骨质疏松性骨折风险预测模型的系统评述： 算法比较、临床适用性与发展路径

Systematic Review of Osteoporotic Fracture Risk Prediction Models: Algorithm Comparison, Clinical Applicability, and Development Pathways

摘要: 骨质疏松性骨折是骨质疏松症的严重后果，早期识别高危人群并实施个性化干预是降低其发生风险的关键，因此开发高效、精准的风险预测模型具有重要临床意义。本文系统回顾了近10年国内外骨质疏松性骨折风险预测模型的研究进展。当前模型构建主要围绕年龄、骨密度、跌倒史等核心风险因素展开，并在特定人群中纳入共病、用药等复杂变量。在方法学上，研究从传统的Logistic回归、COX比例风险模型，发展到集成学习、深度学习等多种机器学习算法，后者在捕捉复杂非线性关系和高维数据方面展现出优势。然而，该领域仍面临模型可解释性不足、外部验证缺乏、数据标准化程度低等关键挑战，制约其临床实用化。未来，通过深度融合多模态数据、发展动态预测能力，并加强模型的验证与评估，是实现精准预防和临床转化的关键方向。

Abstract: Osteoporotic fractures represent a severe consequence of osteoporosis, making the early identification of high-risk individuals and implementation of personalized interventions crucial for reducing their incidence. This underscores the clinical significance of developing efficient and accurate risk prediction models. This review synthesizes recent advances over the past decade in risk prediction models for osteoporotic fractures. Current models primarily incorporate core risk factors such as age, bone mineral density, and fall history, while also integrating more complex variables like comorbidities and medication use in specific populations. Methodologically, research has evolved from traditional logistic regression and COX proportional hazards models to various machine learning approaches, including ensemble learning and deep learning, which demonstrate advantages in capturing complex nonlinear relationships and high-dimensional data. Nevertheless, the field still faces key challenges—such as limited model interpretability, insufficient external validation, and low data standardization—that hinder clinical translation. Moving forward, key directions for achieving precise prevention and clinical application include deeper integration of multimodal data, development of dynamic prediction capabilities, and enhanced model validation and evaluation.