On the calculation of reference change values, with examples from a long-term study.

Reference change values (sometimes called critical differences) indicate statistically important changes between test values obtained on two occasions. They are commonly computed from the median (or mean) within-subject variance observed in repeated test measurements on a number of subjects. With this computational approach, all observed within-subject variances are assumed to be estimates of a constant true variance, the same for all individuals. Moreover, any possible correlation between successive values is almost always ignored. This simplified methodology differs from the method originally proposed for computing reference change values, which accounts for variability in true variances and for serial correlation. From data obtained from repeated measurements over 2 to 5 years in 72 physically healthy subjects, we computed and compared reference change values in 18 serum analytes, using the simplified method and the originally proposed procedure. Although the original method is more complicated and requires a computer program, we believe that it produces more-reliable reference change values than those obtained by the simplified approach. The former are generally larger, but remain sensitive to clinically important changes in the individual.