Fine M J, Hanusa B H, Lave J R, Singer D E, Stone R A, Weissfeld L A, Coley C M, Marrie T J, Kapoor W N
Department of Medicine, University of Pittsburgh, Pennsylvania, USA.
J Gen Intern Med. 1995 Jul;10(7):359-68. doi: 10.1007/BF02599830.
To compare the performances of a disease-specific severity of illness index and a prototypical generic severity of illness measure, MedisGroups Admission Severity Groups (ASGs), for patients with community-acquired pneumonia.
A retrospective database study.
Adult patients (aged > or = 18 years) with an ICD-9-CM principal diagnosis of pneumonia in 78 MedisGroups Comparative Database hospitals.
The pneumonia severity of illness index (PSI) was developed to predict hospital mortality using logistic regression analyses in a 70% random sample of study patients. The performances of the PSI and the generic severity measure were assessed among the remaining 30% of patients by comparing observed mortalities within the five PSI and ASG severity classes, and areas under their receiver operating characteristic (ROC) curves. Both the PSI and the generic severity measure were used to estimate the 95% confidence interval of the expected number of deaths in each of the 78 study hospitals. Hospitals with an observed number of deaths outside these limits were identified as outliers.
There were 14,199 study patients who had community-acquired pneumonia, and 1,542 (10.9%) died during hospitalization. In comparison with the generic severity measure, the PSI more accurately identified patients at extremely low risk of death, and had a larger area under its ROC curve (0.84 vs 0.79; p < 0.0001). Of the 78 study hospitals, 17 (21.8%) were classified as outliers for mortality by at least one severity adjustment system. Among the 11 low-outlier hospitals, six were classified by the generic severity measure alone, two by the PSI alone, and three by both systems; among the six high-outlier hospitals, one was classified by the generic measure alone, three by the PSI alone, and two by both systems.
The PSI provided more accurate estimates of hospital mortality and classified different hospital outliers for mortality than did the generic severity of illness measure for patients with community-acquired pneumonia.
比较疾病特异性疾病严重程度指数与典型的通用疾病严重程度测量方法（MedisGroups入院严重程度分组，即ASG）在社区获得性肺炎患者中的表现。
一项回顾性数据库研究。
78家MedisGroups比较数据库医院中，国际疾病分类第九版临床修订本（ICD-9-CM）主要诊断为肺炎的成年患者（年龄≥18岁）。
在70%的随机抽样研究患者中，采用逻辑回归分析制定肺炎疾病严重程度指数（PSI）以预测医院死亡率。通过比较五个PSI严重程度等级和ASG严重程度等级内的观察到的死亡率以及它们的受试者操作特征（ROC）曲线下的面积，在其余30%的患者中评估PSI和通用严重程度测量方法的表现。PSI和通用严重程度测量方法均用于估计78家研究医院中每家医院预期死亡人数的95%置信区间。观察到的死亡人数超出这些范围的医院被确定为异常值。
有14199例研究患者患有社区获得性肺炎，1542例（10.9%）在住院期间死亡。与通用严重程度测量方法相比，PSI更准确地识别出死亡风险极低的患者，并且其ROC曲线下的面积更大（0.84对0.79；p<0.0001）。在78家研究医院中，17家（21.8%）被至少一种严重程度调整系统归类为死亡率异常值医院。在11家低异常值医院中，6家仅被通用严重程度测量方法归类，2家仅被PSI归类，3家被两种系统归类；在6家高异常值医院中，1家仅被通用测量方法归类，3家仅被PSI归类，2家被两种系统归类。
对于社区获得性肺炎患者，与通用疾病严重程度测量方法相比，PSI能更准确地估计医院死亡率，并对不同的医院死亡率异常值进行分类。
