A non-parametric method of reconstructing single-dose survival curves from multi-fraction experiments

Purpose: A method of estimating the single-dose curve from designed multifraction experiments is described and applied to three datasets. Materials and methods: The method, which is non-parametric and based on standard statistical regression techniques, can be used for functional endpoints which are either continuous or binary. The datasets are concerned with wound healing on mice, myelopathy in guinea pigs and spermatogenesis in mice. The results are compared with the results from fitting the linear quadratic model. The statistical methods of Bootstrapping and residual plots are illustrated. Results: The method is based in part on an assumed statistical model, however, exact knowledge of the correct statistical model is not necessary to obtain an estimate of the shape of the single-dose survival curve. We find no good evidence from the reconstructed single-dose survival curve of an 'induced repair' phenomena at low doses for the wound healing and spermatogenesis experiments. For the myelopathy experiment the data are consistent with the LQ, model with a low alpha-beta ratio down to doses of at least 1.5 Gy per fraction. Conclusions: A robust statistical method of estimating the shape of the single-dose survival curve is demonstrated using standard statistical software on three datasets.