QA/QC for ANRS Soil, Water and Plant Testing Laboratories

1. QA/QC for ANRS Soil, Water and Plant Testing Laboratories Presented by: Farzad Dadgari Soils and Environmental Specialist, SWHISA

2. Overview • Why QC? • What’s in it for me? • QC vs. QA ...what’s the difference? • General Components of a QA/QC Program • Examples of QC • Setting up an Effective QA Plan

3. QualityControl

4. Why QA/QC? • Fundamental concept applies far beyond laboratory • Would you .... buy used car from a car rental company? • .... drive cross-country before checking the oil, brake fluid, water, etc.? • Why QA/QC? • Common sense • “Checks and balances” is a universal concept. • “To err” is human

5. Why QA/QC? (cont.) • Prove data valid mostly through documentation • “He who has the best documentation wins.” • Important to data user andthe laboratory • User wants to make correct decisions…and sleep at night! • Laboratory wants to produce a good product…and pass the audits! Yeah…but what’s in it for me?

6. What’s in it for me? • Millions spent on, say, irrigation development projects based on the lab data--Consider the troubles of irrigation project and farmers when project is designed and built using inaccurate soils and water quality data that is later determined to be inaccurate; • Better data is needed to maintain lab certification when such program starts;

7. What’s in it for me? (cont.) • Quality data is required to ensure recommendations are accurate; • Policy and guidelines promote uniformity • Decreased learning curve for new employees • Fewer repeated analyses • and…?

8. OK, so we agree that QC and QA are important...but aren’t they the same thing?

9. QC v. QA QA General managementfunction to ensure data quality relies on: • documentation and establishment of QC protocols, • evaluation and summarization of their outcomes. QC Specific technical,operational measures or activities to ensure lab data quality.

10. QC vs. QA QCQA 100 mg weight = 83 mg OK No spike control limits 90-110% OK OK barometer reads 72 cm OK No no spikes for ammonia, TP No No TP spike limits -275 to 3047% OK No TSS oven temp. kept 180 + 1oC --- No Filling out bench sheets in advance with OK No dates and names in advance of collection With clear distinction....you can build a QA/QC program

11. A good QA Program Components The foundation: • Good facilities and equipment; • Training of personnel; • Operation plan (assigned responsibilities); and • Methods documentation and follow ups. The structure: • Rigorous QC procedures; • Precision; • Accuracy; and • Documentation to ensure traceability.

12. Type and Uses of QC samples Blanks • Laboratory reagent water; • Used to verify the absence of contamination in the lab; • Particularly important in phosphorus and ammonia testing.

13. Type and Uses of QC samples (cont.) Known Standards • Used to verify calibration curve accuracy, or • Absence of bias in laboratory procedure (vs. matrix-effects) • Best if these are prepared from a different standard than is used for calibration standards.

14. Type and Uses of QC samples (cont.) Replicates (Dups) • Used to measure the ability to reproduce your results. Maybe you got it right once, but can you do it again?

15. Type and Uses of QC samples (cont.) Spikes • Used to evaluate bias (i.e., the recovery of the analyte from the specific sample matrix). • If you only get 25% spike recovery, …and your sample concentration is close to a permit limit …isn’t it likely that the permit limit has actually been exceeded?

16. Type and Uses of QC samples (cont.) Reference Samples • Annual requirement • “Show me you can do this test right” Blind Standards • Same as reference samples, but more timely. However, the materials must be used correctly to serve their purpose: precision and accuracy

17. Precision vs. Accuracy Accuracy (Bias) • How close you can get to the true value. You want LOW bias (bias is not a good thing) Precision • Reproducibility of the method. The ability to get the right answer – again and again You want HIGH Precision

18. Precision High Low High Farzad shoots like this Bias Low Mengistu shoots like this You want high precision and low bias!

19. Pitfalls of Poor P&A • Report results that show high soil salinity (when the soils is actually non-saline) • Possibility of periodic, unexplainable limit violations. • Bring overall ability to provide meaningful soil fertilizer recommendation into question.

20. Setting up an effective QA Plan • Standard Operating Procedures (SOPs) should be available for anything not self-explanatory Ex. How do you clean the phosphorus glassware? Essentially, if someone unconnected to the lab were to perform this task, what guidance would they need to do it?

21. Setting up an effective QA Plan (cont.) • You can simply reference SOPs, rather than including them in your QA Plan • If you are not doing it, DON'T include it in the QA Plan.

22. QA Plan’s “DON’Ts” • DON’T allow your QA Manual to read like ... Quality assurance is a systematic design plan incorporating a number of related laboratory aspects.

23. QA Plan’s “DON’Ts” (cont.) • We know what QA is, but it’s too complicated to explain here. “First you have to calibrate” Accurate and precise analytical data can only be realized by systems that are capable of comparing response of a real world sample to the response of a known standard.

24. Setting up an effective QA Plan • Tables are better than lots of text! • the old “a picture is worth 1000 words” concept • Tables FORCE you to be brief

25. Setting up an effective QA Plan (cont.) 3 rules for building a QA Plan by tables • What am I looking at? (parameter) • What am I looking at it for (criteria) • What if it doesn’t meet specifications? (Corrective Action) Sound easy enough? Let’s see some real-life examples…

26. Setting up an effective QA Plan

27. QA PLANS - The Bottom Line • Brief NOT Volumes • Realistic NOT Marketing "fluff" • Guidance NOT Philosophy • Decision trees NOT Generic options • Reference NOT Paperweight • Tables NOT Text

28. Quality ContorlControl

29. Calibration

30. CALIBRATION - Discussion points (cont.) • Initial vs. continuing calibration • How many standards to use? • To include...or not to include (a blank)? • Processing the data • internal calibrations • graph paper • linear regressions • software

31. CALIBRATION - Discussion points (CONT.) • Evaluating a calibration • Visual • Statistical • analytical

32. CALIBRATION - Initial Considerations Frequency • For best results, should be run daily. • Alternatively, a “full” calibration can be analyzed initially and verified (with one or more standards) each day of analysis. Use an appropriate number of standards • Calibrations must be constructed using at least 3 standards and a blank.

33. CALIBRATION - Initial Considerations (CONT.) Know when to include a zero • A good rule of thumb: If you can adjust your instrument to read zero in the presence of a blank, then include a zero point in your calibration curve. • Including a zero is generally appropriate for colorimetric procedures.

34. CALIBRATION - Initial Considerations (CONT.) Define your calibration range properly • Range should be appropriate for the samples being analyzed (i.e. don't calibrate from 1- 5 mg/L when all samples are between 0.05 - 0.5 mg/L). • Better results are obtained when sample response is close to response of standards used to establish the calibration curve.

35. CALIBRATION - Initial Considerations (CONT.) • Optimal results ==> when sample results fall near the mid-point. • Standards should also be evenly spaced. (1, 2, and 500 are NOT good levels for a calibration) • Whenever possible….bracket samples with calibration standards. • Low standard not more than 2 to 5 times the LOD (best is = LOQ).

36. CALIBRATION - Processing the data Pre-programmed Calibrations • Use of pre-programmed calibrations is unacceptable • Laboratory must generate its own standard curve. NOTE: A manufacturer’s claims that their method is approved or acceptable does not mean that the approval extends to pre-programmed calibrations.

37. CALIBRATION - Processing the data (CONT.) Hand-drawn Calibration Curves • Plot concentration on the x-axis and absorbance on the y-axis. • A straight line which best fits the data points is then drawn. • The "best fit" line used to convert absorbance into concentration.

38. CALIBRATION - Initial Considerations (CONT.) • Makes traceability virtually impossible • Significant variability in how the scale of the graph is constructed; and • Significant variability in how any individual draws the "best fit" line.

39. CALIBRATION - Initial Considerations (CONT.) Scientific Calculators & Software • Using a standard procedure can eliminate sources of variability. • Linear regression = one of the most widely recognized calibration means. • Linear regression equations can be generated with an inexpensive scientific calculator, or most spreadsheet programs (Excel, Lotus, etc.).

40. CALIBRATION - Processing the data Calibration exercises 1. Graph paper Calibration Data Find concentration for mg/L Abs. For these Absorbances: 0 0 Then... 0.118 0.1 0.051 0.531 0.5 0.25 0.770 2 0.72 0.853 5 1.24 1.092

41. CALIBRATION - Processing the data • Draw the X and Y axis (graph frame) • Number your x-axis to fit the range of the data. • Number your y-axis to fit the range of the data. • Plot each of the data points.

42. CALIBRATION – Data Processing by graph

43. CALIBRATION – Data Processing by graph 5. Draw the line that “best fits all the data points. Or this?... Or is it this?...You Decide!

44. CALIBRATION – Data Processing by graph Draw a line over to your “best fit” line Drop a line from that point down to x-axis ...and read concentration off the x-axis.

45. CALIBRATION – Data processing by graph Observations • Some prepared graph landscape vs. portrait • Some use the whole page, others just part of the page • Quite a bit of spread in the data....range of 1 mg/kg or more • Some read to nearest 0.01 mg/kg; others to nearest 0.1

46. CALIBRATION – Data processing by graph (CONT.) Food for thought.... • What if your permit limit is right about 2 mg/kg? • Should the line pass through the origin (0,0)?

47. CALIBRATION - Processing the data Calibration exercises 2. Linear Regression Calibration Data Find concentration for mg/L Abs. For these Absorbances: 0 0 Then... 0.118 0.1 0.051 0.531 0.5 0.25 0.770 2 0.72 0.853 5 1.24 1.092

48. CALIBRATION - Regression using calculator • You need a basic calculator with statistical analysis capability (regression, mean, standard deviation, and correlation coefficient). • Value of the Correlation coefficient (r). • Looking for r > 0.995 • Tells you how closely points fit the regression (best fit) line

49. CALIBRATION - Regression using calculator (CONT.) • Value of the Slope (b) • Can keep records to show when the analysis is changing Value of the Intercept (a) • Represents the concentration associated with no (0) response • Thus gives an approximation of detection limit. If your intercept exceeds your LOD, there may be contamination

50. CALIBRATION - Regression using calculator (CONT.) • The ONLY downside of using a calculator vs. a spreadsheet (computer) program is that you do NOT get the visual evaluation power afforded by charting the data and regression line.