Impact of sample size and data origin on the simulation-based analytical performance specification derivation

Indirect clinical outcome-based analytical performance specifications (APS) may offer a feasible alternative to direct studies but raise questions regarding data requirements and transferability. This study aimed to determine the minimum sample size for simulation-driven APS and to investigate the impact of data origin.

We analysed data for six laboratory measurands from four distinct datasets: three Turkish hospitals (Istanbul, Ankara, Van) and the US National Health and Nutrition Examination Survey (NHANES) 2017-2020 dataset. Using the 'APS Calculator', we determined APS for measurement uncertainty (MU). The effect of sample size was evaluated by subsampling the Istanbul dataset (n=50,000 to n=500) and calculating the Mean Absolute Percentage Error (MAPE) relative to the baseline (n=50,000) (acceptable <2 %). Precision of APS estimate was evaluated using nonparametric 95 % confidence intervals from simulation results and summarized as relative margin of error (RME; acceptable <10 %). Data-origin effects were assessed by Pearson correlation analysis of APS for MU across the four datasets.

A sample size of n=5,000 met stability and precision criteria (MAPE 1.99 %; mean RME 6.86 %), whereas n=2,000 exceeded thresholds (MAPE 2.32 %; mean RME 10.67 %). APS for MU were highly consistent across datasets (Pearson r 0.977-0.994), with the main divergence observed for total folate in NHANES.

A minimum sample size of 5,000 results appears sufficient for a reliable simulation-driven APS determination. APS were largely transferable across populations when decision limits and agreement targets were constant, with caution for analytes influenced by population exposure and method selectivity.