Outcomes and prognostic factors in patients with haematological malignancy admitted to a specialist cancer intensive care unit: a 5 yr study

Long-held assumptions of poor prognoses for patients with haematological malignancies (HM) have meant that clinicians have been reluctant to admit them to the intensive care unit (ICU). We aimed to evaluate ICU, in-hospital, and 6 month mortality and to identify predictors for in-hospital mortality. A cohort study in a specialist cancer ICU of adult HM patients admitted over 5 yr. Data acquired included: patient characteristics, haematological diagnosis, haematopoietic stem cell transplant (HSCT), reason for ICU admission, and APACHE II scores. Laboratory values, organ failures, and level of organ support were recorded on ICU admission. Predictors for in-hospital mortality were evaluated using uni- and multivariate analysis. Of 199 patients, median age was 58 yr [inter-quartile range (IQR) 4666], 51.7 were emergency admissions, 42.2 post-HSCT, 51.9 required mechanical ventilation, median APACHE II was 21 (IQR 1625), and median organ failure numbered 2 (IQR 14). ICU, in-hospital, and 6 month mortalities were 33.7, 45.7, and 59.3, respectively. Univariate analysis revealed bilirubin 32 mol litre(1), mechanical ventilation, epsilon 2 organ failures, renal replacement therapy, vasopressor support (all P0.001), graft-vs-host disease (P0.007), APACHE II score (P0.02), platelets 2010(9) litre(1) (P0.03), and proven invasive fungal infection (P0.04) were associated with in-hospital mortality. Multivariate analysis revealed that epsilon 2 organ failures [odds ratio (OR) 5.62; 95 confidence interval (95 CI), 2.3013.70] and mechanical ventilation (OR 3.03; 95 CI, 1.336.90) were independently associated with in-hospital mortality. Mortality was lower than in previous studies. Mechanical ventilation and epsilon 2 organ failures were independently associated with in-hospital mortality. oTraditional' variables such as neutropenia, transplantation status, and APACHE II score no longer appear to be predictive.