The data to be derived in this Program Project have multivariate-multi-source variation features. The EPR pO2 measurements combine longitudinal changes of tumor oxygen on the scale of minutes, in the case of hyperoxygenation experiments, with chronic changes on the scale of days and weeks compounded with variation across patients with the same tumor, variation of pO2 across sensor locations within tumor, and variations across tumor types. The traditional t-test is simply not applicable to capture all these features and therefore more sophisticated statistical models and techniques are required. We will be using the theory of mixed models that combines nonlinear modeling of pO2 over the time with variation within and between tumors to promptly address the heterogeneity of oxygen distribution in vivo. The rationale of having a stand-alone Biostatistics Core at Dartmouth is two-fold: (1) the data from different projects will be analyzed under one methodological umbrella, (2) Dartmouth biostatisticians are the most experienced in handling multi-source variation nonlinear pattern data with a long-year track record of working with cancer biologists and clinicians. The following tasks, among others, will be performed at Biostatistics Core: (a) estimation of intra and inter tumor pO2 variation at the baseline using multisite measurements, (b) develop predictive models for increase of oxygen due to hyperoxygenation on the scale of minutes, (c) estimation of longitudinal patterns of tumor oxygen over the course of days and weeks as a consequence of cancer therapy such as radiation, surgery or chemo, (d) classification of cancer patients as responsive or non-responsive to hyperoxygenation by statistical means. These tasks will be carried out for different types of tumor and using different EPR techniques, such as india ink, OxyChip and deep- tissue implanted resonators. Biostatistics Core at Dartmouth will work with close collaboration with researchers from all three projects. We are planning to have regular meetings to discuss statistical analyses and findings along with educational sessions for clinicians and biologists to explain advanced statistical methods and models.
