Risk assessment of patients with hematologic malignancies who develop fever accompanied by pulmonary infiltrates: a historical cohort study.

BACKGROUND The mortality rate associated with fever accompanied by pulmonary infiltrates after chemotherapy for hematologic malignancies remains higher than the corresponding rate associated with febrile neutropenia without pulmonary infiltrates. Nonetheless, few studies have focused on the factors that predict outcome for patients with lung infiltrates. The purpose of the current study was to construct a risk model for clinical use by assessing the factors that affect outcome for patients with fever and pulmonary infiltrates. METHODS A historical cohort of 110 patients with hematologic malignancies who developed fever and pulmonary infiltrates was examined. Using parameters for which data were available at the onset of lung infiltrates, univariate and multivariate analyses were performed to assess factors affecting outcome. After a value of one point was assigned to each significant variable, a prediction score was calculated for each patient; scores were used to generate a system for identifying patients with a low risk of death due to fever accompanied by pulmonary infiltrates. RESULTS The crude mortality rate associated with pulmonary infiltrates was 23%; factors associated with cure included a favorable change in white blood cell counts (odds ratio [OR], 5.6; 95% confidence interval [CI], 1.7–18.9; P = 0.001), C-reactive protein levels < 10 mg/dL (OR, 4.6; 95% CI, 1.6–13.8; P = 0.001), and serum albumin levels ≥ 3 g/dL (OR, 3.2; 95% CI, 1.4–7.3; P = 0.004). Low-risk patients (risk score, 2–3) and high-risk patients (risk score, 0–1) had survival rates of 95% and 46%, respectively (P < 0.0001). The risk model had a specificity of 88% and a positive predictive value of 95%. CONCLUSIONS The risk model tested in the current study accurately predicted the survival of patients with hematologic malignancies who developed fever with pulmonary infiltrates. Once prospectively validated, the model could be used to select patients for trials involving novel diagnostic and therapeutic strategies. Cancer 2004. © 2004 American Cancer Society.