A study on reference interval transference via linear regression.

Background: Reference intervals (RIs) transference can expand the applicability of established RIs. However, the study on transference methodology is insufficient, and RIs validation based on small samples cannot adequately identify transferred risk under complex situations. This study aimed to find appropriate conditions to ensure the effect of transference. Methods: We established the RIs of Roche and Beckman systems for 27 analytes based on 681 healthy individuals. Roche RIs were converted into the Beckman RIs using linear regression (least squares method) which is divided into two methods - Method(ref) (500 test numbers with relatively narrow data range) and Method(ep) (80 test numbers with relatively wide data range). Taking the RIs established by Beckman results as standard, we assessed the accuracy, precision and trueness of transferred results under various conditions. Results: A total of 29.6% and 48.1% of analytes were consistent between the two systems for the lower and upper reference limits, respectively. The concordance rates between transferred and measured RIs for Method(ref) were up to 74.1% and 92.6%, which were better than Method(ep) (4/1.4% and 59.3%). The CV of transferred reference limits decreased gradually with increasing test number under the same data range. For most analytes, excluding some electrolyte tests, we could obtain accurate results when r > 0.800 and the test number was sufficient regardless of the regression equation types. Conclusions: Transferability of RIs is affected by many factors, such as correlation, test number, regression equation type, and quality requirement. To reduce the risk of transference, it is very important to select right method with reasonable conditions.