Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
An Introduction to Quality Assurance in Analytical Science Dr Irene Mueller-Harvey Mr Richard Baker Mr Brian Woodget University of Reading Part 2 - The Analytical Method Contents: • • • • • • VAM principles (slides 3&4) Fit for purpose (slide 5) Choice of method (slide 6) Method development and validation (slide 7) Method performance characteristics (slides 8 - 15) Method validation (slides 16 - 25) The presentation contains some animation which will be activated automatically (no more than a 2 second delay), by mouse click or by use of the ‘page down’ key on your keyboard. VAM Principles (1) The DTI’s programme on Valid Analytical Measurement (VAM) is an integral part of the UK National Measurement System. The aim of the VAM programme is to help analytical laboratories to demonstrate the validity of their data and to facilitate mutual recognition of the results of analytical measurements. You can find out more about VAM by logging onto the web site [http://www.vam.org.uk/] The VAM Bulletin is published half yearly and is sent free to all registered subscribers. You may subscribe via the web site. VAM Principles (2) There are six VAM principles: Analytical measurements should be made to satisfy an agreed requirement Analytical measurements should be made using methods and equipment which have been tested to ensure they are fit for their purpose Staff making analytical measurements should be both qualified and competent to undertake the task There should be a regular independent assessment of the technical performance of the laboratory Analytical measurements made in one location should be consistent with those elsewhere Organisations making analytical measurements should have well defined quality control and quality assurance procedures Fit for Purpose You have seen in part 1 of this presentation that it can be expensive to generate good quality analytical data. The cost of producing data can often be reduced by selecting analytical methods and technologies that produce data in accordance with the stated objectives for carrying out the analysis or test. Consider two examples: A commercial limescale remover states that it contains between 5 to 15% w/v of sulphamic acid. Therefore any analysis for quality control purposes needs only to show that the quantity present is indeed between these two limits and does not need to be accurate to the nearest + 0.1% The EU current statutory limit for lead in potable water is 25 μg/l. Thus the use of a method to perform the analysis, which only provides a ‘less than’ value of 100 μg/l, is not suitable for this purpose Choice of Method It is important to appreciate the difference between an ‘analytical method’ (combination of steps illustrated by the ‘analytical process’) and an ‘analytical technique’ (chemical or instrumental procedure by which analytical data is eventually obtained). In selecting a method we shall need to consider the following parameters: • sample type (matrix) and size (lot or a little); • data required (qualitative/quantitative); • expected level(s) of analyte(s); • precision & accuracy expected; • likely interferences; • number and frequency of samples for analysis. Always use a standard method if one is available as this will save on development time. However the method must be checked to prove that it suitable for your laboratory/situation. Modification may well be required. Method development and validation ‘Never attempt to re-invent the wheel!’ Before embarking on the development of a new method, always research the chemical literature to see if a suitable one already exists. If a suitable one is found, it will still be necessary however to perform some method validation to prove that the method can be successfully adapted to your laboratory, equipment and personnel. More extensive validation is required for a brand new method. Methods in any field of analysis may be defined in terms ‘Method performance characteristics’ and it is these parameters plus a few others, that are quantified during a method validation exercise. Method performance characteristics A method’s performance is defined by a number of important individual characteristics. There are: Sensitivity Accuracy Precision Limit of Detection (LoD) Limit of Determination Selectivity Linear Range Dynamic range Bias Accuracy and precision The dictionary definition of both ‘accuracy’ and ‘precision’ are roughly the same, indicating that these words may be used synonymously. However in ‘Analytical Science’ they have two separate meanings, the difference between them is best illustrated by using target diagrams Poor precision poor accuracy Good precision poor accuracy Good mean accuracy poor precision Good accuracy good precision Accuracy and precision (2) You saw from the previous slide, a set of results can be either accurate and/or precise or can be neither accurate nor precise. Thus accuracy may be defined as: The closeness of the mean value from a replicate set of results to the true or accepted value Precision may be defined as: The spread of results from a replicate set of measurements The difference between the true value and the mean measured value is termed bias. The spread of replicate data is measured in terms of standard deviation (s) or variance (s2) Random and systematic errors There are 2 types of error for which allowance may be made: Random error Systematic error Random error arises from variations in parameters which are outside the control of the analyst, but which influence the value of the measurement being made. Because these errors are statistically random, the mean error should be zero if sufficient measurements are made. mean Systematic error remains constant or may vary in a predictable way over a series of measurements and cannot be reduced by making replicate measurements. In theory, if known, this error can be allowed for. Eg: subtraction of blank values mean Bias and variance A solution containing copper was analysed 10 times using atomic absorption spectroscopy. The results obtained in ppm were: 10.08, 9.80, 10.10, 10.21, 10.14, 9.88, 10.02, 10.12, 10.11, 10.09 We can now calculate the precision of the data as standard deviation If the true value is known to be 10.00 ppm, we can also calculate the bias Cu in ppm Cu by AAS 10.4 10.2 10 9.8 9.6 0 2 4 6 8 10 12 Replicate sample Bias = Mean value - true value = 10.06 - 10.00 = 0.06 ppm Standard deviation (SD) = 0.12(4) Relative SD = 100 X SD/10.00 = 1.2% Conclusion - the method gives both good accuracy (low bias) and acceptable precision (RSD of 1.2%) Sensitivity and selectivity 1200 1000 Signal Sensitivity is the change in measured signal for unit change in concentration and can be obtained from the calibration graph Assessment of method sensitivity 800 600 dy 400 200 dx 0 Sensitivity = dy/dx 0 10 20 Concentration Selectivity is the ability of a method to discriminate between the target analyte and other constituents of the sample. In many instances selectivity is achieved by high performance separation using chromatographic or electrophoretic techniques. Hplc chromatogram 30 Limits of detection (LoD) and determination Example: In an analysis of trace Cd by plasma emission These values refer to the statistical limits spectrometry the following data were obtained: below which results of detection or • mean blank (Bl) signal 4 accurate quantitative measurements 12 (determination) should not be reported. • SD of blank signal 2000 The levels of both are dependent upon • 500 ppb Cd the variability of the signal when a blank containing none of the analyte is being measured. The signal generated under these conditions is mostly signal noise and is assumed to exhibit a normal distribution pattern. Both the blank signal and the standard deviation of the blank signal need to be measured. From this data we can calculate both limits. LoD = Bl signal + 3(SD of Bl) = 4 + 3(12) = 40 This equates to: [40/2000] X 500 ppb = 10 ppb Cd The limit of determination uses a similar formula, replacing the 3 SD’s by 10. This gives the limit of determination as 31 ppb Cd Linear and Dynamic ranges These terms refer to the extent to which the method may be used to produce accurate quantitative data From the graph, it would appear that the data is linear to about 25 ppm and dynamic until about 75 ppm. After 75 ppm there is only minimal increase in signal for increased concentration. Graph showing linear & dynamic ranges 1 0.6 Linear & dynamic ranges 0.4 0.2 0 0 50 100 150 200 Concentration/ppm Signal Signal 0.8 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0 20 40 Concentration/ppm Extent of dynamic range 60 Expanding the lower section of the graph however, shows that nonlinearity starts at around 20 ppm. Top of linear range Method validation Validation, is the proof needed to ensure that an analytical method can produce results which are reliable and reproducible and which are fit for the purpose intended. The parameters that need to be demonstrated are those associated with the ‘Performance characteristics’ together with robustness, repeatability and reproducibility. Many analytical methods appearing in the literature have not been through a thorough validation exercise and thus should be treated with caution until full validation has been carried out. Validation of a new method (new to your laboratory), is a costly and time-consuming exercise, however the result of not carrying out method validation could result in litigation, failure to get product approval, costly repeat analysis and loss of business and prestige. You can now consider in more detail how validation is carried out Method validation linearity Calibration graph 0.6 0.5 Signal Check linearity between 50 - 150% of the expected analyte concentration 0.4 0.3 0.2 0.1 0 0 5 10 15 20 Concentration Linearity Most analytical methods are of a comparative type and thus require calibration against accurately known standards to generate quantitative data. Where possible calibration data should show a linear relationship between analyte concentration and measured signal, however it is acceptable under some circumstances, to use a non-linear relationship up to the limit of the dynamic range. 25 Method validation specificity Loss of specificity can be due to interferences and matrix effects. All likely interferences should be investigated and their effects on analyte response determined over a range of concentrations. Measures can then be put into place to mask, eliminate or separate them from the analyte. Standard addition procedures can be used to identify matrix effects no yes Normal distribution curve Method validation precision You have seen already that, precision is measured in terms of standard deviation (SD). Assuming that the variability of the measurements is totally random (obeys a normal distribution curve) then a formula derived from this distribution may be used to calculate standard deviation. Frequency of occurrence 100 80 60 40 20 0 0 50 100 150 Data points Estimation of true mean SD = [ Σ(xi - x)2/(n - 1)]1/2 where: xi = individual data point x = mean value of the data n = the total number of data points Σ = the sum of In practice around 8 - 10 data points are used normally to calculate the SD, although statisticians would recommend 50 Method Validation repeatability & reproducibility Methods need to be shown to be both repeatable and reproducible. A replicate set of data produced at a particular time point by an operator working with a particular set of equipment in a given laboratory will verify repeatability. To show reproducibility, the method must produce similar results when any of these parameters are changed. The most likely changes are to time and operator. Two different operators analysing milk using different pieces of equipment at different times. The laboratory is the same. Method validation reliability The reliability of a method can be tested in a number of ways Test results from the new method against an existing method which is known to be accurate Add a known quantity of pure analyte (spike) to a real sample or real sample matrix and check that all of the added substance can be measured (recovered) Selection of reference materials from LGC The best way of demonstrating accuracy is to analyse a reference material or certified reference material (CRM) if one is available [see part 3 of this presentation] Method validation detection & quantitation limits A method is not acceptable for accurate detection or quantitation if the analyte level is likely be fall beneath the limit(s) calculated based upon the blank signal and its standard deviation (Refer to the formula given in slide 14, in this part of the presentation). Analyte pre-concentration then becomes necessary. Variation in sample signal Variation in blank signal Mean blank signal Mean sample signal Mean sample signal must be sufficiently larger than the blank so that positive detection or accurate quantitation is possible Method validation robustness Robustness of an analytical method refers to it’s ability to remain unaffected when subjected to small changes in method parameters. For example In an hplc analysis the mobile phase is defined in terms of % organic modifier, pH of the mobile phase, buffer composition, temperature etc. A perfect mobile phase is one which allows small changes in the composition without affecting the selectivity or the quantitation of the method. Alter all major parameters in order to ascertain when the method ceases to function in accordance with specifications Method validation establish stability In routine analysis where numerous samples and standards are measured each day, it is essential to assess the stability of the prepared solutions. Stability of these solutions should be tested by repeat analysis over at least a 48 hour period. Apparent onset of solution instability Signal Time Method validation additional reading An article entitled “A Practical Guide to Analytical Method Validation” was published in Analytical Chemistry in 1996 [ Anal. Chem. (68) 305A309A] The article may be downloaded free from the acs web site: http://pubs.acs.org/hotartcl/ac/96/may/may.html