Background: Falls among older adults are a significant challenge to global healthy aging. Identifying key factors and differences in fall risks, along with developing predictive models, is essential for differentiated and precise interventions in China's urban and rural older populations.

Methods: The data of 5,876 older adults were obtained from the China Health and Retirement Longitudinal Survey (Waves 2015 and 2018). A total of 87 baseline input variables were considered as candidate features. Predictive models for fall risk over the next 3 years among urban and rural older populations were developed using five machine learning algorithms. Logistic regression analysis was employed to identify key factors influencing falls in these populations.

Results: The fall incidence among older adults was 22.4%, with 23.2% in rural areas and 20.9% in urban areas. Common risk factors across both settings include gender, age, fall history, sleep duration, activities of daily living questionnaire scores, memory status, and chair stand test time. In rural areas, additional risks include being unmarried, having diabetes, heart disease, memory-related medication use, and living in houses built 6-20 years ago. For urban, liver disease, arthritis, physical disabilities, depressive symptoms, weak hand strength, poor relations with children, and digestive medication use are significant risk factors while living in a tidy environment is protective. Random Forest models achieved the highest AUC-ROC and sensitivity in both rural (AUC = 0.732, 95% CI: 0.69-0.78; sensitivity = 0.669) and urban (AUC = 0.734, 95% CI: 0.68-0.79; sensitivity = 0.754) areas. Decision curve analysis confirmed the model's clinical utility across a range of threshold probabilities. Key predictors included prior experience of falling, gender, and chair stand test performance in rural areas, while in urban areas, experience of falling, gender, and age were the most influential features.

Conclusion: The key factors influencing falls among older people differ between urban and rural areas, and the predictive models effectively identify high-risk populations in both settings. This facilitates targeted prevention and precise interventions, supporting healthy aging in China.

Keywords: aging; fall risk; machine learning; older people; public health; rural–urban difference.

Copyright © 2025 Lin, Liu, Cai, Zheng, Li and Hu.

背景: 老年人跌倒是对全球健康老龄化的一个重要挑战。识别关键因素和城乡老年人群中跌倒风险的差异，并开发预测模型，对于中国城市和农村老年群体的差异化精准干预至关重要。

方法: 本研究的数据来自2015年和2018年的中国健康与退休纵向调查（共涉及5,876名老年人）。总计考虑了87个基线输入变量作为候选特征。使用五种机器学习算法开发了针对城市和农村老年群体未来3年内跌倒风险的预测模型，并采用逻辑回归分析来识别这些人群中影响跌倒的关键因素。

结果: 老年人群中的跌倒发生率为22.4%，其中农村地区为23.2%，城市地区为20.9%。城乡环境中常见的风险因素包括性别、年龄、以往的跌倒历史、睡眠时长、日常生活活动问卷得分、记忆力状况和椅子站立测试时间。在农村地区，额外的风险因素还包括未婚状态、糖尿病、心脏病、与记忆相关的药物使用以及居住在6-20年前建造的房子中。对于城市地区，肝病、关节炎、身体残疾、抑郁症状、手部力量弱、子女关系不好及消化系统用药是显著风险因素，而生活在整洁环境中则具有保护作用。随机森林模型在农村（AUC = 0.732, 95% CI: 0.69-0.78; 敏感性 = 0.669）和城市（AUC = 0.734, 95% CI: 0.68-0.79; 敏感性 = 0.754）地区均取得了最高的AUC-ROC和敏感性。决策曲线分析确认了模型在不同阈值概率范围内的临床实用性。农村地区的关键预测因素包括以往的跌倒经历、性别和椅子站立测试表现，而在城市地区，跌倒经历、性别和年龄是最具影响力的特征。

结论: 影响老年人群中跌倒的关键因素在城乡之间存在差异，并且预测模型能够有效识别两地中的高危人群。这有助于实施有针对性的预防措施并进行精准干预，从而支持中国健康老龄化的目标。

关键词: 老化；跌倒风险；机器学习；老年人群；公共健康；城乡差异。