CLUSTERS OF HISTOLOGIC CHARACTERISTICS IN CHILDREN WITH INFRATENTORIAL NEUROGLIAL TUMORS

Five quantitative histologic factors, differing linear combinations of 26 reliably recognized histologic features, account for much of the h istologic variance in 1068 children with infratentorial neuroglial tum ors in the Childhood Brain Tumor Consortium (CBTC) database. In this s tudy, we used the scores on the Spongy, Proliferative, Ring, Fibrillar y, and Nuclear factors in cluster analyses and identified 11 clusters of children's tumors. Each had statistically significant differences i n histology and relative histologic homogeneity. Three clusters had ep endymoma-like histologic features; 4 had astrocytoma-like features; an d 4 had primitive neuroectodermal-like (PNET or medulloblastoma) featu res. Each cluster had a unique high/low mean factor score pattern. Mul tiple operative and other clinical features characterized the three gr oups of clusters. We used Kaplan-Meier survival models to test for dif ferences in survival among clusters and proportional hazards survival models to adjust for associated covariates. Among the 'ependymoma' clu sters the 5 year survival probability ranged from 0.25 to 0.54. Among the 4 'astrocytoma' clusters, 5 year survival probability ranged from 0.59 to 0.94. The 5 year survival probability for the 'medulloblastoma ' clusters ranged from 0.20 to 0.44. Within the three groups, clusters had differing covariates associated with survival. The tumor clusters identified in this study ensure relatively homogeneous histologic sub sets. The five factor scores of a child's tumor provide the basis for finding the cluster nearest to that tumor. We propose that this tumor clustering strategy be employed for selection of children and for anal yses of therapeutic clinical trials.