Predicting Poor Response to Anti-Osteoporosis Therapy with Clinical and Biomarker Data

A retrospective analysis of 543 primary osteoporosis patients admitted between January 2021 and December 2024. The dataset was used to develop and validate machine learning models for predicting poor response to 12 months of standard anti-osteoporosis therapy. It was authored by Yannan Bi and published on figshare.

Use Cases

• Training classification models to predict poor treatment response based on clinical characteristics like diabetes comorbidity and fracture history.
• Evaluating the importance of novel biomarkers such as serum β-CTX and osteocalcin levels for treatment outcome prediction.
• Comparing the performance of Random Forest, Support Vector Machine, and K-Nearest Neighbors algorithms on a clinical prediction task.
• Applying model interpretation techniques like SHAP analysis to identify key predictive variables in a medical context.

Strengths

• Dataset includes 543 patient records with a defined 7:3 split for training (n=380) and validation (n=163).
• Model development process is documented, with the Random Forest model achieving an AUC of 0.856 in training and 0.825 in validation.
• Eight independent predictors were identified, integrating clinical factors and specific serum biomarker levels.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect temporal and source bias inherent to the single-center, retrospective study design.

Provenance

Source

figshare

Collection Method

Retrospective analysis of patient records.

Time Range

January 2021 to December 2024

Freshness

Last updated 2026-05-14 05:43:22; freshness should be verified.