We conducted a cohort study of 423 intensive care unit (ICU) admissions with a primary clinical diagnosis of acute respiratory failure, a PaO2/FIO2 on ICU admission of < 300 mm Hg, and an ICD-9 discharge diagnosis of adult respiratory distress syndrome (ARDS) (518.5 or 518.82) drawn from a nationally representative database of 17,440 ICU admissions to evaluate current and proposed revisions for definitions of ARDS. A variety of nonpulmonary physiologic risk factors, from shock to elevated serum bilirubin measurements, were significant (p < 0.01) for hospital mortality. Multivariable analysis using the admission APACHE III score, primary ICU admission diagnosis, and treatment location before ICU admission provided greater accuracy in prediction (ROC = 0.80) than the individual PaO2/FIO2 (ROC = 0.68). Patients were given an individual risk of hospital mortality based on their admission APACHE III score, treatment location before ICU admission, and ICU admitting diagnosis. Dividing the patient population into groups using a PaO2/FIO2 < or = 150 resulted in a wide range of individual risk for hospital mortality, from < 10 to > 90% in both groups. We conclude that ARDS is a complex clinical entity with a variety of pulmonary and nonpulmonary risk factors for both its development and its prognosis. Current and proposed categorical definitions based on the severity of hypoxemia result in a wide distribution of individual patient risks. Use of these findings in the design and conduct of future clinical trials would improve the evaluation of new therapies.