Childhood brain tumor update.

In this review of recent studies of the cerebellar tumors of childhood we have discussed the limitations of traditional approaches to the classification of childhood brain tumors for the purposes of estimating prognosis. While these traditional approaches have provided considerable information about cells of origin, they have left behind the ancient tradition of pathology of considering the tumor in all its aspects, not just its histologic features. As an alternative we have introduced an approach that does not depend ona priori assumptions but rather allows prognosis to be estimated directly from histologic features or from clusters of histologic, clinical, and surgical features-- namely the use clustering strategies as applied to the cerebellar tumors. We have also discussed the preliminary work of the Childhood Brain Tumor Consortium, which has collected a large amount of data to serve as a source of prognostic information in the design of studies of alternative therapies. The preliminary data indicate that there is a disquietingly wide array of overlapping histologic features in three common childhood brain tumors-- medulloblastomas, pilocytic astrocytomas, and ependymomas. Clinicians who seek to reduce the number of deaths among children with brain tumors need specific prognostic information to decide among alternative therapies and to design studies of therapies for biologically homogeneous tumors, rather than studies of therapies of tumor "names" that encompass extraordinarily wide ranges of histologic features. Clinicians must also be extremely wary of "historical" controls; they cannot allow themselves to fall into the trap of believing that a new therapy is better because their patients "would have been expected" to have succumbed sooner. For each proposed therapy new studies must be designed, each with its control population, no matter how difficult this is. Historical controls assume that 1) the tumors were classified properly, 2) the sites of the tumors in different groups of patients were comparable, and 3) tumors with the same name (each encompassing an extremely wide range of histologic features) are comparable in every histologic respect to the study population, and 4) the outlook for patients with tumors with this name has not changed over the half century since the name came into common usage (i.e., that the morbidity associated with these tumors has not changed). For these reasons the use of historical controls must be condemned and all studies of therapies based on such controls discarded. Studies of therapies must compare populations of patients homogeneous as to site, histologic features, and such clinical features as age.