基于老年综合征构建急诊老年肺炎合并脓毒症患者28d死亡预测模型并验证

目的 构建并验证基于老年综合征的急诊老年肺炎合并脓毒症患者28d死亡的预测模型。方法 连续纳入2022年1月至11月就诊于北京朝阳医院急诊科年龄65岁的社区获得性肺炎合并脓毒症患者607例。使用R软件随机将患者分为训练集425例(70%)和测试集182例(30%)。根据随访28d生存情况,将患者分为生存组421例和死亡组186例。记录患者的一般资料、入院时生命体征、实验室指标及老年综合征相关内容。在训练集中,采用最小绝对收缩算子回归及Cox回归进行变量筛选,确定患者28d死亡的独立影响因素并构建列线图。通过时间依赖的受试者工作特征曲线下面积(AUC)和校准曲线在测试集中对模型的区分度和校准度进行验证。采用临床决策曲线分析(DCA)验证模型的临床有效性。结果 多因素Cox回归分析显示,近1年跌倒史(HR=1.613,CI 1.098~2.730；P=0.015)、Glasgow评分(HR=0.923,95%CI 0.880~0.968；P=0.001)、临床衰弱量表评分(HR=1.265,95%CI 1.081~1.481；P=0.003)、外周血氧饱和度(HR=0.966,95%CI 0.949~0.983；P＜0.001)、乳酸(HR=1.159,95%CI 1.095~1.226；P＜0.001)、白蛋白(HR=0.963,95%CI 0.930~0.998；P=0.037)、降钙素原(HR=1.010,95%CI 1.001~1.020；P=0.026)、白细胞(HR=1.028,95%CI 1.007~1.050；P=0.009)和红细胞(HR=0.731,95%CI 0.600~0.891；P=0.002)为急诊老年肺炎合并脓毒症患者28d死亡的独立影响因素。纳入上述9个变量构建的列线图模型预测患者28d死亡的时间依赖的AUC在训练集和测试集中均＞0.80,提示该模型在训练集和测试集中均具有较好的判别能力。校准曲线显示,列线图模型在训练集和测试集中的预测概率与观察概率一致性较好。DCA证实,列线图模型在测试集及训练集中均具有较好的临床获益。结论 基于老年综合征(跌倒、衰弱和营养不良)的列线图模型可有效预测急诊老年肺炎合并脓毒症患者28d死亡,有利于临床医生早期进行危险分层并采取干预措施。

Development and validation of a prediction model for 28-day mortality among older patients with pneumonia-induced sepsis based on geriatric syndrome

Objective To construct and validate a geriatric syndrome-based prediction model for 28-day mortality in older patients with pneumonia-induced sepsis admitted in emergency department. Methods A total of 607 patients aged ≥65 years with pneumonia-induced sepsis visiting the emergency department of Beijing Chaoyang Hospital from January to November 2022 were consecutively enrolled in this study. According to their survival status at 28-day follow-up, they were divided into survival group (421 cases) and death group (186 cases). With the aid of R software, they were randomly assigned into a training set (425 cases, 70%) and a test set (182 cases, 30%). General data, vital signs at admission, laboratory indicators and geriatric syndrome-related indicators were recorded. Least absolute shrinkage and selection operator regression and Cox regression were used to determine independent variables related to 28-day mortality in training set, and a nomogram was then constructed. Time-dependent area under receiver operating characteristic curve (AUC) and calibration curve were employed to evaluate the discriminability and calibration of the model, respectively. Decision curve analysis (DCA) was applied for clinical effectiveness of the model. Results Multivariate Cox regression analysis indicated that the history of falls in the past year (HR=1.613,95%CI 1.098-2.730; P=0.015), Glasgow coma scale (HR=0.923,95%CI 0.880-0.968; P=0.001), score of clinical frailty scale (HR=1.265,95%CI 1.081-1.481; P=0.003), peripheral capillary oxygen saturation (HR=0.966,95%CI 0.949-0.983; P<0.001), levels of lactate (HR=1.159,95%CI 1.095-1.226; P<0.001), albumin (HR=0.963,95%CI 0.930-0.998; P=0.037) and procalcitonin (HR=1.010,95%CI 1.001-1.020; P=0.026), and counts of white blood cells (HR=1.028,95%CI 1.007-1.050; P=0.009) and red blood cells (HR=0.731,95%CI 0.600-0.891; P=0.002) were independent influencing factors for 28-day mortality among older patients with pneumonia-induced sepsis. The time-independent AUC value of our nomogram based on above 9 variables was>0.80 in predicting 28-day mortality in both the training set and the test set, indicating the model showing good discriminability in the two tests. Calibration curve analysis revealed that the predicted probability was in consistent with the observed probability in the two tests. DCA confirmed that the nomogram had good clinical benefits in both the training and test sets. Conclusion Our nomogram based on geriatric syndrome (fall, frailty and malnutrition) can effectively predict the 28-day mortality in older patients with pneumonia-induced sepsis in emergency department, which is beneficial for clinical physicians to conduct early risk stratification and implement intervention measures.