Objectives
To develop a machine learning algorithm with prognosis data to identify different clinical phenotypes of biliary atresia (BA) and provide instructions for choosing treatment schemes.
Methods
Six hundred thirty-nine cases of type III BA were retrospectively collected from the Children’s Hospital of Fudan University from Jan 1st, 2017 to Dec 1st, 2019 as a training dataset, and a survival-based forward clustering method, which can also be used to predict the subtype of a new patient was developed to identify BA subtypes.
Results
A total of 2 clusters were identified (cluster 1 = 324 and cluster 2 = 315), where cluster 2 had a lower 2 y native liver survival post-Kasai rate. The infant patients in cluster 2 have higher weight, liver, and spleen volume, wider portal vein width, and older operative age; worse coagulation and liver function results; higher grade of liver fibrosis and detection rate of hepatic portal fibrous mass, and higher recent infection detection rate of herpes simplex virus type I. With the proposed prognostic classification system, the authors predicted the subtypes of the 187 cases of type III BA in a testing dataset collected from the whole year of 2020. The p-value computed from the log-rank testing for the Kaplan–Meier survival curves of the predicted two testing groups was 0.0113.
Conclusions
This classification system would be a convenient tool to choose appropriate treatment and accelerate the choice-making between clinicians and infant patients.
