目的：（1）对MCI老年人进行为期6个月的追踪，建立人群队列，描述MCI老年 人在6个月内出现有临床意义的认知功能下降的发生率，并分析人群特征及风险因 素；（2）分别采用Logistic回归分析和机器学习算法 构建短期内MCI进展高危老年 人群风险预测模型，以AUC值为主要评价指标，选择预测效果最好的模型。 方法：于2023年3月-2023年6月，采用方便抽样法，在北京市东花市、方庄社区 卫生服务中心，招募符合纳排标准的 MCI 老年人作为研究对象，进行一般资料、 既往疾病史、感知觉、精神心理、生活行为、认知功能、日常生活活动能力的基 线数据采集，并对其进行 6个月的追踪，测评其认知功能变化。分别采用Logistic 回归分析和随机森林、极端梯度提升、轻量级梯度提升机器学习、支持向量机构 建MCI进展高危老年人群风险预测模型，以AUC值和Brier分数评价模型的区分 度和校准度。 结果：共招募符合纳排标准的MCI老年人498例，其中随访成功390例。在完成 随访的390例MCI老年人中，有100例（25.6%）发生有临床意义的认知功能下降。 Logistic 回归分析显示：嗅觉障碍[OR=1.846，95%CI=（1.065，3.200）]、抑郁 [OR=11.844，95%CI=（2.579，54.390）]、被动认知[OR=4.928，95%CI=（1.319， 18.420）]、日常生活活动能力受损[OR =1.557，95%CI=（1.044，2.321）]是 MCI 老年人6个月内发生有临床意义的认知功能下降的危险因素，文化程度{中学/中专 [OR =0.186，95%CI=（0.050，0.686）]；大专及以上[OR=0.150，95%CI=（0.035， 0.614）]}、家庭人均月收入{3000~5999 元[OR =0.233，95%CI=（0.071，0.700）]； ≥6000 元[OR =0.137，95%CI=（0.038-0.495）]}、社交参与[OR =0.304，95%CI= （0.169-0.548）]是 MCI 老年人 6 个月内发生有临床意义的认知功能下降的保护性 因素。Logistic 回归模型以及基于机器学习算法构建的随机森林、极端梯度提升、 轻量级梯度提升机器学习、支持向量机模型的AUC分别为0.78、0.66、0.65、0.66、 0.69，Brier 分数分别为 0.16、0.18、0.22、0.19、0.17。综合考虑区分度 AUC 值以 及校准度Brier分数，Logistic回归模型的效能最好。 结论：本研究显示，基于Logistic回归构建的预测模型在MCI进展高危老年人群的 识别中更具有优势，能够有效识别短期内可能出现有临床意义的认知功能下降的 MCI老年人。但该模型在进行临床运用前，仍需进行外部验证研究。

Objective:(1) To follow up the elderly with MCI for 6 months, establish a population cohort, describe the incidence of clinically significant cognitive decline in the elderly with MCI within 6 months, and analyze the population characteristics and risk factors; (2) Logistic regression analysis and machine learning algorithm were used to construct a risk prediction model for the elderly population at high risk of MCI progression in the short term, and the AUC value was used as the main evaluation index to select the model with the best prediction effect. Method：From March 2023 to June 2023, the elderly with MCI who meet the inclusion criteria were recruited as the research subjects at the Donghuashi and Fangzhuang Community Health Service Centers in Beijing by convenience sampling method, and the baseline data collection of general data, past disease history, perception, mental psychology, life behavior, cognitive function, and daily living activities were carried out, and the changes in their cognitive function were evaluated for 6 months. Logistic regression analysis, random forest, extreme gradient boosting, lightweight gradient boosting machine learning, and support vector mechanism were used to construct a risk prediction model for the high-risk elderly population with MCI progression, and the AUC value and Brier score were used to evaluate the discrimination and calibration degree of the model. Results：A total of 498 elderly patients with MCI who met the inclusion criteria were recruited, of which 390 were successfully followed-up. Of the 390 older adults with MCI who completed follow-up, 100 (25.6%) experienced clinically significant cognitive decline. Logistic regression analysis showed that olfactory impairment [OR=1.846, 95%CI=(1.065, 3.200)], depression [OR=11.844, 95%CI=(2.579, 54.390)], passive cognition [OR=4.928, 95%CI=(1.319, 18.420)], impaired activities of daily living [OR =1.557, 95%CI=(1.044, 2.321)] It was a risk factor for clinically significant cognitive decline in the elderly with MCI within 6 months, and the education level {middle school/technical secondary school [OR = 0.186, 95% CI = (0.050, 0.686)]; College degree or above [OR=0.150, 95%CI=(0.035, 0.614)]}, per capita monthly household income {3000~5999 yuan [OR =0.233, 95%CI=(0.071, 0.700)]; ≥6000 yuan [OR = 0.137, 95%CI=(0.038-0.495)]} and social participation [OR = 0.304, 95%CI=(0.169-0.548)] were protective factors for clinically significant cognitive decline within 6 months in older adults with MCI. The AUC value of the logistic regression model and the random forest, extreme gradient boosting, lightweight gradient boosting machine learning and support vector machine model constructed based on the machine learning algorithm were 0.78, 0.66, 0.65, 0.66 and 0.69, respectively, and the Brier scores were 0.16, 0.18, 0.22, 0.19 and 0.17, respectively. Considering the discrimination AUC value and the Brier score of calibration, the logistic regression model had the best performance. Conclusion：This study shows that the prediction model based on logistic regression has more advantages in the identification of elderly people at high risk of MCI progression, and can effectively identify the elderly with MCI who may have clinically significant cognitive decline in the short term. However, the model still needs to be externally validated before it can be used in clinical practice.