BACKGROUND: We sought to identify risk factors for mechanical ventilation in patients with malignancies and acute respiratory failure (ARF). METHODS: We analyzed data from a previous randomized controlled trial in which nonintubated oncology and hematology subjects with ARF were randomized to early bronchoalveolar lavage or routine care in 16 ICUs in France. Consecutive patients with malignancies were admitted to the ICU for ARF in 2005 and 2006 with no intervention. RESULTS: During the study period, 219 patients were admitted to the ICU for ARF, and 8 patients were not included due to a nonintubation order. Data on the underlying disease, pulmonary involvement, and extrapulmonary organ dysfunctions were recorded at admission in the 211 remaining subjects. Ventilatory support included oxygen only (49 subjects), noninvasive ventilation (NIV) only (81 subjects), NIV followed by invasive mechanical ventilation (49 subjects), and first-line invasive mechanical ventilation (32 subjects). The 81 subjects who required invasive mechanical ventilation were compared with the 130 subjects who remained on oxygen or NIV. Factors associated with invasive mechanical ventilation by multivariate analysis were the oxygen flow required at ICU admission, the number of quadrants involved on chest x-ray, and hemodynamic dysfunction. Mortality rates for subjects who had NIV failure were 65.3% compared with 50% for subjects who were first-line intubated ( P = .34). CONCLUSIONS: In cancer patients with ARF, hypoxemia, extent of pulmonary infiltration on chest x-ray, or hemodynamic dysfunction are risk factors for invasive mechanical ventilation. Mortality was not significantly different between NIV failure and first-line intubation.