Comparison of reference intervals derived by direct and indirect methods based on compatible datasets obtained in...

Area of Science:

• Clinical Chemistry
• Laboratory Medicine
• Biostatistics

Background:

• Reference intervals (RIs) are crucial for interpreting laboratory test results.
• Indirect derivation of RIs from laboratory information system (LIS) data is an emerging approach.
• Evaluating the accuracy of indirect RIs against directly established RIs is necessary.

Purpose of the Study:

• To assess the accuracy of indirectly predicted RIs compared to directly established RIs.
• To evaluate four indirect univariate methods (Hoffmann, Bhattacharya, Arzideh, Wosniok) for RI prediction.
• To analyze prediction biases in 25 major chemistry analytes using Turkish nationwide data.

Main Methods:

• Retrieved and cleaned LIS data from outpatients aged 18-65, with one record per patient per year.
• Applied four indirect univariate methods, including power transformation with IFCC or predicted lambda values.
• Compared indirectly predicted RIs with RIs established directly from healthy subjects.

Main Results:

• LIS data exhibited peak location and shape alterations compared to direct study data.
• Indirect methods resulted in lowered or raised RI limits for various analytes (e.g., sodium, albumin, triglyceride).
• Overall, 72% of predicted RI limits showed significant biases; this reduced to 47% after excluding age-biased results.

Conclusions:

• Indirectly derived RIs can exhibit significant biases compared to direct RIs.
• Age bias in LIS data contributes to prediction inaccuracies.
• More rigorous data cleaning is essential to minimize biases in indirectly predicted reference intervals.