Bootstrap resampling: a powerful method of assessing confidence intervals for doses from experimental data

Publisher Website
Google Scholar
Abstract
Bootstrap resampling provides a versatile and reliable statistical method for estimating the accuracy of quantities which are calculated from experimental data. It is an empirically based method, in which large numbers of simulated datasets are generated by computer from existing measurements, so that approximate confidence intervals of the derived quantities may be obtained by direct numerical evaluation. A simple introduction to the method is given via a detailed example of estimating 95% confidence intervals for cumulated activity in the thyroid following injection of Tc-sodium pertechnetate using activity-time data from 23 subjects. The application of the approach to estimating confidence limits for the self-dose to the kidney following injection of Tc-DTPA organ imaging agent based on uptake data from 19 subjects is also illustrated. Results are then given for estimates of doses to the foetus following administration of Tc-sodium pertechnetate for clinical reasons during pregnancy, averaged over 25 subjects. The bootstrap method is well suited for applications in radiation dosimetry including uncertainty, reliability and sensitivity analysis of dose coefficients in biokinetic models, but it can also be applied in a wide range of other biomedical situations.