Explainable machine learning for osteoporosis detection in patients with osteopenia: model development and validation
Xiuzhen Zhang1, Li Zhao2, Han Wu1
1Department of Endocrinology and Metabolism, Shenzhen People's Hospital (The First Affiliated Hospital, Southern University of Science and Technology; The Second Clinical Medical College, Jinan University), Shenzhen, Guangdong, China.
Background:
Osteopenia is a critical precursor to osteoporosis (OP), yet accurately discriminating OP from osteopenia among individuals with low bone mass remains challenging. While dual-energy X-ray absorptiometry (DXA) provides definitive diagnosis, accessibility limitations necessitate alternative screening approaches. We therefore aimed to develop an algorithm based on readily available clinical data to discriminate between OP and osteopenia in this population.
Methods:
We conducted a retrospective diagnostic study to develop a model for discriminating osteoporosis from osteopenia within a cohort of 1,203 Asian adults with low bone mass. Eleven machine learning algorithms were trained and validated for this diagnostic task (case: osteoporosis [T-score ≤ -2.5]; control: osteopenia [T-score -2.5 to -1.0]).Performance was evaluated using area under the curve (AUC). The interpretability and clinical utility of the selected model were respectively enhanced and validated by SHAP analysis, nomogram calibration, and decision curve analysis (DCA).
Findings:
The Linear Discriminant Analysis model demonstrated superior and consistent performance. It achieved a mean cross-validated AUC of 0.738 (95% CI: 0.736-0.741) and showed excellent calibration (Hosmer-Lemeshow p = 0.266). On an independent validation set, the model maintained robust performance with an AUC of 0.710 (95% CI: 0.686-0.734). Key predictors included waist-to-height ratio, body weight, serum uric acid, age, and alkaline phosphatase. DCA indicated a positive net benefit across a wide range of risk thresholds.
Interpretation:
This study developed a practical and interpretable tool for discriminating osteoporosis from osteopenia among individuals with low bone mass, using only routinely available clinical data. This approach may serve as a preliminary screening tool to identify high-risk individuals within primary care populations for further definitive testing.
