Scholarlink

Background <p>Osteoporosis is a common chronic condition in older adults, characterized by reduced bone mineral density and increased fracture risk. Dual-energy X-ray absorptiometry is the diagnostic gold standard, but its cost and limited availability restrict large-scale screening in primary and community care. There is a need for auxiliary diagnostic systems that use routinely collected clinical data to identify individuals at high risk of osteoporosis.</p> Methods <p>Data from the 2017–2020 National Health and Nutrition Examination Survey were analysed. Adults aged 50 years or older with valid dual-energy X-ray absorptiometry measurements at the femoral neck or lumbar spine were included. Osteoporosis was defined by a T-score of − 2.5 or lower at either site. After structured preprocessing and imputation, a penalised logistic regression method was used for feature selection. A two-level stacking ensemble learning was then constructed, combining five base learners (logistic regression, decision tree, gradient boosting, extreme gradient boosting and multilayer perceptron) with logistic regression as the meta-learner. Models were trained with stratified five-fold cross-validation in the training set, and performance was assessed in an independent test set.</p> Results <p>A total of 3,735 participants were included, of whom 390 had osteoporosis. Participants with osteoporosis were older and had lower body mass index, hip circumference and serum uric acid levels than those without osteoporosis. Feature importance analysis indicated that age, sex, body mass index and uric acid were the most influential predictors. In the test set, the stacking ensemble achieved an accuracy of 0.96, an area under the receiver operating characteristic curve of 0.95, a precision of 0.83, a recall of 0.81 and an F1-score of 0.82, outperforming all individual base models and maintaining a low false-positive rate.</p> Conclusions <p>A stacking ensemble learning based on routinely available clinical and laboratory variables provided robust discrimination of osteoporosis risk under real-world class imbalance. This approach may serve as a practical, low-cost auxiliary diagnostic tool to support targeted dual-energy X-ray absorptiometry referral in primary and community care. External and prospective validation in diverse populations is needed before widespread clinical use.</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 intelligent auxiliary diagnostic system for early osteoporosis screening using stacking ensemble learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月6日

查看全文
  • Qian Wang,
  • Jun Liu,
  • Min Zhu,
  • Hong-ying Ye,
  • Yan Shang

摘要

Background

Osteoporosis is a common chronic condition in older adults, characterized by reduced bone mineral density and increased fracture risk. Dual-energy X-ray absorptiometry is the diagnostic gold standard, but its cost and limited availability restrict large-scale screening in primary and community care. There is a need for auxiliary diagnostic systems that use routinely collected clinical data to identify individuals at high risk of osteoporosis.

Methods

Data from the 2017–2020 National Health and Nutrition Examination Survey were analysed. Adults aged 50 years or older with valid dual-energy X-ray absorptiometry measurements at the femoral neck or lumbar spine were included. Osteoporosis was defined by a T-score of − 2.5 or lower at either site. After structured preprocessing and imputation, a penalised logistic regression method was used for feature selection. A two-level stacking ensemble learning was then constructed, combining five base learners (logistic regression, decision tree, gradient boosting, extreme gradient boosting and multilayer perceptron) with logistic regression as the meta-learner. Models were trained with stratified five-fold cross-validation in the training set, and performance was assessed in an independent test set.

Results

A total of 3,735 participants were included, of whom 390 had osteoporosis. Participants with osteoporosis were older and had lower body mass index, hip circumference and serum uric acid levels than those without osteoporosis. Feature importance analysis indicated that age, sex, body mass index and uric acid were the most influential predictors. In the test set, the stacking ensemble achieved an accuracy of 0.96, an area under the receiver operating characteristic curve of 0.95, a precision of 0.83, a recall of 0.81 and an F1-score of 0.82, outperforming all individual base models and maintaining a low false-positive rate.

Conclusions

A stacking ensemble learning based on routinely available clinical and laboratory variables provided robust discrimination of osteoporosis risk under real-world class imbalance. This approach may serve as a practical, low-cost auxiliary diagnostic tool to support targeted dual-energy X-ray absorptiometry referral in primary and community care. External and prospective validation in diverse populations is needed before widespread clinical use.