目的：

（1）分析中国60岁及以上老年人的失能现状及其影响因素，重点分析老年人内在能力与失能之间的相关性；

（2）构建并验证基于老年人内在能力水平的失能风险预测模型，为开展失能高风险老年人群筛查提供科学依据。

方法：

本研究基于中国健康与养老追踪调查（China Health and Retirement Longitudinal Study, CHARLS）数据开展两部分研究。基础性日常生活活动能力和工具性日常生活活动能力的12项活动中，任何一项存在困难或无法完成即定义为失能。内在能力从运动、感官、活力、认知和心理五个维度进行综合评估。

（1）利用CHARLS 2015年横断面数据，系统分析老年人失能现状及影响因素。通过多因素Logistic回归（LR）模型，分析内在能力与失能的相关性，计算调整后的比值比及其95%置信区间。

（2）将CHARLS 2015-2018年纵向数据按7:3随机划分为训练集和测试集，分别用于失能风险预测模型的构建与内部验证；采用研究对象无交叉的CHARLS 2011-2013数据进行外部验证。通过LASSO（Least Absolute Shrinkage and Selection Operator）回归筛选预测因子，采用LR、分类回归决策树（Classification And Regression Decision Tree, CART）、随机森林（Random Forest, RF）和人工神经网络（Artificial Neural Network, ANN）4种方法构建失能风险预测模型，评估模型的区分度、校准度和临床实用性，并采用列线图、特征重要性排序等方法进行模型展示。

结果：

（1）在横断面研究中，共纳入4843名60岁及以上老年人，中位年龄为66.0岁（P₂₅=62.0, P₇₅=71.0）。共有1720名老年人存在失能，失能率为35.5%。多因素LR结果显示，内在能力受损（OR=1.786, 95%CI: 1.663-1.918）、年龄增长（OR=1.039, 95%CI: 1.027-1.051）、女性（OR=1.542, 95%CI: 1.344-1.770）、居住乡村（OR=1.365, 95%CI: 1.185-1.572）、跌倒史（OR=1.587, 95%CI: 1.342-1.877）和腰围增大（OR=1.015, 95%CI: 1.008-1.021）是老年人失能的重要危险因素，而较高的教育水平（OR=0.615, 95%CI: 0.471-0.805）、更高的自我健康评价（不好、一般、好、很好的OR分别为0.589、0.296、0.192、0.192, P均＜0.005）、充足的睡眠（OR=0.939, 95%CI: 0.906-0.973）则是保护因素。

（2）在风险预测模型研究中，共纳入2895名老年人，包括训练集（n=1893）、测试集（n=812）和外部验证集（n=190）。三个数据集的失能发生率存在显著性差异，验证集的失能率高于训练集和测试集（38.4% vs 24.5% vs 24.0%, P<0.001）。经LASSO回归筛选出内在能力受损评分、性别、教育水平、居住地、自评健康、慢病状态、跌倒史、饮酒、收缩压、握力减退、肺功能、平均红细胞体积、高密度脂蛋白、尿酸等15个预测因子，构建了5个预测模型（包括2个LR模型）。其中，RF模型在训练集中的表现最佳（AUC=0.871, Brier评分=0.124），但在测试集（AUC=0.685, Brier评分=0.172）和验证集（AUC=0.707, Brier评分=0.215）中性能有所下降。全变量模型LR1和精简模型LR2均表现出良好的稳定性，且预测性能相当。

结论：

（1）失能是老年人面临的常见健康问题，内在能力受损对失能风险有显著影响。

（2）以内在能力为关键预测变量的LR模型和RF模型，在预测老年人失能风险方面具有较大潜力。未来研究可考虑拓展数据来源，并纳入更多具有重要影响的预测变量，进一步提升模型的整体性能。

Objective：

(1) To analyze the prevalence and influencing factors of disability among Chinese older adults aged 60 and above, with a focus on the association between intrinsic capacity and disability.

(2) To develop and validate risk prediction models for disability based on intrinsic capacity levels, providing a scientific basis for screening the older adults at a high risk of disability.

Methods：

This study consisted of two parts based on data from the China Health and Retirement Longitudinal Study (CHARLS). Disability was defined as experiencing difficulty or being unable to perform any of the 12 activities of daily living (ADL), which include 6 basic ADL and 6 instrumental ADL. Intrinsic capacity was assessed comprehensively across five dimensions: locomotion, sensory, vitality, cognition, and psychological.

(1) Using the CHARLS 2015 data, the prevalence and influencing factors of disability among older adults were systematically analyzed. Multivariable logistic regression (LR) models were employed to examine the association between intrinsic capacity and disability, with adjusted odds ratios (OR) and 95% confidence intervals (CI) calculated.

(2) The CHARLS 2015-2018 longitudinal data were randomly divided into a training set (70%) for model development and a test set (30%) for internal validation. External validation was conducted using completely independent CHARLS 2011-2013 data, ensuring no overlap with the model development cohort. Predictors were selected using LASSO (Least Absolute Shrinkage and Selection Operator) regression. LR and three machine learning algorithms, including Classification and Regression Decision Tree (CART), Random Forest (RF), and Artificial Neural Network (ANN), were employed to develop the risk prediction models. Model performance was evaluated in terms of discrimination, calibration, and clinical utility. Visualization tools, such as nomograms and feature importance rankings, were employed to present the models.

Results：

(1) Cross-sectional study: A total of 4,843 adults aged 60 years and older (median age 66.0, interquartile range [IQR]: 62.0–71.0) were included, of whom 1,720 (35.5%) had a disability. Multivariable logistic regression analysis revealed that impaired intrinsic capacity (OR=1.786, 95% CI: 1.663–1.918), older age (OR=1.039, 95% CI: 1.027–1.051), female sex (OR=1.542, 95% CI: 1.344–1.770), rural residence (OR=1.365, 95% CI: 1.185–1.572), history of falls (OR=1.587, 95% CI: 1.342–1.877), and larger waist circumference (OR=1.015, 95% CI: 1.008–1.021) were significant risk factors for disability. Protective factors included higher education (OR=0.615, 95% CI: 0.471–0.805), better self-rated health (ORs for "poor", "fair", "good", and "very good": 0.589, 0.296, 0.192, 0.192, respectively; all P < 0.005), and adequate sleep duration (OR = 0.939, 95% CI: 0.906–0.973).

(2) Prediction Model Study: Among 2,895 older adults (training set n=1,893, test set n=812, external validation set n=190), the prevalence of disability varied significantly across the datasets (validation set: 38.4% vs. training/test sets: 24.5%/24.0%, P<0.001). LASSO regression identified 15 predictors, including intrinsic capacity impairment score age, sex, education, residence, self-rated health, chronic conditions, fall history, alcohol consumption, systolic blood pressure, handgrip strength weakness, lung function, mean corpuscular volume, HDL cholesterol, and uric acid levels. Five models were developed, including two LR models. The RF model exhibited the best performance in the training set (AUC=0.871, Brier score=0.124) but demonstrated reduced performance in the test (AUC=0.685, Brier score=0.172) and validation sets (AUC=0.707, Brier score=0.215). Both the full-variable (LR1) and simplified (LR2) LR models showed stable and comparable predictive performance.

Conclusion：

(1) Disability is a prevalent health challenge among older adults, with impaired intrinsic capacity significantly increasing the risk.

(2) LR and RF models, especially those that incorporate intrinsic capacity as a key predictor, demonstrate promise in predicting disability risk. Future studies should broaden data sources and include additional relevant predictors to further improve model performance.