Utilization of Accuracy Profiles as a Tool for the Validation of Analytical Methods A. Ceccato, P. Jacobs, A. Flament, M. Gibella and W. Dewé Lilly Development Centre, B-1348 Mont-Saint-Guibert, Belgium 1. Introduction The objective of the validation is to demonstrate that the method is suitable for its intended use. The validation presented here is based on the strategy presented by Hubert et al. [1,2] and consists in 2 steps : prevalidation and validation. Prevalidation –Selection of the most suitable calibration model using an accuracy profile as a selection tool. Validation –Evaluation of the selectivity of the method. –Assessment of precision, trueness and accuracy at different concentration levels. –Evaluation of the linearity. –Determination of the limits of detection and quantification. The GC method was developed in order to determine impurities around the 0.1 % level in methanol, acetone, toluene and dichloromethane. 2. Analytical Conditions Column : CP-Sil-8CB Low/MS (30mx0.32mmx1.00 µm) Carrier gas : Helium Inlet Pressure : 10.0 psi (constant pressure mode) Injector T° : 280°C Detector T° : 320°C H2 Flow : 30 ml/min Air flow : 400 ml/min Make up flow : 25 ml/min Injection volume : 0.2 µl Split ratio : 50/1 T° program :35°C for 10 min, then raised to 120°C at 40°C/min, then 120°C for 8 min then to 300°C at 20°C/min. Concentration range, from 0.01 % to 2.0 % V/V Chromatographic separation of all compounds investigated. (1) methanol – (2) ethanol – (3) acetone – (4) isopropanol – (5) dichloromethane – (6) proprionitrile (IS) – (7) MEK – (8) ethyl acetate – (9) chloroform – (10) benzene – (11) toluene – (x) endogenous compound from ethylacetate 4. Pre-validation The accuracy profile is used as a tool to decide the capability of the method to give results inside the acceptance limits : it is obtained by linking on one hand the lower and on the other hand the upper bounds of the  -expectaton tolerance tolerance limits at each concentration level. The regression model can be used if the  -expectation tolerance interval (blue) stays within a predefined acceptance criteria (i.e. 10 %, black dotted lines) since it guarantees that the method will be able to give a result within the  -expectation tolerance interval 95 times out 100 experiments. All calculations were perfromed on Benzene using the e-noval ® software from Arlenda. 3. Selectivity Weighted Linear Regression Quadratic Regression Linear Regression Weighted Linear Regression after logarithm transformation Linear Regression Through 0 Fitted using the highest level only Weighted Quadratic Regression Weighted Linear Regression after square root transformation Linear Regression after square root transformation Linear Regression after logarithm transformation Linear Regression Through 0 Fitted using the level (3) only 5. Validation The response functions are determined using the selected model : linear regression through 0 using one calibration level. The precision of the method is evaluated by measuring repeatability and intermediate precision. The trueness is evaluated by calculating the relative bias (%) at each concentration level. Accuracy is assessed from the accuracy profiles. The accuracy profiles of benzene and ethanol show that the determination of benzene is accurate over the whole concentration range while the determination of ethanol is only accurate between 0.05 and 2.0 % (50 and 2000 µl/100 ml). The linearity is established by fitting the regression line between the back-calculated concentrations versus the introduced concentrations. The Lower (LLOQ) and Upper Limits of Quantitation (ULOQ) are obtained by calculating the smallest and the highest concentrations beyond which the  -expectation tolerance limits (dotted blue lines) go outside the acceptance limits (dotted black lines). 7. References Ph. Hubert et al., STP Pharma Pratiques 13, 101 (2003) Ph. Hubert et al., J. Pharm. Biomed. Anal., 36, 579 (2004) 6. Conclusion A GC method for the determination of impurities in methanol, acetone, toluene and dichloromethane was validated using a new approach based on the accuracy profile determination and was found to meet the requirements for a further investigation of traces impurities in solvents. The accuracy profiles are used as a tool to select the best regression model. A graphic visualization is constructed in order to help the operator to decide if the method is suitable to do what it is supposed to do. The accuracy prodiles are also used to determine the accuracy of the method : if the lines corresponding to the  -expectation tolerance interval stay within the predefined acceptance limits, the method can be declared as valid. Ethanol Benzene