Advances in our understanding of the genetic landscape of cancer has allowed for a more sophisticated understanding of carcinogenesis and progression, as well as a more nuanced approach to treatment. Various cellular environments can favor sensitivity or resistance to irradiation, and recent investigations have suggested that certain genetic markers and transcriptomic nodes are associated with this response. Contrary to the standard null hypothesis in biology that any polygenic trait would exhibit a normal distribution, in a previous investigation of the variance in radiosensitivity across cancers we observed bi- or multi-modal distributions within classically defined cancer types. We hypothesize that these disease sites are aggregations of disparate sub-classes of disease, each with their own normally distributed radiation sensitivity. We utilized the measured distributions of radiation sensitivity, as derived from the well-validated radiation sensitivity index (RSI) from an IRB-approved tissue biorepository of 8,271 cancer patients with a variety of tumors characterized by gene expression. To address the hypothesis that these observed multi-modal distributions were indeed aggregations of normal distributions, and therefore inform clinical sub-classification, we reanalyzed these data. For each site of origin, we minimized the Bayesian information criteria (BIC) and Akaike information criteria (AIC) of multiple mixed-Gaussian models of the calculated RSI. We constructed mixture models for tumor tissue sampled from 20 sites of origin. The corresponding optimization illustrated varying degrees of heterogeneity in radiosensitivity across cancer types. In order to limit the effects of sample size, we restricted our results to sites with greater than 50 samples available for analysis. We subsequently identified varying numbers of distinct subpopulations of disparate radiosensitivity within a specific cancer type, ranging between: 1 [glioma, sarcoma, melanoma], 2 [lung, prostate, esophageal, gastric, H&N], 3 [endometrial, rectal, pancreatic], and 4 [kidney, breast] Gaussian clusters via posterior probability as predicted by the resultant mixture model. Lastly, cumulative distribution functions were constructed and related to a simplified linear-quadratic model to illustrate the potential for dose adjustment in order to maximize treatment efficacy and minimize patient harm. Our results suggest that the dose-response relationship can be better understood as a representation of the cumulative distribution function across the patient population, within which lie discrete sub-populations of disparate sensitivity to radiation. This interpretation indicates further opportunities for biomarker discovery to allow for a more informed approach to dosing decisions and clinical trial stratification in radiation therapy, and provides clues to differences in biology which can be explored in the laboratory.