Qin Guo,1,* Shengying Xia,2,* Yan Ding,3 Fan Liu3 

1Department of General Surgery, Wuhan Children’s Hospital, Tongji Medical College, Huazhong University of Science & Technology, Wuhan, 430016, People’s Republic of China; 2Department of Emergency and Critical Care Center, Wuhan Children’s Hospital, Tongji Medical College, Huazhong University of Science &Technology, Wuhan, 430016, People’s Republic of China; 3Department of Rheumatology and Immunology, Wuhan Children’s Hospital, Tongji Medical College, Huazhong University of Science &Technology, Wuhan, 430016, People’s Republic of China

*These authors contributed equally to this work

Correspondence: Fan Liu; Yan Ding, Department of Rheumatology and Immunology, Wuhan Children’s Hospital, Tongji Medical College, Huazhong University of Science &Technology, No. 100 hong Kong Road, Jiang’an District, Wuhan, 430016, People’s Republic of China, Tel +86 02782824022, Email LFwhetyy@outlook.com; DYwhetyy@outlook.com

Background: Henoch-Schönlein Purpura (HSP) is a common systemic vasculitis in children that often involves the gastrointestinal system (GIS). Identifying reliable predictive markers for GIS complications is crucial for early intervention and improved patient outcomes.
Objective: This study aims to identify laboratory markers predictive of GIS complications in children with HSP using a machine learning approach.
Methods: This retrospective study included children diagnosed with HSP and a control group from May 2020 to January 2024. Detailed demographic and laboratory data, including WBC count, lymphocyte count, neutrophil count, platelet count, hemoglobin, NLR, PLR, MPV, MPR, C-reactive protein, ESR, albumin, BUN, creatinine, sodium, potassium, calcium, IgA, PT, aPTT, and INR, were collected. GIS complications was classified based on clinical symptoms and diagnostic findings. Patients were categorized into groups without GIS complications, with mild GIS complications, and with severe GIS complications. We compared laboratory parameters across these groups to identify significant differences associated with GIS complications. Furthermore, a predictive model was developed by a Random Forest classifier to identify key markers and assess their ability to distinguish between patients with and without GIS complications.
Results: Significant differences were observed in several laboratory parameters between HSP patients and the control group, and between patients with and without GIS complications. Key predictive markers identified included neutrophil count, NLR, WBC count, PLR, and platelet count. The RandomForest model achieved an accuracy of 91% and an AUC of 0.90.
Conclusion: Our findings highlight the importance of specific laboratory markers in predicting GIS complications in HSP. The use of machine learning models can enhance the early identification and management of high-risk patients, potentially improving clinical outcomes.

Keywords: Henoch-Schönlein Purpura, gastrointestinal complications, laboratory markers, machine learning, random forest classifier