Influence of selection bias on dose-time analysis: a theoretical investigation.

Abstract

We investigate the effect of patient selection factors on the interpretation of dose-time analyses of retrospective head and neck data treated with external beam radiation therapy. We use a computer simulation model to ascertain the influence of individualizing the fractionation regimen on the estimated steepness of dose-response and time-response estimates, on the probability of local control and on the validity of statistical tests and confidence intervals. We assume that the total dose and overall treatment time for each patient are altered by consideration of three possible factors, the pre treatment size of the tumor, the response of the tumor to therapy or the amount of acute effects. We find that individualizing treatment leads to very small improvements in the local control rate, but substantial biases in the estimates of the doubling time and the effective D0 value. Patient selection also compromises the validity of confidence intervals and tests. Despite this selection bias our results suggest that if there were no real effect of treatment time then the influence of individualizing treatments would not cause treatment time to be found to be as important as it has been shown to be in many studies. Assuming the fundamental radiobiological assumptions in the model are appropriate, individual alterations of the fractionation schedule due to the response to therapy can compromise the interpretation of dose-time analyses of retrospective data, without giving much improvement in the control rate.