Our study’s sample size of 58 physicians was determinedusing an exact McNemar test. The best would have been tobase sample size on GEE method by Pan[14]to align withour planned analysis. We recommend a two-step processwhen estimating sample size for paired binary data whenGEE is the planned analytic method. First, one needs tospecify the marginal proportions and rather than specify thecorrelation as zero, calculate the correlation between themarginal proportions using the SAS program inAppendixCatwww.jclinepi.com. And second, use the SAS macrofor a two-period crossover design availablehttp://www.imbs-luebeck.de/imbs/de/node/30 [31]. The required samplesize based on GEE method by Pan for our study was 52. Assample size estimates based on the asymptotic unconditionalMcNemar test are good approximations of those based onGEE method by Pan, this test is a simpler alternative. Thistest is both convenient and practical, as one need only specifythe inside proportions. A SAS macro to calculate sample sizebased on the asymptotic unconditional McNemar testavailable athttp://www2.bsc.gwu.edu/bsc/docs/JohnLachin/mcnemarn.sas [27]. The required sample size based on theasymptotic unconditional McNemar test for our study was52 (odds ratio53.3,p2150.093,a50.05, andb50.20).