Search for a command to run...
Hepatocellular carcinoma (HCC) remains one of the leading causes of cancer-related deaths worldwide, with approximately 900,000 new cases and 830,000 deaths per year, making it the sixth most common cancer and the third most lethal cancer worldwide [1, 2]. Using both alpha-fetoprotein (AFP) measurement and ultrasound scans to screen patients in dedicated cirrhosis clinics has yielded a steady rise in the incidence of HCC but without necessarily improving the mortality outcomes [3]. Besides cirrhosis, other key risk factors for HCC include chronic viral hepatitis, haemochromatosis and, increasingly, metabolic dysfunction-associated steatotic liver disease, MASLD [2, 3]. It is, therefore, understandable that attempts continue to be made to predict HCC and identify it in its early and potentially treatable stages. Predictions have involved the analysis of demographic, aetiological, laboratory and imaging findings. Machine Learning (AI) Models have been tested, such as the Convolutional Neural Networks (CNN) and Random Survival Forests (RSF), which are used to analyse electronic health records, focusing on age, gender and comorbidities, often without using laboratory results. Risk Scores and biomarkers were also used including the GALAD Score (combining Gender, Age and three biomarkers [AFP, AFP-L3, DCP] for high accuracy); aMAP Score (uses Age, Male gender, Albumin-bilirubin and Platelets); and AAAPD-C Score (focusing on patients with hepatitis C after treatment [Age, Albumin, AFP, Platelets, Diabetes status]) [4-8]. Despite the existence of this multitude of scoring systems, in this issue of AP&T, Dr. Ying and colleagues shift the focus to the prediction of HCC after hepatitis B surface antigen seroclearance in patients with chronic hepatitis B (CHB) [9]. This is justifiable by the ongoing common nature of hepatitis B, affecting 254 million people globally, and the still rising death rates from viral hepatitis despite recent advances in prevention and management [10]. Other risk models have already been proposed for the prediction of HCC development, including PAGE-B, modified PAGE-B (mPAGE-B) for CHB and Yang's model, CAMP-B model for HBsAg seroclearance patients; but as Dr. Ying and colleagues point out, most existing models were derived from cohorts of patients with ongoing CHB infection, and their applicability to patients who have achieved HBsAg seroclearance remains uncertain. The authors found six independent risk factors capable of predicting HCC occurrence, including age ≥ 50 years, male, platelet count ≤ 150 × 109/L, albumin level ≤ 44 g/L, with cirrhosis at HBsAg seroclearance and AVT-induced HBsAg seroclearance. They concluded that their six-factor model enables risk stratification among patients achieving HBsAg seroclearance and may assist in informing surveillance strategies in clinical practice. Their data do not adjust for potentially protective factors such as the use of aspirin, and there is incomplete information on patients' baseline performance status (e.g., ECOG score) [3] or viral factors, as a higher baseline HBV DNA level (up to 106 units) may indicate a higher risk for HCC development. However, besides being simple, the proposed model was able to predict 67.9%–100% of the early stages of the Barcelona Clinic Liver Cancer (BCLC) stage [3]. This supports its relevance for surveillance strategies in clinical practice. Ali S. Taha: conceptualization, writing – review and editing, writing – original draft, investigation, methodology. The author has nothing to report. The author declares no conflicts of interest. This article is linked to Ying et al. paper. To view this article, visit https://doi.org/10.1111/apt.70637. Research data are not shared.