Using pilot data to estimate sample size and compare question forms for a crossover study

A pilot study was conducted on thirty office workers to help determine if a scannable form of symptom severity questions would yield similar results as a nonscannable form. There were three goals of the pilot study: first, to observe if, in a questionnaire using two forms of Visual Analog Scares (VAS), questions using a scannable sequence of "boxes" for responses would elicit different mean responses than questions using unbroken "lines"; second, to observe if questions using a sequence of "boxes" would elicit different within person and week response variability than questions using unbroken "lines"; and third, to estimate the sample size needed for a crossover study, depending on the particular form of the question used, and the number of crossovers. The pilot study, consisting of three sequential weekly questionnaires, provided week, subject, and error variance components for each of three dependent variables from the two different VAS forms. Most of the calculations were performed with a log transformation of the data. For each VAS form, the number of subjects necessary for desired study power for each symptom was calculated. Based on this pilot study, neither the mean nor the within person and week variance component was consistently larger or smaller for the VAS(box) form than for the VAS(line) form. The linear models analysis showed that the two forms filled out by the same person on the same day had similar mean Values and were highly correlated for all symptoms (R(2)greater than or equal to 0.95). Thus we chose the VAS(box) form because of scanner compatibility and estimated the required number of subjects for our full-scale study based on this chosen form.