How should the repeatability of clinical measurements be analysed? An assessment of analysis techniques with data from cardiovascular autonomic function tests.

A reliable clinical test should give similar results when repeated on the same patient. Repeatability data, however, can be presented in several ways, giving apparently different results. We compared two approaches--the coefficient of variation, and the SD ratio (SD of repeat measurements/SD of subject means). Repeat measurements of two published studies of cardiovascular autonomic function response were analysed. The test measurements were divided into two groups: group A tests, for which no response results in a zero output (includes measurements of differences or changes); and group B tests, for which no response results in a unity or other non-zero output (includes measurements of ratios). Data were obtained from 10 normal and 25 diabetic subjects. There were significant differences between the coefficients of variation of the two groups of tests for the normal subjects (p < 0.01) and for the diabetic subjects (p < 0.01). Low coefficients of variation were more likely to belong to group B (ratio measurements) than to be an estimate of good repeatability. The calculation of the coefficient of variation was adapted to ensure that the test calculation tended to zero with low autonomic function. There was a strong relationship, especially with the diabetic subjects, between repeatability calculated from the adapted coefficient of variation and from the SD ratio (normal subjects, r = 0.63; diabetic subjects, r = 0.91). The calculation of coefficient of variation is often misused, and low coefficients of variation presented in the literature may not represent good repeatability of a clinical test.