Bioeffect modeling of nonuniform dose distributions for paired organs

Evaluating treatment plans on the basis of risk/benefit to the patient depends on accurate characterization of dose-response distributions from clinical data. Typically, response data is defined as a binary complication variable according to a pre-determined threshold value. This binning of the data into two groups can mask gradual dose-response changes, and may make the dose-response model overly sensitive to the choice of complication threshold. Here, the authors explore two new fitting functions, the Generalized Mills Ratio (GMR) and the log-normal integral (LNI), with the goal of modeling the bioeffect indicator without the use of binning. These new fitting functions give the modeling a wide range of response curves of the type the authors are working with. These functions are applied to dose-response data for parotid glands treated with 3D treatment planning techniques, primarily intensity modulated radiation therapy (IMRT). Stimulated saliva flow 6 months after treatment is used as the bioeffect indicator. Preliminary analysis indicates that for the parotid glands, the lesser of the two mean doses may be an important indicator. In other words, the response may be correlated, and new methods of data analysis are important to analyze the correlation. An example of the computation of confidence intervals is given using bootstrap resampling techniques