Gut microbiome-based machine learning for diagnostic prediction of liver fibrosis and cirrhosis: a systematic review and meta-analysis.

BACKGROUND

Invasive detection methods such as liver biopsy are currently the gold standard for diagnosing liver cirrhosis and can be used to determine the degree of liver fibrosis and cirrhosis. In contrast, non-invasive diagnostic methods, such as ultrasonography, elastography, and clinical prediction scores, can prevent patients from invasiveness-related discomfort and risks and are often chosen as alternative or supplementary diagnostic methods for liver fibrosis or cirrhosis. However, these non-invasive methods cannot specify the pathological grading and early diagnosis of the lesions. Recent studies have revealed that gut microbiome-based machine learning can be utilized as a non-invasive diagnostic technique for liver cirrhosis or fibrosis, but there is no evidence-based support. Therefore, this study conducted a systematic review and meta-analysis for the first time to investigate the accuracy of machine learning based on the gut microbiota in the prediction of liver fibrosis and cirrhosis.

METHODS

A comprehensive and systematic search of publications published before April 2th, 2023 in PubMed, Cochrane Library, Embase, and Web of Science was conducted for relevant studies on the application of gut microbiome-based metagenomic sequencing modeling technology to the diagnostic prediction of liver cirrhosis or fibrosis. A bivariate mixed-effects model and Stata software 15.0 were adopted for the meta-analysis.

RESULTS

Ten studies were included in the present study, involving 11 prediction trials and 838 participants, 403 of whom were fibrotic and cirrhotic patients. Meta-analysis showed the pooled sensitivity (SEN) = 0.81 [0.75, 0.85], specificity (SEP) = 0.85 [0.77, 0.91], positive likelihood ratio (PLR) = 5.5 [3.6, 8.7], negative likelihood ratio (NLR) = 0.23 [0.18, 0.29], diagnostic odds ratio (DOR) = 24 [14, 41], and area under curve (AUC) = 0.86 [0.83-0.89]. The results demonstrated that machine learning methods had excellent potential to analyze gut microbiome data and could effectively predict liver cirrhosis or fibrosis. Machine learning provides a powerful tool for non-invasive prediction and diagnosis of liver cirrhosis or liver fibrosis, with broad clinical application prospects. However, these results need to be interpreted with caution due to limited clinical data.

CONCLUSION

Gut microbiome-based machine learning can be utilized as a practical, non-invasive technique for the diagnostic prediction of liver cirrhosis or fibrosis. However, most of the included studies applied the random forest algorithm in modeling, so a diversified prediction system based on microorganisms is needed to improve the non-invasive detection of liver cirrhosis or fibrosis.