Validation of a pneumonia prognostic index using the MedisGroups Comparative Hospital Database

Abstract

Purpose: Our purpose was to validate a previously developed pneumonia-specific prognostic index in a large, multicenter population.

Patients and methods: We developed a pneumonia-specific prognostic index in a prospective, multicenter study of 346 patients with clinical and radiographic evidence of pneumonia admitted to 3 Pittsburgh hospitals (the derivation cohort), and validated the index in 14,199 patients with a principal ICD-9-CM diagnosis of pneumonia admitted to 78 hospitals in the 1989 MedisGroups Comparative Hospital Database (the validation cohort). The prognostic index classified patients into five ordered risk classes based on six predictors of mortality: age greater than 65 years, pleuritic chest pain, a vital sign abnormality, altered mental status, neoplastic disease, and high-risk pneumonia etiology. Each patient in the validation cohort was assigned to a risk class by obtaining values for the index's six predictors in the MedisGroups population. The performance of the prognostic index in the derivation and validation cohorts was assessed by comparing hospital mortality rates within each of the index's five prognostic risk classes.

Results: The hospital mortality rate was 13.0% in the derivation cohort, and 11.1% in the validation cohort (p = 0.26). The agreement in the risk class-specific mortality rates was striking with the exception of class V: in class I, mortality was 0% in the derivation cohort versus 1% in the validation cohort; in class II, 0% versus 1.1%; class III, 10.9% versus 8.6%; class IV, 21.8% versus 26.2%; and class V, 73.7% versus 37.7%. There were no statistically significant differences in mortality rates within the first four risk classes, which represented the vast majority of patients in the derivation (94%) as well as the validation (98%) cohorts.

Conclusions: These data support the generalizability of a pneumonia-specific prognostic index. This index, which performs exceptionally well in classifying low-risk patients, may help physicians identify patients with community-acquired pneumonia who could safely be managed in the ambulatory setting, or if hospitalized, the patients that could be treated with abbreviated inpatient care.