Predicting those individuals with primary biliary cholangitis (PBC) who progress to cirrhosis can be assisted with the use of machine learning–based models of gut microbiota, according to findings from an observational, prospective study conducted in China and published in the journal Microbes and Infection.
The chronic immune-mediated disorder PBC is reported mainly among middle-aged females. Patients with the disease exhibit inflammation of intrahepatic bile ducts, which leads to the development of cholestasis. Although epidemiologic information on population-based PBC in China is scarce, the estimated prevalence of the condition is 20.5 cases per 100,000 individuals.
Currently, ursodeoxycholic acid (UDCA) is the first-line therapy for patients with PBC. Treatment with UDCA is associated with delayed histologic progression of the disorder and prolonged survival among patients without undergoing liver transplantation. If left untreated, however, PBC eventually progresses to cirrhosis and its related complications.
Thus, the need exists to monitor the evolution of PBC in a timely fashion, in order to prevent the development of PBC to the greatest extent possible. Recognizing that PBC is linked closely to the gut microbiota, the researchers sought to explore functional- and metabolic-related changes in the gut microbiota of patients with PBC that had progressed to cirrhosis.
Read more about PBC pathophysiology
DNA was extracted from fecal samples, and the quality of the DNA was analyzed. The V3-V4 region of the bacterial 16S rRNA gene was sequenced via use of polymerase chain reaction primers 515F and 806R, in order to screen for differences among individuals with PBC who progressed to cirrhosis. The data derived from these analyses were grouped into training and verification sets. The investigators utilized seven different machine learning models to extract the features of intestinal flora that impact PBC cirrhosis:
- Lasso Regression
- K-Nearest Neighbor
- Support Vector Machine
- Random Forest
- Naïve Bayes
- Decision Tree
- Linear Discriminant Analysis
Among all study participants, accuracy, F1 score, precision, and recall were compared using the various machine learning models, along with screening for the dominant intestinal flora that were linked to the development of PBC cirrhosis. Noninvasive markers of PBC cirrhosis were evaluated as well.
The current study was carried out at the Capital Medical University Affiliated Beijing You’an Hospital, between September 2019 and September 2022. The participants in the study comprised 105 patients with PBC, who were divided into the PBC_cirrhosis_yes group (n=31) and the PBC_cirrhosis_no group (n=74), along with 26 healthy controls. All participants with PBC were treated with UDCA 13 to 15 mg/kg/day for 12 months.
Results of the study showed that among participants with PBC cirrhosis, decreased diversity and richness of the gut microbiota were observed, together with alterations in the composition of gut microbiota in these patients. The abundance of Faecalibacterium and Gemmiger bacteria declined significantly, whereas the abundance of Streptococcus and Veillonella bacteria rose significantly.
Based on machine learning methods, Streptococcus and Gemmiger bacteriawere recognized as the predominant gut microbiota among patients with PBC and cirrhosis, and served as noninvasive biomarkers (area under the curve, 0.902). Thus, these two bacteria might be able to serve as indicators for predicting PBC cirrhosis severity.
Among patients with PBC cirrhosis, “gut microbiota composition and function have significantly changed,” the authors noted. “Streptococcus and Gemmiger may become a [noninvasive] biomarker for predicting the progression of PBC . . . to cirrhosis.”
