|
Abstract INTRODUCTION:
Review of chromatographic LC-MS/MS results has historically relied on skilled scientists to interpret the peak shape and peak integration of every chromatogram to certify both are of sufficient quality to substantiate the quantitative result produced. This review process tends to be qualitative in nature due to subjectivity and training, making it highly susceptible to inter-person variability. Acknowledging this, some regulations require review of all chromatograms by two individuals to corroborate the chromatographic “ball” or “strike” call (i.e., 200% data review). A decade ago, our clinical mass spectrometry laboratory adopted the use of ASCENT (Indigo BioAutomation) which not only applies a sophisticated peak modeling algorithm for robust peak detection/integration, but also supports a rule-based flagging system to facilitate more objective determination of chromatographic successes. This largely obviates the need for secondary review.
However, we are now aiming to automate the majority of chromatographic calls in a review-by-exception (autoverification) workflow as the lab transitions to ASCENT v4 which offers both an enhanced peak selection/integration and a deliberate review-by-exception user interface. To that end, we will present our process for developing and validating the robust autointegration settings which provide accurate and precise quantitative measurements. We also present our framework of chromatographic flagging rules designed to autodetect unacceptable chromatography/integration and validate that the error rate achieved by these rules is acceptably low.
METHODS:
Two hundred batches acquired over the past 12 months and analyzed in ASCENT v3 were randomly selected and split (4:1) into ‘training’ (N = 160 batches) and ‘testing’ (N = 40 batches) groups for ASCENT v4. Of the 160 training batches, a sub-set of 20 ‘tuning’ batches were randomly selected to optimize the autointegration settings. Subsequently, all training batches were analyzed using these tuned settings to establish the acceptable range for chromatographic features (peak width, asymmetry, retention time, ion ratio, etc.). Observations outside that range are identified with rule-based flagging as potential missed integration or poor chromatographic performance.
The 40 testing batches were first used to validate the precision and accuracy in ASCENT using the optimized autointegration settings. Imprecision across the AMR was demonstrated using back-fit concentration of the calibration standards in each batch, as well as analysis of quality control specimens. Quantitative accuracy was demonstrated by comparison to the previously certified results. Finally, autoverification of chromatographic peak selection/integration in ASCENT using the ‘trained’ flagging rules was validated by comparison of the automated flags to manual flags (annotated via 200% blinded manual review).
RESULTS:
Converting from a review-with-assistance workflow to a review-by-exception (autoverification) workflow was piloted using our existing assay for Vitamin B1 in whole blood. Tuning the autointegration settings was guided by comparison to observed retention times, peak areas and quantitative results from 20 tuning batches. Robust peak selection for the analyte quantifier and qualifier chromatograms was achieved by referencing the internal standard quantifier retention time within a fixed tolerance. Otherwise, default settings for peak integration using the exponentially modified gaussian peak model were sufficient when applying dynamic peak detection thresholds referencing the lowest calibrator.
After tuning, comparison of results observed in ASCENT v4 to certified results in ASCENT v3 produced a Deming regression slope of 1.0052, correlation coefficient of 0.9898, and mean bias of +0.05% (N=3173). Over 99.8% of the results demonstrated quantitative agreement within 10% and >99.7% also provided qualitative agreement based on the normal reference interval. Manual interrogation of discordant outliers revealed poor chromatography and misleading autointegration of the internal standard that is readily flagged.
Over 30 custom rules were developed for autoverification of the results. Autoverification rules were categorized as 1) batch, 2) chromatographic, and 3) reporting. Batch rules were designed for autoverification of calibration curves, quality controls, and other pass/fail QA metrics such as contamination or internal standard response variance. Chromatographic rules were designed to autoverify peak selection/integration and permits the identified samples to be manually adjudicated (i.e., review-by-exception). Otherwise, autoverification of subsequent reporting decisions were fully automated based on objective information.
Thresholds for 15 chromatographic rules were established empirically using a custom RShiny script to identify normal ranges for various characteristics related to peak shape and modeling. In the majority of cases, these characteristics were heteroskedastic and thresholds were bifurcated by response or concentration. Under the premise that outliers would be most susceptible to missed integration, thresholds for each rule were initially set to flag 1% of samples.
DISCUSSION:
Missed integrations represent the major risk to autoverification of LC-MS/MS results. We have developed a suite of rules to autoverify the chromatographic peak selection/integration in ASCENT v4. Our workflow for development of these rules and their validation demonstrates an error rate (i.e., unflagged missed integrations) of <0.1%. Results will be discussed in the context of current, relevant guidelines including CLSI C62 and AUTO15.
|
= Discovery stage. (53.14%, 2025)
= Translation stage. (22.33%, 2025)
= Clinically available. (24.53%, 2025)
