Confidence limits of the probability of success in animal experiments and clinical studies: A Bayesian approach

Analysis of tumour dose-response data from animal experiments via two TCP models accounting for tumor hypoxia and resensitization

To treat animal dose-response data exhibiting inverse dose-response behavior with two tumor control probability (TCP) models accounting for tumor hypoxia and re-oxygenation leading to resensitization of the tumor. One of the tested TCP models uses a modified linear-quadratic (LQ) model of cell survival where both α and β radiosensitivities increase in time during the treatment due to re-oxygenation of the hypoxic tumor sub-population. The other TCP model deals with two types of hypoxia-chronic and acute-and accounts for tumor re-sensitization via oxygenation of the chronically hypoxic and fluctuating oxygenation of the acutely hypoxic sub-populations. The two models are fit using the maximum likelihood method to the data of Fowler et al. on mice mammary tumors irradiated to different doses using different fractionated schedules. These data are chosen since as many as five of the dose-response curves show an inverse dose behavior, which is interpreted as due to re-sensitization. The p-values of the fits of both models to the data render them statistically acceptable. A performed comparison test shows that both models describe the data equally well. It is also demonstrated that the most sensitive (oxic) tumor component has no impact on the treatment outcome. The ability of the tested models to predict and describe the impact of re-sensitization on the treatment outcome is thus proven. It is also demonstrated that prolonged treatment schedules can be more beneficial than shorter ones. However, this may be true only for schedules with small number of fractions, i.e. for hypo-fractionated treatments only.