Statistical power analysis for PET studies in humans.

Abstract

Although simple techniques have been established to determine statistical power when comparing, for example, the means of two groups of sampled data, the analysis is more complicated when establishing a trend in the data, such as with a linear regression. We present an approach to calculate the sample size necessary to reject the hypothesis that there is no trend in the data (slope is not different from zero) at a given level of statistical significance, given the intra- and inter-subject variability of the measurement. We have derived analytically the distribution of the t statistic, for a given non-zero slope, and integrated this distribution to determine in what fraction of trials a real trend in the population would be missed. We illustrate our approach by re-examining the issue of an age-related impairment in presynaptic dopamine metabolism as measured by PET. We showed that the sample size necessary to determine whether 6-18F-fluoro-L-dopa retention decreases with age depends critically on both the variability of the quantitative method used and on the magnitude of the expected change. The method we have illustrated is a simple statistical test that allows investigators to be certain that an experimental design has a sufficient sample size to demonstrate the effect under study.