Authors (including presenting author) :
Chan PL (1), Lam CW (1)(2)
Affiliation :
(1) Division of Chemical Pathology, Department of Pathology, Queen Mary Hospital, (2) Department of Pathology, Li Ka Shing Faculty of Medicine, University of Hong Kong
Introduction :
Reference intervals (RIs) are essential for clinical laboratories to interpret biomarker levels. Traditional RI establishment requires extensive sampling from healthy populations, which poses significant challenges due to ethical constraints and resource limitations. With advances in Artificial Intelligence (AI), the automated identification of ‘normal’ and ‘abnormal’ distributions from mixed patient data can now be achieved using data mining techniques. Indirect RI estimation has emerged as a powerful alternative, allowing the establishment of RIs through the analysis of existing clinical laboratory data.
Objectives :
(1) To validate an advanced indirect RI algorithm (refineR) that uses routine laboratory data to establish RIs; (2) To explore continuous RIs that provide smooth and dynamic boundaries across age ranges, allowing for more precise and individualized clinical interpretations compared to traditional discrete RIs
Methodology :
Plasma serine levels were used as an example to illustrate the process of indirect RI establishment. Data were collected from QMH Chemical Pathology between 2009 and 2023, comprising 3253 samples. The refineR algorithm was used to evaluate indirect RIs. For continuous RI establishment, a regression curve of log-transformed age and concentration was fitted using the least squares method. The regression equation for the RI bounds was: ln(RI Bound) = β0 + β1⋅ln(Age) + β2⋅[ln(Age)]2 + k, where β0, β1, and β2 are coefficients obtained from the quadratic regression, and klwr and kupr are global adjustments for the lower and upper bounds respectively. Outliers were removed using the interquartile range (IQR) method. The adjusted RI bounds were then transformed back to the original scale, providing smooth and continuous RIs across age ranges.
Result & Outcome :
The indirect method (refineR) produced RIs that closely matched the traditional direct method, validating the indirect approach. Quadratic regression established continuous RIs with lower bound coefficients β0 = 4.2477, β1 = 0.130, and β2 = −0.040, and upper bound coefficients β0 = 5.495, β1 = −0.034, and β2 = −0.008. Global adjustments (????lwr = −0.147, ????upr = 0.061) refined the boundaries. Continuous RIs demonstrated the potential to replace age-stratified RIs by providing smoother and dynamic boundaries, offering the possibility of improving diagnostic accuracy in clinical practice.
