Feasibility of using real-world free thyroxine data from the US and Europe to enable fast and efficient transfer of reference intervals from one population to another

Abstract

Objectives: The direct approach for determining reference intervals (RIs) is not always practical. This study aimed to generate evidence that a real-world data (RWD) approach could be applied to transfer free thyroxine RIs determined in one population to a second population, presenting an alternative to performing multiple RI determinations.

Design and methods: Two datasets (US, n = 10,000; Europe, n = 10,000) were created from existing RWD. Descriptive statistics, density plots and cumulative distributions were produced for each data set and comparisons made. Cumulative probabilities at the lower and upper limits of the RIs were identified using an empirical cumulative distribution function. According to these probabilities, estimated percentiles for each dataset and estimated differences between the two sets of percentiles were obtained by case resampling bootstrapping. The estimated differences were then evaluated against a pre-determined acceptance criterion of ≤7.8% (inter-individual biological variability). The direct approach was used to validate the RWD approach.

Results: The RWD approach provided similar descriptive statistics for both populations (mean: US = 16.1 pmol/L, Europe = 16.4 pmol/L; median: US = 15.4 pmol/L, Europe = 15.8 pmol/L). Differences between the estimated percentiles at the upper and lower limits of the RIs fulfilled the pre-determined acceptance criterion and the density plots and cumulative distributions demonstrated population homogeneity. Similar RI distributions were observed using the direct approach.

Conclusions: This study provides evidence that a RWD approach can be used to transfer RIs determined in one population to another.

Keywords: Direct approach; Free thyroxine; Indirect approach; Real-world data; Reference intervals.

Graphical abstract

Fig. 1

FT4 data from existing RWD sources according to site. A) descriptive statistics of FT4 values (US, n = 10,000; Europe, n = 10,000 samples); B) absolute estimated FT4 values (US, n = 1000; Europe, n = 1000 bootstrapped samples); C) density plots showing distribution of FT4 values (US, n = 10,000; Europe, n = 10,000 samples); D) cumulative distributions of FT4 values (US, n = 10,000; Europe, n = 10,000 samples). ^aThe minimum value measured in the European dataset has been rounded up from 0.499 pmol/L to 0.5 pmol/L to reflect the lower end of the measuring range of the Elecsys FT4 III, which is 0.5 pmol/L [[16], [17]]. The minimum value measured in the US dataset was 1.3 pmol/L, which is the lower end of the measuring range of the Elecsys FT4 II [18]. ^bThe maximum values measured in the European and the US datasets have been rounded down from 101 pmol/L to 100 pmol/L to reflect the upper end of the measuring ranges of the Elecsys FT4 II and the Elecsys FT4 III, both of which are 100 pmol/L [[16], [17], [18]]. ^cThe pre-determined acceptance criterion only applied to the 9.76th and the 93.03rd percentiles; the percentiles closest to the lower and upper end of the RI. CI, confidence interval; FT4, free thyroxine; NA, not applicable; RI, reference interval; RWD, real-world data; SD, standard deviation; US, United States.

Fig. 2

FT4 concentrations measured in samples from the US (n = 150) and Europe (n = 150) using the Elecsys FT4 IV assay. A) descriptive statistics and percentiles for FT4 measurements; B) boxplots showing distribution of FT4 measurements. CI, confidence interval; CV, coefficient of variation; FT4, free thyroxine; SD, standard deviation; US, United States.