Search

Objectives: To describe the application of a data-mining statistical algorithm for calculation of clinical laboratory tests reference intervals. Methods: Reference intervals for eight different analytes and different age and sex groups (a total of 11 separate reference intervals) for tests that are unlikely to be ordered during routine screening of disease-free populations were calculated using the modified algorithm for data mining of test results stored in the laboratory database and compared with published peer-reviewed studies that used direct sampling. The selection of analytes was based on the predefined criteria that include comparability of analytical methods with a statistically significant number of observations. Results: Of the 11 calculated reference intervals, having upper and lower limits for each, 21 of 22 reference interval limits were not statistically different from the reference studies. Conclusions: The presented statistical algorithm is shown to be an accurate and practical tool for reference interval calculations.

To the Editor We applaud the work by Katayev and colleagues1 reflected in their article on data mining for reference intervals. The development of robust automated tools to perform data mining in a standardized manner represents a significant step forward in the field. We would like to point out that the authors’ application of the Hoffmann2 method seems to deviate from the original in one important aspect. The original Hoffmann method involved manually plotting data on normal probability paper. When the cumulative frequency distribution of a Gaussian data set is plotted on this type of paper, a straight line is obtained. As Hoffmann pointed out, even when a data set includes a significant minority of data from a …