Prognostic scores, such as the PRISM and APACHE II, have been established, predicting with reasonable accuracy the outcome of patients admitted to intensive care units (ICU). In keeping with previous reports, we found, however, that these scores failed to perform in a series of 28 recipients of hematopoietic auto- or allografts (BMT) who required ICU admission for reasons including respiratory (82%) and multi-organ (36%) failure. We therefore retrospectively analyzed the charts of these patients, evaluating predisposing factors and prognostic variables which might confound the validity of these ICU tools which in other clinical scenarios have proven so valuable. Of all the parameters tested, logistic analysis established the following as predictors for poor outcome: increased C-reactive protein (CRP) to > 10 mg/dl (P = 0.04), macroscopic hemorrhage (P = 0.04), hypotension (mean arterial pressure < normal) (P = 0.04) and GVHD > or = III (P = 0.002). Most of these factors are not accounted for by the standard prognostic questionnaires. The development of an 'oncological' or 'post-BMT' risk of mortality score, taking into account these patients' specific clinical problems, might improve the risk assessment for this patient group, and might thus facilitate the timely recognition of those patients most in need of more intensive therapeutic measures.