To identify predictors of mortality, the records of 133 elderly patients with pneumonia admitted to a small rural midwestern hospital were examined using a retrospective cohort design. All recorded clinical information available to the patient's physician within the first hours of admission was reviewed. Twenty-one (15.8%) patients died during the hospitalization. Patients with preexisting coronary heart disease, dementia, urinary incontinence, and impaired mobility were more likely to die. Impaired mental status, absence of fever, rapid respiratory rate, hypotension, cyanosis, and diffuse abnormalities on chest examination were also associated with mortality. Logistic regression analysis revealed five predictive indicators of mortality: impaired level of consciousness (odds ratio [OR] = 11.3), tachypnea (OR = 10.8), temperature lower than normal (OR = 14.2), white cell count higher than 20 X 10(9)/L (20,000 mm-3) (OR = 12.2), and cyanosis (OR = 8.6). A risk score based on this regression model demonstrated that 1 of 95 patients with a score lower than 3 (1%), 7 of 22 with a score of 3 (32%), and 13 of 15 patients with a score higher than 3 (87%) died during their hospitalization. The validity of this risk-scoring system was confirmed in another sample of 40 patients. Studies such as this may be useful in identifying information of important prognostic value that enables physicians, patients, and family members to make more effective decisions.