| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Letters to the Editor |
1 Research & Development
2 and Central Laboratory, Centro Diagnostico Italiano, Milan, Italy
3 Lombardia Cancer Registry, and Environmental Epidemiology, National Cancer Institute, Milan, Italy
4 Università degli Studi di Milano, Department of Chemistry, Biochemistry and Biotechnologies, for Medicine, Milan, Italy
bAddress correspondence to this author at: Università degli Studi di Milano. Dipartimento di Chimica, Biochimica e Biotecnologie per la Medicina. Via Saldini 50, 20133 Milan, Italy. Fax 39-02-5031-6059; e-mail carlo.franzini{at}unimi.it.
To the Editor:
We appreciate the interest of colleagues in our approach to the definition of reference values (1). They point out the complexity of our suggested statistical procedure compared with a simpler approach, and they highlight differences between our and their reference intervals for serum thyrotropin (TSH), with special reference to the upper reference limit (URL).
Concerning the first point, we recognized in our report that the application of our procedure needs the cooperation of an expert statistician and, of course, the availability of a large computerized database. However, our approach aimed to define the reference limits for several quantities commonly measured in serum (actually 36), based on the large nondiseased population of individuals referred to our center, without having detailed clinical information on each individual. To achieve this, we developed a method based on a specified rationale, including a multistep selection of nondiseased individuals, followed by statistical, nonparametric definition of percentile-based limits. This is not simple, of course, but it has a rational basis: simpler approaches are not applicable. By contrast, Giavarina et al. do not provide any rationale for their simple approach. Looking at their graphs, we presume that they locate the URL at the local minimum on the right side of the major peak of the distribution, where it appears as the cross-point of 2 underlying subdistributions (are these healthy and diseased individuals?). The method does not locate exactly the percentile-related URL. This criterion is hardly related to the theory of reference values—more complicated patterns might be very difficult to explore; therefore, we have concerns about the possibility of generalizing this technique. Moreover, the method is intrinsically univariate, wasting highly relevant information provided by other available quantities, as we showed with our correlation analysis.
Concerning the second point, it must be considered that, together with several variables (e.g., ethnic group and environment), sex and age distributions are very important factors in determining the population’s values (2). Actually, a simple linear regression of TSH values over age performed on our raw data shows slopes of 0.01229 (P <0.0001) for females and 0.02324 (P <0.0001) for males. To give just a crude interpretation of our regression data, for every year of increasing age, the TSH mean value increases by 0.01229 mIU/L in females and by 0.02324 mIU/L in males. Independently from age distribution, the different clinical settings should also be considered in comparing results. On one extreme there is a trial specifically aimed at estimating TSH reference limits (3), whereas on the opposite extreme there is our retrospective (a posteriori) analysis, based on opportunistic data. Results from settings so different can hardly be compared.
The report based on the National Health and Nutrition Examination Survey (NHANES) III study results (2) includes a long list of distribution parameters (2.5, 50, and 97.5 centiles) of TSH values by age class, sex, and ethnicity in 2 different populations: a disease-free population and a reference population (risk factors excluded). Leaving out ethnic group- and age-related differences (which are not insignificant), we found in the entire disease-free population that the 97.5 centile is 6.10 mIU/L for women and 4.81 mIU/L for men, whereas in the whole reference population the corresponding values are 4.09 and 4.12 mIU/L, respectively. These last values compare acceptably with our URL values (4.45 and 4.08 mIU/L, respectively), indirectly confirming the power of the multivariate algorithm in selecting the disease-free portion of the population. There is still controversy about the correct reference interval of serum TSH. A recent report (4) debated the need for a narrower TSH reference interval and recommended lowering the URL to 2.5 mIU/L. With this limit, the sensitivity of the test is increased, but its specificity is inevitably decreased.
References
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
