Indirect Methods for TSH Reference Interval: At Last Fit for Purpose?The Author’s ReplyThe Author’s Reply

Abstract

To the Editor We read with great interest the article by Katayev et al1 and the companion editorial by Horowitz2 that discuss the production of reliable reference intervals for common tests such as calcium, creatinine, mean corpuscular volume, and thyroid-stimulating hormone (TSH). Katayev et al1 used a computerized version of the technique proposed almost 50 years ago by Hoffmann3 in a large number of results stored in the laboratory information system serving 6 laboratories. According to Katayev et al,1 “The computerized Hoffmann method for the indirect determination of RIs [reference intervals] produced intervals that were remarkably similar to peer-reviewed RIs.” Horowitz2 disagrees and states “… the reference intervals generated in this way are strikingly different from the reference intervals in use….” The articles disagree because TSH is being measured more and more precisely, but different assays show a relevant bias; manufacturers, laboratory professionals, and clinicians rarely appreciate the effect of this bias on the cutoffs quoted in guidelines. In some cases, we agree that TSH could indeed be unfit for purpose.4 Katayev et al1 reported upper reference limits in 2 large sets of data using the ADVIA Centaur analyzer (Siemens Medical Solutions Diagnostics, Tarrytown, NY) of 3.05 and 3.19 mIU/L consistent with that proposed by the American Association of Clinical Endocrinologists (3 mIU/L).5 Since 2000, we have been using the same analyzer and the program GraphROC, which implements the “indirect” method proposed by Kairisto and Poola6 based on the Hoffmann method.7 In synthesis, the distribution is split and the mode (rather than the mean) of the hypothesized health-related distribution is forced to be the same as the mode in the original distribution. The health-related distribution consisted of 2 halves of 2 different gaussian distributions, with the same mode and …