Development and evaluation of automated screening algorithms for pre-analytical errors in pharmacokinetic data: a...

Area of Science:

• Pharmacokinetics and Pharmacometrics
• Clinical Pharmacology
• Data Quality Assurance

Background:

• Pre-analytical errors, such as sample mix-ups, can severely compromise pharmacokinetic (PK) data integrity.
• Ensuring the accuracy of PK data is crucial for reliable clinical pharmacology research and drug development.
• Existing methods for detecting pre-analytical errors in PK datasets may lack sufficient sensitivity or specificity.

Purpose of the Study:

• To develop and validate automated methods for detecting pre-analytical errors in simulated pharmacokinetic data.
• To compare the performance of a Mahalanobis distance-based method against a traditional run-test method.
• To assess the robustness of these methods under various challenging conditions, including different PK models and data imperfections.

Main Methods:

• Simulated vancomycin PK data were generated and intentionally corrupted with two types of pre-analytical errors: time point swaps (TS) and concentration swaps (CS).
• Two automated detection approaches were developed: a run-test non-parametric method and a Mahalanobis distance-based method employing leave-one-out cross-validation (LOOCV).
• Performance was evaluated on 12,500 simulated PK profiles, comparing error detection rates, specificity, precision, recall, and area under the ROC curve (AUC).

Main Results:

• The Mahalanobis distance-based method demonstrated a significantly higher error detection rate (79.7%) compared to the run-test method (59.6%, p < 0.001).
• The distance-based method achieved a superior AUC (0.856 vs. 0.574) and better recall for TS errors (80.4% vs. 71.9%) and precision for CS errors (86.1% vs. 59.8%).
• Detection performance remained robust across alternative population PK models, a different drug (theophylline), contaminated training sets, missing data, and sampling time variations.

Conclusions:

• The Mahalanobis distance-based method with LOOCV is a more effective tool for identifying pre-analytical errors in PK datasets than the conventional run-test method.
• This distance-based approach offers a practical and reliable quality control solution for enhancing the integrity of pharmacokinetic data in research.
• Implementing such automated detection tools is vital for ensuring the accuracy and validity of findings in clinical pharmacology.