RESUMO
OBJECTIVES: Undernourished children with cancer are at major risk for adverse outcomes. We intended to model nutritional status as risk factor for hospital death. METHODS: Retrospective analysis of two cohorts of children admitted in an oncology/hematology intensive care unit. Logistic regression was used for model building. RESULTS: The derivation cohort had 155 patients, with a median age of 104.9 mo., and 36 deaths (23.2%). Twenty-eight children (18.1%) had the z score of the body mass index < -2. A multivariate model with the variables "relapse of oncological disease" (Odds Ratio 3.14, P = 0.025), "surgical case" (OR 0.22, P = 0.002), "intubation/mechanical ventilation" (OR 8.38, P = 0.000) and "body mass index z score < -2" (OR 3.21, P = 0.024) generated a logit with good predictive capacity for "hospital death". In the validation cohort, with 450 patients (median age of 92.7 mo.), the model was able to predict 55.5 of the 58 observed deaths (Standardized Mortality Rate = 1.04, 95% CI 0.80-1.34, P = 0.72), with P = 0.68 in the Hosmer test. The AUC was 0.90 (95% CI 0.86-0.93). CONCLUSION: Undernutrition increases the risk of death, and is a variable that should be included in predictive mortality algorithms.