Comparison of a disease-specific and a generic severity of illness measure for patients with community-acquired pneumonia

Abstract

Objective: 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.

Design: A retrospective database study.

Patients: Adult patients (aged > or = 18 years) with an ICD-9-CM principal diagnosis of pneumonia in 78 MedisGroups Comparative Database hospitals.

Methods: 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.

Results: 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.

Conclusions: 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.