QP-12: Immunoassay Applications
|
|
Quality planning for immunoassays poses some special difficulties because there are usually multiple decision levels, different quality requirements at these different decision levels, and different method CVs at different decision levels. The title of an April 2000 survey of immunoassay instruments in CAP TODAY makes the point - "Of all analyzers, immunoassay the trickiest" [1]. QC designs may be more complicated because immunoassay measurement procedures are generally not as precise as the highly automated chemistry and hematology methods, therefore it may be necessary to utilize multi-level or multi-stage designs, often with multirule control procedures and a higher number of control measurements.
Endocrine tests and some toxicology tests are commonly performed by immunoassay methods. CLIA's list of regulated endocrine tests is short - only seven tests - cortisol, free thyroxine, human chorionic gonadotropin, T3 uptake, triiodothyronine, thyroid stimulating hormone, and thyroxine. For all but cortisol and thyroxine, the quality requirements are given as Target Value plus and minus 3 SD, where the SD here is estimated from the group of laboratories participating in the proficiency testing survey. For Cortisol, the allowable total error is given as 25%; for thyroxine, the allowable total error is given as 20% or 1.0 mcg/dL, whichever is greater. CLIA's requirements for toxicology are generally stated as Target Value plus and minus 20 or 25%, the higher figure being used most often.
For other assays, it would seem appropriate to extend the 3 SD peer group concept and utilize the data from inter-laboratory peer comparison programs and external quality assessment programs as the starting point for defining the quality requirements. Clinical quality requirements may also be used, but their application requires access to more sophisticated planning tools, such as a more complex clinical quality-planning model and a computer program, such as the QC Validator program [2-3], to carry out the calculations.
Imprecision is estimated from replication studies that are typically carried out with three or more levels of control materials. The SDs observed and the CVs calculated are likely to vary widely from low to high end of the reportable range. CVs may be as large as 10 to 15% at one end of the range and typically will be about 5% in the most precise part of the range. Inaccuracy can be assessed from comparison of methods experiments, but the lack of reference methodology makes it difficult to assign systematic errors to one method or the other. Therefore, it is widespread practice to assess systematic errors by comparison to like methods in monthly peer-review programs or periodic proficiency testing surveys.
A study published by Mugan et al [4] describes applications for 7 different tests that were performed on an automated analyzer - prolactin, Total b-hCG, CEA, FSH, LH, TSH, and b2-Microglobulin. Seth [5] has illustrated how QC procedures can be selected with the aid of critical-error graphs. Carey has provided some general QC planning guidelines, as well as some detailed examples for theophylline, cortisol, thyroxine, and folate, that are available on the Internet [6].
NOTE:
- Normalized OPSpecs operating points can be calculated on this website using the Normalized OPSpecs Calculator. Click here to see the calculator.
- The Normalized OPSpecs charts can be download from this website in PDF format. Click here to download the charts.
A method has a CV of 5.5% and a bias of 0.0% at a decision level of 5.0 ug/dL. CLIA defines the allowable total error as 20%. Given 3 control materials are to be analyzed, select an appropriate QC procedure and Total QC strategy.
- The normalized operating point would have an x-coordinate of 27.5% [(5.5%/20.0%)100] and a y-coordinate of zero
- Two solutions are identified when the operating point is plotted on the normalized OPSpecs chart for N=6 and 90% AQA (see figure above). Multirule procedures are necessary to provide the desired 90% detection of medically important systematic errors.
- The Total QC strategy could depend on the high error detection of the multirule QC procedure, however, it would be advantageous to improve method performance and reduce the number of control measurements from 6 to 3 per run. Inspection of the N=3 90% AQA chart shows that the x-coordinate can have a maximum value of about 23%, which corresponds to improving the method CV to 4.6%.
A method has a CV of 5.3% at a medical decision level of 30 ug/dL. Method bias is assumed to be zero. The CLIA PT criterion is 25%. Given 3 control materials are to be analyzed, select an appropriate QC rules, N, and TQC strategy.
- The normalized operating point would have an x-coordinate of 21.2% [(5.3%/25%)100] and a y-coordinate of 0%.
- The normalized OPSpecs chart for N=3 with 90%AQA shows 4 possible solution (see figure above), but the 12s procedure should not be considered because of its high rate of false rejections, about 0.14 or 14%. A 12.5s single-rule procedure would be the simplest to implement and is perfectly adequate for detecting medically important systematic errors in the first run in which they appear.
- The Total QC strategy can depend on the statistical QC procedure to detect problems.
The method CV is 7.5% at a level of 0.8 uIU/mL, 6.0% at 4.8 uIU/mL, and 6.0% at 26.6 uIU/mL. Method bias is assumed to be zero. TSH is a regulated analyte with a CLIA PT requirement in the form of TV +/- 3SD. Data available from a CAP peer-review program yields calculated TEa values of 28% at 1.0 uIU/mL, 19% at 5.0 uIU/mL, and 19% at 25 uIU/mL. Given three control materials are to be analyzed, select an appropriate QC rules, N, and TQC strategy.
A multilevel QC design would be useful here to provide one QC procedure for monitoring performance in the range of 5 to 25 uIU/mL and another to monitor performance at the low end. The two highest control materials can be used in one design and the lowest material in the other.
Upper control level
- To select a QC procedure to monitor the middle to upper end of the reportable range, the normalized operating point would have an x-coordinate of 31.6% [(6.0/19)100] and a y-coordinate of zero.
- OPSpecs charts for Ns of 2 and 4 are used because the quality requirements and the CVs are the same for control materials at 5.0 and 25.0 uIU/mL. Three possible solutions are found on the normalized OPSpecs chart for N=4 with 50%AQA (see figure above). A multirule procedure is the best choice here and the 8x rule should be added to permit application across two runs.
Lower control level
- To monitor the low range of the reportable range using the control material whose mean is 1.0 uIU/dL, the normalized operating point would have an x-coordinate of 26.8% [(7.5/28)100] and a y-coordinate of zero.
- No solution can be found on the normalized OPSpecs chart for N=2 with 90%AQA or N=2 with 50%. However, the N=2 chart with 50%AQA shows that a multirule procedure with 13s/22s/R4s will provide at least 50% error detection. The 41s rule can be added to look-back at control data from a previous run, which should provide nearly 90% error detection over two runs. Two observations would be collected on the low control material each run and the multirule procedure applied to the data from the current and immediate previous run.
- The TQC strategy should take into account that less than 50% error detection is available in a single run. Considerable attention must be paid to non-statistical QC procedures to prevent problems from occurring.
Quality-planning skills and experience are needed to deal with immunoassays. Basic quality-planning tools will get you started, help you develop your skills, and motivate you to learn about clinical quality requirements and the more advanced planning tools that support their use.
- Start with peer group SDs to calculate quality requirements. The use of peer group SDs to calculate an allowable total error is a rational starting point in US laboratories because of the way that CLIA initially sets quality requirements.
- Consider more advanced quality-planning tools that can utilize clinical quality requirements. An alternate approach would be to use clinical quality requirements in the form of medically important changes in test results that lead to changes in interpretation and treatment. The use of clinical quality requirement also requires knowledge of intra-individual biologic variation to account for an important preanalytic factor that affects test results. Incorporating that information also means a more complicated quality-planning model, which in turn complicates the calculations enough that computer support is needed, such as provided by the QC Validator computer program.
- Utilize multilevel QC designs for different performance and different quality at different levels. When the quality requirements and method CVs are similar across the whole reportable range, a single QC design can be used and all levels of control materials incorporated in that design. This premise is less likely to hold for immunoassay tests than for general chemistry and hematology tests, therefore it may be effective to design different QC procedures for different medical decision levels or different parts of the reportable range. However, implementation of two different QC designs may be difficult with the current generation of QC software.
- Utilize higher N multirule procedures to increase error detection. Achieving adequate error detection may often require a higher number of control measurements. Multirule designs have been optimized for up to 8 control measurements to maximize error detection and minimize false rejections [7]. A 13s/3of82s multirule with N=8 provides very high error detection while keeping false rejections at 2-3%.
- Utilize multi-stage QC designs to monitor regular system changes. In situations where it is difficult to achieve high error detection AND low false rejection, two different QC designs can be employed - one for high error detection (with some increase in false rejection - a startup design) and another for low false rejection (with moderate error detection - a monitor design). Weekly changes in reagent lots might be checked at the beginning of the week using the startup design (high error detection and moderate false rejection), then followed throughout the week using the monitor design (low false rejection and moderate error detection).
