Establishment of a pediatric diagnostic model for McCune-Albright syndrome based on bone metabolism indicators and machine learning
10.3760/cma.j.cn311282-20250110-00017
- VernacularTitle:基于骨代谢指标与机器学习建立儿童MAS诊断模型
- Author:
Jie LU
1
;
Ni ZHEN
;
Wenli LU
;
Congcong XIA
;
Yunzhe WU
;
Jian WEI
Author Information
1. 上海交通大学医学院附属瑞金医院检验科，上海 200025
- Publication Type:Journal Article
- Keywords:
McCune-Albright syndrome;
Diagnostic model;
Bone metabolic markers;
Children
- From:
Chinese Journal of Endocrinology and Metabolism
2025;41(10):823-829
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To develop a multi-parameter diagnostic model for pediatric McCune-Albright syndrome(MAS) using machine learning techniques based on laboratory data from MAS patients, with the goal of providing a rapid and reliable auxiliary diagnostic tool for clinical practice.Methods:In this retrospective study, 232 children diagnosed with MAS at the Department of Pediatrics, Ruijin Hospital Affiliated to Shanghai Jiao Tong University School of Medicine from March 2023 to November 2024 were enrolled as the positive group. After removing duplicate or missing data, 119 cases were finally selected for statistical analysis as the positive group. Meanwhile, 113 children with normal physical examinations during the same period were selected as the control group. The clinical manifestations of the classic " triad" in the positive group were documented. Fasting serum samples were obtained from both groups at 8: 00 AM for laboratory testing, including bone metabolism-related and hormone-related indicators, which served as candidate features. Baseline descriptive analysis was conducted on the hormone-related indicators. For the bone metabolism indicators, six machine learning models—support vector machine(SVM), XGBoost, decision tree, random forest, Logistic regression, and K-nearest neighbor(KNN)—were constructed using R software. XGBoost subgroup analysis was performed based on the triad symptoms. The contribution of individual features to model predictions was visualized using SHAP diagrams. Results:SHAP visualization indicated that age, serum phosphorus, osteocalcin, and β-C-terminal cross-linked telopeptide of type Ⅰ collagen had the greatest average impact on model predictions. Among the six models, the SVM model achieved the highest diagnostic performance, with a sensitivity of 0.742 9, a specificity of 0.909 1, and an area under the curve (AUC) of 0.917.Conclusion:This study demonstrates that machine learning models, based on data from the positive patients and normal controls, can effectively distinguish MAS patients from healthy controls. The diagnostic model developed offers clinicians a valuable tool for early detection of MAS in children, contributing to earlier diagnosis, timely intervention, and improved clinical management.
