Original title: A primary care-oriented framework for osteoporosis prediction: integrating multi-algorithm feature selection and machine learning model benchmarking
The researchers built models to estimate osteoporosis risk without using bone-density measurements. They used routine health-check data and kept eight measures selected by at least four of seven feature-selection methods. They compared six types of machine-learning models. In the validation set, a random forest and an artificial neural network were the strongest at separating people with and without osteoporosis; the neural network also matched predicted risks more closely to observed outcomes.
Imagine sorting a pile of health forms to decide who should get an extra check-up. The hard part is choosing a small set of details that gives a useful warning without having the special scan result. In this paper, the forms are routine health checks, the chosen details are the eight measures, and the warning helps identify people who may need a bone scan.
DXA scans are not widely available, which makes osteoporosis harder to screen for in some primary-care settings. A model using routine health-check information could help clinicians decide who may need referral for a scan.
The abstract does not report external validation or establish how the models perform in other populations or health systems. It describes prediction from routine health-check data and presents the models as a potential aid for pre-DXA triage, not as a replacement for DXA diagnosis.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
A primary care-oriented framework for osteoporosis prediction: integrating multi-algorithm feature selection and machine learning model benchmarking — Xiaofang Guan, Xiaoqi Wang, Wenjing Zhang et al. — Frontiers in Endocrinology — doi:10.3389/fendo.2026.1917386
Abstract licence: CC BY
Explanation generated by Synthesis from the abstract — it is not the original paper.