Scholarlink

Objective <p>This study aimed to develop and validate an interpretable machine learning model using readily available biochemical indicators to predict the 1-year risk of osteoporosis, thereby offering an efficient and accessible alternative to tools like FRAX and resource-intensive methods like radiomics.</p> Methods <p>A prospective cohort study was conducted using electronic medical records from a hospital in Wuhan, China (Jan 2020-Jan 2025). Patients with at least two DXA scans and blood tests were included. After data cleaning and handling missing values, 40 initial features were analyzed. Feature selection involved LASSO and univariate regression. Eleven machine learning models were developed and compared, with XGBoost ultimately selected. Model performance was evaluated using AUC, sensitivity, specificity, etc., on internal and external validation sets. SHAP analysis provided model interpretability. The final model was deployed as an online tool.</p> Results <p>The final model utilized only 17 routinely available features. The XGBoost model demonstrated superior performance, achieving an AUC of 0.966 for predicting 1-year spine osteoporosis in the derivation set. This high performance was consistently replicated in the external validation set (AUC = 0.969). The model also showed robust predictive power for left and right hip osteoporosis. SHAP analysis confirmed the strong influence of age and gender, among other features, and provided both global and local interpretations.</p> Conclusion <p>This study successfully developed a highly accurate, interpretable, and clinically practical machine learning model for 1-year osteoporosis prediction using only 17 routine clinical features. Its performance surpasses many previous models and is comparable to complex radiomics-based approaches, while being far more accessible. The online tool facilitates clinical application, offering a valuable and efficient strategy for early osteoporosis screening and risk assessment.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An explainable machine learning model for predicting one-year osteoporosis risk: development and validation in a prospective cohort

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月1日

查看全文
  • Jinyi Wu,
  • Dan Yu,
  • Dan Huang,
  • Chunyan Yin,
  • Lingli Li

摘要

Objective

This study aimed to develop and validate an interpretable machine learning model using readily available biochemical indicators to predict the 1-year risk of osteoporosis, thereby offering an efficient and accessible alternative to tools like FRAX and resource-intensive methods like radiomics.

Methods

A prospective cohort study was conducted using electronic medical records from a hospital in Wuhan, China (Jan 2020-Jan 2025). Patients with at least two DXA scans and blood tests were included. After data cleaning and handling missing values, 40 initial features were analyzed. Feature selection involved LASSO and univariate regression. Eleven machine learning models were developed and compared, with XGBoost ultimately selected. Model performance was evaluated using AUC, sensitivity, specificity, etc., on internal and external validation sets. SHAP analysis provided model interpretability. The final model was deployed as an online tool.

Results

The final model utilized only 17 routinely available features. The XGBoost model demonstrated superior performance, achieving an AUC of 0.966 for predicting 1-year spine osteoporosis in the derivation set. This high performance was consistently replicated in the external validation set (AUC = 0.969). The model also showed robust predictive power for left and right hip osteoporosis. SHAP analysis confirmed the strong influence of age and gender, among other features, and provided both global and local interpretations.

Conclusion

This study successfully developed a highly accurate, interpretable, and clinically practical machine learning model for 1-year osteoporosis prediction using only 17 routine clinical features. Its performance surpasses many previous models and is comparable to complex radiomics-based approaches, while being far more accessible. The online tool facilitates clinical application, offering a valuable and efficient strategy for early osteoporosis screening and risk assessment.