imrmc: Size and Analyze Multi-reader Multi-case Reader Studies, 5.5 Sizing a Future Study
5.5 Sizing a Future Study
The software calculates the expected statistical power of the new study. The software implements two kinds of hypothesis testing. One is for a single modality; the null hypothesis is that the AUC is equal to a specified value. The second compares modalities, the null hypothesis is that the two modalities have the same AUC. The statistical power is computed in two ways. One uses the Z-test, and the other uses the F-test from Hillis [9].
Z-test:
Assumes the variance is known even though it is actually estimated. The Z-test uses the ratio of the affect size over the square root of its estimated variance as the test statistic, and assumes the test statistic follows a standard normal distribution.
F-test:
Does not assume the variance is known. It estimates the non-centrality parameter and the denominator degrees of freedom values using the components of variance (input or estimated) for the specified case and reader sample size and effect size. The degrees of freedom is also determined with an alternate calculation from Obuchowski et al. 2012 [11]. The statistical power is then computed using F distributions, unless the degrees of freedom is greater than 50, at which point the power is computed using a standard normal distribution.
Status
The sizing analysis has been debugged and enabled. We believe all elements are now functioning properly. Confirmatory validation of statistical and sizing analysis by Roe and Metz simulation is in processing. Please contact us with any issues via the issues page.
