Models to predict outcome from childhood neuroblastoma: the role of serum ferritin and tumor histology.

We report on the development of three multiple logistic regression models to predict death from childhood neuroblastoma in patients treated without bone marrow transplantation. The models have been developed using a data set of 125 patients for whom age, stage, serum ferritin, and/or histology were available from diagnosis. Seventy-seven patients had all four variables recorded at diagnosis, 34 had age, stage, and serum ferritin, and 14 had age, stage, and histology. Minimum time from diagnosis for all patients was 3 years. The four-variable (full) model showed a predictive value positive rate (or 1 - the false positive rate) of 91.3% and a predictive value negative rate (or 1 - the false negative rate) of 94.4%. Survival curves, based on derived "good" and "poor" prognosis, were constructed for the full model of 77 patients and for the same patients using subset models either without ferritin or without histology. Correcting for prognostic factors noted at diagnosis, no time trend could be identified over the study period. Point estimates for the probability of death in all three models are displayed in graphical form. The results suggest that serum ferritin and tumor histology at diagnosis have independent prognostic significance and that patient outcome in neuroblastoma can be very accurately predicted with a four-variable model. Such information will help sort patients into good and poor prognosis for bone marrow transplant and intensive chemotherapy protocol triage and will help evaluate the efficacy of future therapeutic innovations.