Validation of the Pneumonia Severity Index Among Patients Treated at Home or in the Hospital

摘要

To assess the predictive validity of the pneumonia severity-of-illness index (PSI), a mortalityprediction rule, and extend the work of others by including data on outpatients treated for pneumonia. Methods: Prospective study of 675 consecutive patients with community-acquired pneumonia (CAP) [501 inpatients and 174 outpatients] treated at primary care practice clinics or emergency departments at nine medical centers (five community healthcare systems, three university-affiliated hospital systems, and one Veterans Affairs Medical Center) in Georgia and Virginia in the US between November 1996 and March 1998. Data, including demographic characteristics, co-morbid conditions, laboratory and chest x-ray results, were collected from surveys administered to patients at inception, 2, 15, and 30 days and from retrospective medical chart review. We computed the PSI for each patient using demographic and prognostic factors including age, gender, co-existing illnesses, vital signs, laboratory test results and the corresponding logistic regression parameters from previous research. In addition, the Pneumonia Outcomes Research Team (PORT) prediction rule was used to risk adjust patients for mortality severity by disposition. The PSI performed well in its ability to predict mortality for our sample of patients with an area under the Receiver Operating Curve (ROC) of 0.757, significantly different than chance (p < 0.01). Results of the Homser and Lemeshow goodness of fit test also indicated that the PSI was a reasonably good predictor of mortality for our patients. Twenty-eight patients (4.1%) died within the 30-day observation period. Using the PORT prediction rule we found that 27 of the deaths occurred among inpatients (three in class II, five in class III and 19 in class IV). One of these deaths occurred among outpatients (risk class IV). The PSI is a valid predictor of mortality for outpatients and inpatients treated in various community-based settings.