A data-driven approach for the detection of internal standard outliers in targeted LC-MS/MS assays.

Heavy-labelled internal standard (IS) compounds are commonly used in liquid chromatography-tandem mass spectrometry (LC-MS/MS) assays to control for stochastic and systematic variation. Identifying samples that suffer from unwanted variation is critically important in order to avoid factitiously inaccurate results. Current approaches for outlier detection typically employ arbitrary thresholds and ignore systematic drift. To improve this, we applied robust linear mixed-effects models (LMMs) to capture the within- and between-run variability in IS signal and generate data-driven acceptance ranges for routine use. Data from in-house LC-MS/MS assays for 25-hydroxyvitamin D and D and prednisolone were retrospectively collected. The variation in the percentage deviation of the internal standard area from the mean of the calibrators was modelled through the use of robust LMMs. The fitted LMMs revealed significant positive drift in IS signal over the analytical runs for vitamin D, with slope coefficients of 0.118 (95% CI: 0.098, 0.138) and 0.192 (0.168, 0.215) for D and D, respectively. In contrast, the models for prednisolone demonstrated a significant negative drift in IS signal, with a slope coefficient of -0.164 (-0.297, -0.036). Non-parametric, cluster bootstrap resampling enabled us to define acceptance ranges for use in future assays. Here, we have described a computational approach to extensively characterise the variation in IS signal in routinely-performed LC-MS/MS assays. This approach facilitates a robust quality assessment of IS outliers in routine practice and thus has the potential to improve patient safety. Importantly, this approach is applicable to other MS assays where linear variation in IS signal is observed.