Abstract: Background　As China enters an aging society, the incidence of Alzheimer's Disease (AD) is increasing annually. Concurrently, these patients are susceptible to developing Community-Acquired Pneumonia (CAP), a complication that ultimately leads to adverse outcomes. Therefore, establishing an accurate risk prediction model can assist healthcare professionals in early identification and intervention for high-risk patients, thereby alleviating familial and societal burdens.Objective　To develop a predictive model for the risk of adverse prognosis in AD patients with CAP, analyze the high-risk factors for poor outcomes in this patient population, and provide a basis for formulating targeted treatment measures. Methods　Data retrieval was conducted using the dementia database platform established by Hubei Provincial Hospital of Traditional Chinese Medicine. Hospitalized patients diagnosed with AD and CAP from January 2020 to August 2025 were included as study subjects (n=371). Data collected included demographic and baseline characteristics, comorbidities and behavioral risk factors, assessment scales, and laboratory indicators. The data were randomly split into a training set (n=259) and a validation set (n=112) in a 7: 3 ratio. Subsequently, least absolute shrinkage and selection operator (LASSO) regression with 10-fold cross-validation was employed to screen for optimal predictors. Following this screening process, logistic regression was used to develop the final prediction model, and a nomogram was constructed based on this model. The model performance was evaluated using multiple metrics, including the area under the receiver operating characteristic (ROC) curve, calibration curve, and decision curve analysis (DCA). Results　The incidence of adverse outcomes in AD patients with CAP was 27.76% (103/371). Using LASSO regression, four potential predictors were selected from 20 candidate variables: CDR score, CURB-65 score, ADL score, and dysphagia. Subsequent multivariate logistic regression analysis confirmed that these four factors were independent influencing factors for adverse outcomes (P<0.05), with the following odds ratios: CDR score (OR=2.304, 95%CI=1.486-3.572), CURB-65 score (OR=2.263, 95%CI=1.477-3.468), ADL score (OR=2.337, 95%CI=1.033-5.287), and dysphagia (OR=2.042, 95%CI=1.036-4.023). A clinical prediction model (nomogram) was constructed based on these four predictive variables. The area under the ROC curve for the prediction model was 0.835 (95%CI=0.777-0.892) in the training set and 0.902 (95%CI=0.835-0.968) in the validation set, indicating good discriminative ability. The Hosmer-Lemeshow test results showed χ2 =7.046 (P=0.531) for the training set and χ2 =7.781 (P=0.455) for the validation set, suggesting good model fit. The Brier scores for the calibration curves were 0.137 and 0.108 for the two datasets, further indicating significant consistency between predicted and actual values. Decision curve analysis (DCA) showed that using the nomogram model to predict the risk of adverse outcomes provides clinical benefit when the threshold probability ranges from 0.10 to 0.75. Conclusion　The nomogram model developed in this study effectively predicts the risk of adverse outcomes in AD patients with CAP. This tool facilitates early identification of high-risk populations by clinicians, thereby enabling the formulation of individualized interventions to improve patient prognosis.

Key words: Alzheimer's disease, Community-acquired pneumonia, Poor prognosis, Prediction model, Nomogram, Logistic regression

摘要： 背景　随着我国步入老龄化社会，阿尔茨海默病（AD）发病率逐年上升，后者易合并社区获得性肺炎（CAP），最终导致患者发生不良结局。因此，构建精准的风险预测模型有助于医务人员早期识别高危患者并干预，以减轻家庭与社会压力。目的　构建AD合并CAP患者不良预后风险的预测模型，分析该类患者出现不良结局的高危因素，为制订针对性的治疗措施提供依据。方法　依托湖北省中医院构建的痴呆数据库平台进行数据检索，将2020年1月—2025年8月诊断为AD合并CAP的住院患者作为研究对象（n=371），收集其人口学及基础特征、共病与行为学风险因素、评估量表、实验室检查指标相关数据。将数据集按照7∶3随机拆分为训练集（n=259）和验证集（n=112），采用LASSO回归10折交叉验证筛选最佳预测变量，采用Logistic回归建立最终预测模型，并构建列线图，采用受试者工作特征（ROC）曲线下面积、校准曲线、决策曲线等评价模型性能。结果　AD合并CAP患者不良结局发生率为27.76%（103/371）。利用LASSO回归筛从20个候选变量中选出4个潜在预测因子，分别为CDR评分、CURB-65评分、ADL评分、吞咽障碍。进一步的多因素Logistic回归分析显示，CDR评分（OR=2.304，95%CI=1.486~3.572）、CURB-65评分（OR=2.263，95%CI=1.477~3.468）、ADL评分（OR=2.337，95%CI=1.033~5.287）、吞咽障碍（OR=2.042，95%CI=1.036~4.023）是AD合并CAP患者发生不良结局的影响因素（P<0.05）。根据多因素Logistic回归分析筛选出的4个预测变量，构建临床预测模型。预测模型在训练集和验证集的ROC曲线下面积分别为0.835（95%CI=0.777~0.892）和0.902（95%CI=0.835~0.968），提示模型区分度较好；Hosmer-Lemeshow检验结果显示，训练集：χ2=7.046（P=0.531）和验证集：χ2=7.781（P=0.455）；两个数据集的校准曲线Brier评分分别为0.137与0.108，提示预测值与实际值之间存在显著一致性；DCA结果显示，当阈值为0.10~0.75时，使用列线图模型预测不良结局发生风险可使患者临床获益。结论　本研究构建的列线图模型可预测AD合并CAP患者出现不良结局的风险，便于临床医师早期识别高危人群，并制定个体化干预措施，改善患者预后。

关键词: 阿尔茨海默病, 社区获得性肺炎, 不良预后, 预测模型, 列线图, Logistic 回归