCLC) stage 0-B hepatocellular carcinoma (HCC) and to develop a prognostic nomogram for the low-HALP subgroup.
Patients And Methods:
We retrospectively analyzed 1,281 patients with BCLC stage 0-B HCC treated at Beijing Ditan Hospital between 2015 and 2021. Using a HALP cutoff of 47, patients were divided into a low‑HALP group (n = 457) and a high‑HALP group (n = 824). The low‑HALP cohort was randomly divided into a derivation set (n = 305) and an internal validation set (validation set 1, n = 152). An additional temporal validation set (validation set 2) comprised 251 patients treated between 2022 and 2025. Candidate predictors of 5‑year overall survival (OS) were screened using random survival forest, LASSO regression, and multivariate Cox analysis. A nomogram was then built and assessed by time‑dependent ROC curves, calibration plots, and decision curve analysis (DCA). Survival differences between risk groups were compared using the Kaplan-Meier method.
Results:
A high HALP score was significantly associated with longer OS (p < 0.001). The final nomogram included gamma‑glutamyl transferase, creatinine, C‑reactive protein, and BCLC stage. The AUCs for 1‑, 3‑, and 5‑year OS were 0.80, 0.75, and 0.77 in the derivation set; 0.80, 0.83, and 0.77 in validation set 1; and 0.79, 0.73, and 0.76 in validation set 2. Compared with conventional staging systems, the nomogram showed better discrimination for 5‑year OS. Calibration and decision curve analyses confirmed good agreement and clinical usefulness. Using an optimal cutoff of 9.5, the nomogram stratified patients into high‑ and low‑risk groups with markedly different 5‑year mortality rates: 86.9% versus 46.4% in the derivation set, 90.0% versus 43.4% in validation set 1, and 86.1% versus 43.5% in validation set 2.
Conclusion:
A low HALP score defines a subgroup of BCLC stage 0‑B HCC patients with poorer survival. To our knowledge, this is the first nomogram developed specifically for low‑HALP patients. It provides a tool for individualized prognostic assessment, although external validation is still needed.
