Integrated analysis of tissue histology with the genome-wide array (e.g., OMIC) and clinical data have the potential for hypothesis generation and be prognostic. OMIC and clinical data are typically characterized and summarized at the patient level while whole mount histological sections are often heterogeneous in terms of nuclear morphology and organization. In this paper, we propose a multilevel framework for summarization and association of morphometric data. At the lowest level, each nucleus is segmented and then profiled with a multi-dimensional representation. At the intermediate level, cellular profiles are summarized within a local neighborhood, and further clustered into subtypes. At the highest level, each patient is represented by the composition of subtypes that are computed from the intermediate level, and then integrated with OMIC and outcome data for further analysis. The framework has been applied to Glioblastoma multiforme (GBM) data from The Cancer Genome Atlas (TCGA). Based on cellularity and nuclear size, four subtypes have been identified at the intermediate level. Subsequent multi-variate survival analysis indicates that the patient composition of one of the subtypes, with extremely low cellularity and small nucleus size, has a significantly higher hazard ratio. Further correlation of this subtype with the molecular data reveals enrichment of (i) STAT3 pathway and (ii) common regulators of PKC, TNF, AGT, and PDGF.