Qikuan He

Background: The impact of gut microbiome on hepatitis B virus (HBV)‐related hepatocellular carcinoma (HCC) is unclear. We aimed to evaluate the potential correlation between gut microbiome and HBV‐related HCC and introduced novel machine learning (ML) signatures based on gut microbe to predict the risk of HCC.Materials and Methods: A total of 640 patients with chronic liver diseases or HCC were prospectively recruited between 2019 and 2022. Fecal samples were collected and subjected to 16S rRNA gene sequencing. Univariate and multivariate logistic regression was applied to identify risk characteristics. Several ML methods were employed to construct gut microbe‐based models and the predictive performance was evaluated.Results: A total of 571 patients were involved in the study, including 374 patients with HCC and 197 patients with chronic liver diseases. After the propensity score matching method, 147 pairs of participants were enrolled in the analysis. Bacteroidia and Bacteroidales were demonstrated to exert mediating effects between HBV and HCC, and the moderating effects varied across Bacilli, Lactobacillales, Erysipelotrichaceae, Actinomyces, and Roseburia. HBV, alpha‐fetoprotein, alanine transaminase, triglyceride, and Child‐Pugh were identified as independent risk factors for HCC occurrence. Seven ML‐based HBV‐gut microbe models were established to predict HCC, with AUCs ranging from 0.821 to 0.898 in the training set and 0.813–0.885 in the validation set. Furthermore, the merged clinical‐HBV‐gut microbe models exhibited a comparable performance to HBV‐gut microbe models.Conclusions: Gut microbes are important factors between HBV and HCC through its potential mediating and moderating effects, which can be used as valuable biomarkers for the pathogenesis of HBV‐related HCC.

Kaplan–Meier curves analysis between responders and nonresponders with unresectable HCC treated with lenvatinib monotherapy. (A) Kaplan–Meier curves of overall survival between responders and nonresponders; (B) Kaplan–Meier curves of progression-free survival between responders and nonresponders. Abbreviations: HCC, hepatocellular carcinoma; HR, hazard ratio.

Comparison of predictive value of 10 machine learning algorithms for predicting the response to lenvatinib in the training and external validation cohorts.

Baseline characteristics between two different radiomics-based subtypes for unresectable HCC treated with lenvatinib monotherapy.

The STARD/TRIPOD statement checklists.

Representative segmentation of ROIs and primary CECT images of HCC patients that exhibited PR and PD. (A) Primary image and ROI (red region) marked image from a patient with PR; (B) Primary image and ROI (blue region) marked image from a patient with PD; (C) pretreatment and post-treatment CECT images from a patient who shows PR; (D) pretreatment and post-treatment CECT images from a patient who shows PD. Abbreviations: HCC, hepatocellular carcinoma; ROI, region of interest; CECT, contrast enhanced computed tomography; PR, partial response; PD, progressive disease.

Essential scripts for each model and the specific parameters of the models.