Different percentages of false-positive results obtained using five methods for the calculation of reference change...

Background Reference change values provide objective tools to assess the significance of a change in two consecutive results for a biomarker from an individual. The reference change value calculation is based on the assumption that within-subject biological variation has random fluctuation around a homeostatic set point that follows a normal (Gaussian) distribution. This set point (or baseline in steady-state) should be estimated from a set of previous samples, but, in practice, decisions based on reference change value are often based on only two consecutive results. The original reference change value was based on standard deviations according to the assumption of normality, but was soon changed to coefficients of variation (CV) in the formula (reference change value = ± Z ċ 2^½ ċ CV). Z is being dependent on the desired probability of significance, which also defines the percentages of false-positive results. The aim of this study was to investigate false-positive results using five different published methods for calculation of reference change value. Methods The five reference change value methods were examined using normally and ln-normally distributed simulated data. Results One method performed best in approaching the theoretical false-positive percentages on normally distributed data and another method performed best on ln-normally distributed data. The commonly used reference change value method based on two results (without use of estimated set point) performed worst both on normally distributed and ln-normally distributed data. Conclusions The optimal choice of method to calculate reference change value limits requires knowledge of the distribution of data (normal or ln-normal) and, if possible, knowledge of the homeostatic set point.