Establishing personalized reference intervals based on within-subject biological variation using real-world data and...

This study aimed to establish personalized reference intervals (prRIs) for creatinine, glucose and γ-glutamyl transferase (GGT) using routine laboratory data. The impact of different indirect methods for estimating within-subject biological variation (CV_I) on prRI width and its ratio to population-based reference intervals (popRIs) was evaluated.

A five-year retrospective analysis was conducted on two cohorts: a general adult population and a presumed healthy subgroup. CV_I was estimated from result ratio (RR) distributions between consecutive test results using the refineR algorithm in both groups, and also using an alternative indirect approach in the healthy subgroup. prRIs were calculated from serial test results, adjusted for analytical variation (CV_A), and stratified by age, sex, estimation method, and EFLM reference values.

In the general population, refineR estimated CV_I at 8.6% for creatinine, 7.5% for glucose, and 12.9% for GGT. In the presumed healthy subgroup, CV_I estimates obtained using refineR and the alternative indirect approach were 6.2% and 7.0% for creatinine, 5.0% and 5.1% for glucose, and 14.2% and 15.9% for GGT, respectively. All prRIs were narrower than popRIs.

prRIs can be derived from routine laboratory data using indirect approaches. The chosen CV_I estimation strategy influences the prRI width. CV_I estimation performance appeared to depend on dataset composition and RR distribution. Simpler statistical approaches yielded comparable results for relatively symmetric RR distributions, whereas refineR-based modeling appeared more useful for highly skewed distributions.