As evidenced by the univariate analysis, warfarin use is associated with a number of demographic and clinical characteristics that are known to be related to trauma mortality. To adjust for potential confounding, we created a multivariate logistic model using demographic and clinical characteristics (comorbidities) determined to be related to warfarin use in the univariate analysis. The most parsimonious model adjusted sex, age, Trauma-Related Injury Severity Score, race, chronic obstructive pulmonary disease, nonemphysematous pulmonary disease, coronary artery disease, dialysis dependence, and treatment facility (Table 4). In this model, warfarin use was associated with a significantly higher odds of death both among all patients (OR, 1.72; 95% CI, 1.63-1.81; P < .001; R2 = 0.27; area under the receiver operating characteristic curve = 0.893; Hosmer-Lemeshow goodness-of-fit P < .01) and patients 65 years and older (OR, 1.38; 95% CI, 1.30-1.47; P < .001; R2 = 0.16; area under the receiver operating characteristic curve = 0.792; Hosmer-Lemeshow goodness-of-fit P < .01). For patients with intracranial hemorrhage, chronic obstructive pulmonary disease and coronary artery disease did not contribute significantly to the model and were therefore dropped. Among all patients with intracranial hemorrhage, the OR for death was 1.26 for warfarin users (OR, 1.07-1.48; P = .005; R2 = 0.21; area under the receiver operating characteristic curve = 0.833), but the OR for death increased to 1.46 (OR, 1.00-2.12; P = .05; R2 = 0.24; area under the receiver operating characteristic curve = 0.847) for patients younger than 65 years.