AI Models Predict Osteoporosis Risk in Postmenopausal Women with CKD

Introduction

This study explores the application of artificial intelligence (AI) to enhance the screening and prediction of osteoporosis (OP) risk, particularly in a complex and vulnerable population: postmenopausal women with chronic kidney disease (CKD). Given the limitations in current screening methods for such high-risk groups, the research aimed to develop and interpret an explainable AI model to identify low bone mass (LBM) or OP.

The Study in Detail

The research, titled "Artificial intelligence models for osteoporosis risk prediction in postmenopausal women with CKD: A cross-sectional analysis of NHANES 2005 to 2018," was conducted by Zhu Q, Wang T, Lou N, Yu Z, Wang Z, Guo H, and Li X. It was published in Medicine (2026 Mar 20;105(12):e48107). The authors are affiliated with the Division of Bone and Joint Surgery and Sports Medicine, Orthopedic Center, The University of Hong Kong-Shenzhen Hospital, China.

The study utilized data from the National Health and Nutrition Examination Survey (NHANES) spanning from 2005 to 2018. This nationally representative cross-sectional study provided the dataset for developing and comparing eight different AI models. For the purpose of algorithm development and comparison, complex survey weights of NHANES were not applied. Model performance was evaluated using metrics such as AUC-ROC, recall, precision, F1 score, and Brier score. To ensure interpretability, techniques like SHapley Additive exPlanations (SHAP), local interpretable model-agnostic explanations (LIME), and generalized additive models were employed to identify key features and their non-linear interactions.

The multilayer perceptron model demonstrated the most effective performance, achieving an AUC of 0.72 and a precision of 0.84. Interpretation of this model revealed that key predictors for LBM/OP included:

• Weight
• Age
• Estimated glomerular filtration rate (eGFR)
• Age when heaviest weight
• Total cholesterol (TC)
• Age at last menstrual period

A significant finding was the non-linear relationship between weight and LBM/OP, as well as complex interactions observed between weight, age, and the age at which an individual was heaviest.

Assessment

This study successfully developed and validated an interpretable AI model for screening LBM/OP in a specific high-risk demographic. A key strength is its focus on explainable AI, moving beyond traditional "black-box" approaches. This allows for a better understanding of the factors contributing to risk prediction, which is crucial for clinical acceptance and application. The use of a large, nationally representative dataset (NHANES) enhances the generalizability of the findings to the US population. However, a limitation noted by the authors is the cross-sectional nature of the NHANES data, which prevents the establishment of causal relationships. Additionally, the non-application of complex survey weights during algorithm development might affect the direct representativeness of the model for the entire NHANES population. The authors also emphasize the necessity of further external validation in prospective, independent cohorts to confirm the model's utility in real-world clinical settings.

Practical Relevance

The findings from this research have several practical implications for healthcare. For postmenopausal women with chronic kidney disease, an AI-powered tool could potentially offer a more personalized and accurate assessment of their osteoporosis risk. This could lead to earlier identification of individuals at risk, allowing for timely interventions and management strategies. The identification of specific predictors like weight, age, eGFR, and total cholesterol, along with their non-linear interactions, provides valuable insights for clinicians. It suggests that a holistic view of a patient's health profile, beyond standard bone density measurements, is critical. Understanding these complex relationships could also inform the development of more targeted preventive measures and treatment protocols, potentially improving patient outcomes and reducing the burden of osteoporosis in this vulnerable population.

Conclusion

This study demonstrates the potential of AI models to significantly improve osteoporosis risk prediction in postmenopausal women with chronic kidney disease. By identifying key predictors and their complex interactions, the research offers a valuable, interpretable tool that could enhance screening and facilitate personalized risk evaluation. While further external validation is required, this work represents a step forward in leveraging AI for precision medicine in bone health.