Sample size requirement for comparison of decompression outcomes using ultrasonically detected venous gas emboli (VGE): power calculations using Monte Carlo resampling from real data.
INTRODUCTION: In studies of decompression procedures, ultrasonically detected venous gas emboli (VGE) are commonly used as a surrogate outcome if decompression sickness (DCS) is unlikely to be observed. There is substantial variability in observed VGE grades, and studies should be designed with sufficient power to detect an important effect. METHODS: Data for estimating sample size requirements for studies using VGE as an outcome is provided by a comparison of two decompression schedules that found corresponding differences in DCS incidence (3/192 [DCS/dives] vs. 10/198) and median maximum VGE grade (2 vs. 3, P < 0.0001, Wilcoxon test). Sixty-two subjects dived each schedule at least once, accounting for 183 and 180 man-dives on each schedule. From these data, the frequency with which 10,000 randomly resampled, paired samples of maximum VGE grade were significantly different (paired Wilcoxon test, one-sided P ⋜ 0.05 or 0.025) in the same direction as the VGE grades of the full data set were counted (estimated power). Resampling was also used to estimate power of a Bayesian method that ranks two samples based on DCS risks estimated from the VGE grades. RESULTS: Paired sample sizes of 50 subjects yielded about 80% power, but the power dropped to less than 50% with fewer than 30 subjects. CONCLUSIONS: Comparisons of VGE grades that fail to find a difference between paired sample sizes of 30 or fewer must be interpreted cautiously. Studies can be considered well powered if the sample size is 50 even if only a one-grade difference in median VGE grade is of interest.
Doolette, DJ; Gault, KA; Gutvik, CR
Volume / Issue
Start / End Page
Pubmed Central ID
Electronic International Standard Serial Number (EISSN)
International Standard Serial Number (ISSN)