Estimating Recovery in the Liquid–Liquid Extraction Chemical Space

Chemical characterization studies for medical devices rely on extract preparation techniques such as liquid–liquid extraction to render the extract more amenable to instrumental analysis. Considering these studies are generally non-targeted, the effect of extract preparation on the full range of potential analytes must be considered. However, there is currently no workflow for evaluating the impact of extract preparation on non-targeted extractables. Herein, we present a framework for approaching this problem by defining the applicable chemical space, selecting a model that appropriately predicts recovery from the extract preparation method, selecting example chemicals with a range of expected recoveries, and experimentally measuring the recoveries to verify the model performance under laboratory conditions. The framework is demonstrated for liquid–liquid extraction, with recovery demonstrably dependent on extractable pKa and distribution coefficient. Method parameters including pH, volume ratios, and extraction iterations were also considered critical factors in determining recovery. The analytical method was subsequently verified by incorporating deliberate deviations in the critical parameters as part of robustness determination for the method using a central composite design. Overall, the method parameters selected in this work resulted in coverage of approximately 85% of the selected chemical space. Further, for the various permutations evaluated, the model was found to predict recovery with a root-mean-square error of 19%. A result of this approach is increased clarity in the use of surrogate standards for evaluation of extract preparation methods, indicating that they should be chosen based on the variables included in the predictive model. Additionally, the verified model can be applied to the selected chemical space so that estimations can be made about which relevant analytes are likely to be poorly recovered during extract preparation. This framework is generally applicable to understand the effect of extract preparation on non-targeted analysis and how these considerations can be integrated into method development and validation.