weave
he who
press
Clinical cure is the ideal endpoint for antiviral therapy for chronic hepatitis B (CHB). Clearance of hepatitis B surface antigen (HBsAg) is a core indicator of clinical cure for hepatitis B. Previous studies have shown that HBsAg clearance reduces the risk of poor prognosis, and the risk of hepatocellular carcinoma (HCC) and hepatic decompensation after HBsAg clearance is very low (related links 1 and 2). Predicting the risk of liver-related adverse outcomes in patients with chronic hepatitis B after HBsAg clearance and establishing appropriate monitoring strategies based on risk stratification are the focus of long-term management of chronic hepatitis B.
Recently, the Journal of Hepatology (IF 26.8) published a novel machine learning model to predict liver-related adverse outcomes in patients with HBsAg clearance of chronic hepatitis B, including 2 demographic characteristics (age and sex), 3 liver function-related factors (cirrhosis, albumin and platelet count), and 2 metabolic factors (diabetes and alcohol consumption). Compared with the existing models, the proposed model can significantly improve the accuracy of predicting the risk of liver-related adverse outcomes in patients with HBsAg clearance and chronic hepatitis B, and provide patients with more personalized monitoring strategies based on risk stratification.
Research Methods:
In this multicenter study, CHB patients who received HBsAg clearance from 6 centers in Korea from 2000 to 2022 were consecutively included. HBsAg clearance is defined as at least 2 consecutive negative HBsAg tests (more than 6 months apart) in CHB patients. Patients from both hospitals were included in the training cohort, and the remaining patients were included in the internal validation cohort. Construct independent cohorts for external validation using data from the Clinical Data Analysis and Reporting System (an electronic healthcare database managed by the Hong Kong Hospital Authority). The primary outcomes were the occurrence of liver-related adverse outcomes, including hepatocellular carcinoma (HCC), hepatic decompensation, and liver-related mortality. Three machine learning models were established by gradient booster (GBM), random surviving forest (RSF) and deep neural network (DNN) algorithms.
Patient characteristics
The Korea and Hong Kong cohorts included 2046 and 2741 patients, respectively, of whom 80% were patients with spontaneous HBsAg clearance. At a median follow-up of 55.2 months (interquartile range 30.1-92.3 months), 123 liver-related adverse outcomes (1.1%/person-year) were identified in the Korea cohort. Of these, 59 patients were diagnosed with HCC and 16 had liver-related deaths. In the Hong Kong cohort, the median follow-up was 62.3 months (quartile range 34.2 to 103.0 months), and a total of 100 liver-related adverse outcomes (0.6%/person-year) were identified. Compared with the control group, the proportion of patients with liver-related adverse outcomes was higher in the elderly, males and patients with liver cirrhosis (P <0.05).
Table 1 Baseline characteristics of patients with HBsAg clearance and slow hepatitis B were obtained
Findings:
01
Age, sex, alcohol consumption, diabetes, cirrhosis, albumin, and platelet counts were significantly associated with the risk of liver-related adverse outcomes after HBsAg clearance
Multivariate COX regression analysis showed that male (aHR = 2.07, 95% CI = 1.19–3.60, P = 0.01) and older (aHR = 1.04, 95%CI = 1.01–1.06, P = 0.003) were significantly associated with a higher risk of liver-related adverse outcomes in the training cohort. cirrhosis (aHR = 2.60, 95%CI = 1.57–4.32, P < 0.001), significant alcohol intake (aHR = 2.86, 95%CI = 1.48–5.53, P = 0.002), low serum albumin (aHR = 0.34, 95%CI = 0.22–0.51, P < 0.001), and platelet count (aHR = 0.995, 95%CI = 0.991–0.999, P = 0.01) It is also an independent risk factor for liver-related adverse outcomes. In the training cohort, the 7 factors with the highest prediction accuracy (age, gender, alcohol consumption, diabetes, cirrhosis, albumin, and platelet count) were selected as the final input variables.
02
The new machine learning prediction model performed better than the previous HCC risk prediction model
The GBM model with the best performance in the training queue was selected as the PLAN-B-CURE model. In the training cohort, PLAN-B-CURE was significantly better than the Cox-based scoring model and previous models in the c index (0.82 vs. 0.63 – 0.72, overall P < 0.001), AUROC (0.86 vs. 0.62 – 0.75, overall P < 0.01) and AUPRC (0.53 vs. 0.13 – 0.29, overall P < 0.01). The sensitivity, specificity, and F1 scores of PLAN-B-CURE were 0.78, 0.81, and 0.45, respectively. Internal and external validation queues have similar results.
Patients were divided into four risk groups (i.e., lowest, low, intermediate, and high-risk groups) according to the risk predicted by PLAN-B-CURE, and the lowest-risk and low-risk groups showed a comparable risk of liver-related adverse outcomes (overall P > 0.05). Therefore, patients in the lowest-risk group were pooled into the low-risk group, and the risk of liver-related adverse outcomes was significantly different among the three risk groups (i.e., low, intermediate, and high-risk groups) (overall P < 0.001). In the training cohort, the incidence of 5-year liver-related adverse outcomes in the low-risk group (1.4%) was significantly lower than that in the medium-risk group (6.4%) and high-risk group (25.2%) (P < 0.001).
Liver Linjun has something to say
At present, there are a variety of prediction models for the risk of chronic hepatitis B-related liver cancer at home and abroad, and Gan Linjun has sorted out a number of published prediction models for liver cancer (related links). Constructed using 7 parameters commonly used in clinical practice, this novel machine learning predictive model demonstrated reliable discrimination and calibration, with primary results reproducible in independent validation cohorts and demonstrating superiority over previous models based on conventional statistical methods. Previous models were mostly established in patients with HBsAg-positive CHB, and this model was established in patients with HBsAg clearance. Combined with the results of previous studies and this study, the incidence of long-term adverse liver outcomes in patients with chronic hepatitis B who have HBsAg clearance is extremely low, at about 1%.
Previous studies have shown that patients with chronic hepatitis B who have achieved HBsAg clearance with antiviral therapy have a significantly lower risk of developing HCC or are significantly lower than those with spontaneous clearance (related link), with a higher probability of HBsAg clearance α interferon therapy and a higher proportion of anti-HBs (related link). Therefore, patients with chronic hepatitis B should obtain HBsAg clearance faster and better in a more effective way, and do a good job in monitoring and lifestyle management after HBsAg clearance. The continuous improvement and perfection of the liver cancer prediction model has improved the early diagnosis ability of liver cancer, in order to provide early intervention for people at high risk of chronic hepatitis B-related liver cancer and effectively reduce the risk of liver cancer.
Bibliography:
Hur MH, Yip TC, Kim SU, et al. A machine learning model to predict liver-related outcomes after the functional cure of chronic hepatitis B. J Hepatol. 2024.
Source: Yulu Liver Lin