W. Wegscheider Validation: an Example Validation: an Example Learning Objectives: Student is able to inquire in sufficient depth on the purpose of analytical work is able to translate this information into stepwise strategy keeping in mind suitable performance characteristics at each step can devise experiments to test against these criteria performs statistical analysis of obtained data with the goal to make a valid judgement regarding the suitability of the method B. Neidhart, W. Wegscheider (Eds.): Quality in Chemical Measurements © Springer-Verlag Berlin Heidelberg 2000 1 W. Wegscheider Validation: an Example Validation Must Be Performance Based Definition of the need of the client translated to analytical quality criteria including cost and time Compilation of candidate methods Critical appraisal of candidate methods with respect to expected method performance Design and execution of validation study Evaluation of results, go-ahead or modifications or alternative method Measure the unknown with due regard to previously established quality criteria B. Neidhart, W. Wegscheider (Eds.): Quality in Chemical Measurements © Springer-Verlag Berlin Heidelberg 2000 2 W. Wegscheider Validation: an Example Worked Example: Fe2O3 in MgO For certain industrial applications of MgO the upper limit of Fe oxide needs to be controlled very tightly Definition of the need of the customer translated to analytical quality criteria including cost and time In this application an upper limit of 0.5 % was guaranteed by the supplier, the customer found 0.64 %, the supplier 0.41 %. B. Neidhart, W. Wegscheider (Eds.): Quality in Chemical Measurements © Springer-Verlag Berlin Heidelberg 2000 3 W. Wegscheider 4 Validation: an Example Deriving Minimum Performance Characteristics from Requirements As the differences from one another was about 0.1 % Fe2O3 the proposed method needs to have the following characteristics: overall uncertainty (including sampling): • uncertainty at required level of concentration from analysis itself: limit of quantitation (with matrix): limit of detection (with matrix): selectivity/interferences: linearity: B. Neidhart, W. Wegscheider (Eds.): Quality in Chemical Measurements © Springer-Verlag Berlin Heidelberg 2000 <0.03 % ~0.01% <0.03 % <0.01 % CRM at least 0.3-0.7 % W. Wegscheider 5 Validation: an Example Preliminary Examination of Candidate Methods (1) Method uncertainty limit of (no sampl.) quantitation required (%) 0.01 0.03 limit of linearity detection 0.01 0.3-0.7 selectivity SRM gravimetry ? ? titrimetry ? ? ? ? graphite furnace AAS flame AAS ? WD-XRF ICP-Q-MS ? ? B. Neidhart, W. Wegscheider (Eds.): Quality in Chemical Measurements © Springer-Verlag Berlin Heidelberg 2000 W. Wegscheider 6 Validation: an Example Preliminary Examination of Candidate Methods (2) Method cost gravimetry expertise in currently equipment time to laboratory practiced ready deliver ? ? no titrimetry ? graphite furnace AAS flame AAS ? WD-XRF ? ? ICP-Q-MS ? ? ? B. Neidhart, W. Wegscheider (Eds.): Quality in Chemical Measurements © Springer-Verlag Berlin Heidelberg 2000 ? W. Wegscheider Validation: an Example 7 Validation Plan Produce calibration curves for Fe by flame AA under standard conditions with and without MgO Compare calibration functions Evaluate limit of detection and limit of quantitation Estimate precision for meeting previously established requirements (+ margin): allowance for day to day variability, etc. Produce recovery data by mixing various aliquots of CRM MgO with high purity MgO Compare with alternative method ? Estimate (laboratory) sampling error for sample as received B. Neidhart, W. Wegscheider (Eds.): Quality in Chemical Measurements © Springer-Verlag Berlin Heidelberg 2000 W. Wegscheider 8 Validation: an Example Calibration without MgO 26.12.99 Description Procedure SOP new calibration without MgO flame AA Christof Calibration curve #measurements #replicates #conc. levels 25 5 5 x y1 y2 µg/mL abs. 0 0.5 1 1.5 2 0.0004 0.0175 0.0348 0.0503 0.063 0.0002 0.0172 0.0351 0.0497 0.0646 Working range Profile y3 0 0.0174 0.035 0.0505 0.0649 y4 0.0003 0.0177 0.0348 0.0506 0.0647 B. Neidhart, W. Wegscheider (Eds.): Quality in Chemical Measurements © Springer-Verlag Berlin Heidelberg 2000 0.5 2 y5 y6 0.0004 0.0174 0.0355 0.0502 0.0639 W. Wegscheider 9 Validation: an Example Calibration with MgO 29.12.99 SOP new calibration with MgO (200 mg/L) flame AA Christof Calibration curve #measurements 0 0.5 1 1.5 2 #replicates 0.0002 0.0167 0.0334 0,0502 0.0659 #conc. levels 0.0002 0.0163 0.0339 0,0492 0.066 Working range Profile 0.0000 0.0163 0.0331 0,0491 0.0661 0.0002 0.0161 0.0331 0,0493 0.0658 B. Neidhart, W. Wegscheider (Eds.): Quality in Chemical Measurements © Springer-Verlag Berlin Heidelberg 2000 0.0001 0.016 0.0333 0.0493 0.0655 W. Wegscheider 10 Validation: an Example Graphical Comparison without matrix with matrix calibration curve 1. order calibration curve 2. order 0,07 0,06 0,06 0,05 0,05 0,04 0,04 abs. abs. 0,07 0,03 0,03 0,02 0,02 0,01 0,01 0 0 -0,5 0 0,5 1 1,5 -0,5 2 0 2,5 0,5 µg/mL B. Neidhart, W. Wegscheider (Eds.): Quality in Chemical Measurements © Springer-Verlag Berlin Heidelberg 2000 1 mg/mL mg/l 1,5 2 2,5 W. Wegscheider Validation: an Example EXCEL Macro ValiData Basis of the program is ISO 8466 and DIN 32645 Homogeneity of variances test Calibration functions linear OR quadratic following Mandel´s test weighted linear and weighted quadratic standard additions Performance characteristics Decision limit and detection limit from calibration Limit of quantitation Linear range, sensitivity, confidence limits Recovery studies orthogonal regression robust regression B. Neidhart, W. Wegscheider (Eds.): Quality in Chemical Measurements © Springer-Verlag Berlin Heidelberg 2000 11 W. Wegscheider 12 Validation: an Example Numerical Comparison 2nd orde r ca libra tion function (y=a +b*x+c*x^2) without matrix with matrix Line a r ca libra tion function (y=a +b*x) Slope 0.032916 abs./(mg/L) CI* (slope) 0.032712646 0.033119353 abs./(mg/L) Ordinate intersect 9.6E-05 abs. CI (Ordinate intersect) -0.000153055 0.00034506 abs. Average(x) 1 mg/L Average(y) 0.033012 abs. Residual standard deviation 0.00034755 abs. Std. deviation of the method 0.01055870805 mg/L Coeff. of var. of the method 1.055870805 % t-value (95%) 2.068654794 Qx 12.5 (mg/L)^2 *Confidence interval B. Neidhart, W. Wegscheider (Eds.): Quality in Chemical Measurements © Springer-Verlag Berlin Heidelberg 2000 W. Wegscheider 13 Validation: an Example Limits of Detection Limits of Quantitation Decision and determination limit Blank read. meth. Decision levels (CI) 0.95 Decision limit 0.012155466 Detection limit 0.024310932 Determination limit Decision and determination limit Blank read. meth. Calibr. method Decision levels (CI) 0.95 0.95 Decision limit 0.006345772 0.010236787 mg/l Detection limit 0.012691545 0.020473573 mg/l 0.036786362 mg/l Determination limit B. Neidhart, W. Wegscheider (Eds.): Quality in Chemical Measurements © Springer-Verlag Berlin Heidelberg 2000 without matrix LOD 0.006 % with matrix LOD 0.005 % LOQ 0.018 % W. Wegscheider 14 Validation: an Example Estimation of Recovery serial dilution of Burned Magnesite SRM 104 calibration curve 1. order Fe content of 7.07+0.04 % (1 s) with pure MgO 0,7 between 0.1 and 0.6 % 0,6 1.015187226 found/(given) 0,5 0.994663053 1.035711399found/(given) -0.008726667 found 0,4 -0.016368629-0.001084705found found Slope IC (slope) Ordinate intersect IC (Ordinate intersect) 0,3 0,2 0,1 - 0.009 is about - 1.8 % at 0.5 % level 0 0 0,1 0,2 B. Neidhart, W. Wegscheider (Eds.): Quality in Chemical Measurements © Springer-Verlag Berlin Heidelberg 2000 0,3 0,4 given 0,5 0,6 W. Wegscheider Validation: an Example 15 Alternative Method: Titrimetry 1. dissolve SRM in HCl, dilute to 100 ml 2. reduce with SnCl2 3. add 20 ml HgCl2 and 50 ml Zimmermann-ReinhartSolution 4. titrate with 0.01 n KMnO4 Result: 7.01 % Fe2O3 Certified: 7.07 + 0.04 % B. Neidhart, W. Wegscheider (Eds.): Quality in Chemical Measurements © Springer-Verlag Berlin Heidelberg 2000 W. Wegscheider Validation: an Example 16 Estimation of Uncertainty from Sampling about 500 g laboratory sample delivered sampled “according to ISO” by customer max. concentration about 1 % Fe in dark pieces min. concentration below LOD in light pieces many larger grains: ca 1 cm3 corresponds to 3 g total of 500 g/3 g ~ 160 pieces 80 pieces give a relative uncertainty of 9 % at the 0.5 % level this is 0.045 % (0.02 % requ. !!!) CONCLUSION: the laboratory sample must be about 2000 g B. Neidhart, W. Wegscheider (Eds.): Quality in Chemical Measurements © Springer-Verlag Berlin Heidelberg 2000 W. Wegscheider Validation: an Example Content of Validation Report Introduction Test principle Definition of the type of the analytical test procedure Definition of validation parameters Performance of the test Test outline and acceptance criteria Equipment qualification System suitability Qualification of samples and standards Summary References Results B. Neidhart, W. Wegscheider (Eds.): Quality in Chemical Measurements © Springer-Verlag Berlin Heidelberg 2000 17