Quality Assurance

Haemoglobin reference preparations

The accessibility of an international reference standard of haemiglobincyanide (HiCN), first developed by the ICSH, has improved the accuracy of haemoglobin measurement. In some countries, preparations that conform to the international standard are certified by the appropriate national authorities. An important feature of this material is that it is stable for at least several years. A limited quantity of the international standard can be obtained from the WHO; a comparable certified reference material is available from IRMM, and the ICSH has produced a new preparation with similar specifications (see Table 25-4 ). Where the use of the cyanide reagent for routine haemoglobinometry is prohibited, the haemiglobincyanide standard can still be used to assign a haemoglobin value to a lysate or a whole blood preparation, which is then used as the local secondary standard after appropriate dilution. Undiluted lysate is usually stable for up to 6 months, if frozen for several years. Whole blood is stable for about 3 weeks, but for only a few days after dilution. Both whole blood and lysates are useful for QA of haemoglobinometry; reference samples should be introduced into batches of blood samples with all the samples being assayed together. This applies to both automated and manual methods. Many laboratories are unable to make use of this reference preparation as it is unsuitable for direct use in automated haematology analysers; however, the analyser manufacturer will trace the instrument’s calibration to the reference preparation.

Quality control preparations

Controls are preparations which are used for either IQC or EQA. Some control preparations have assigned values but they should not be used as standards as the assigned values are usually only approximations and the preparations may be stable for a limited time only.

Control materials are available commercially and they can also be made locally, although there are difficulties in the preparation and validation of such ‘homemade’ materials. For example, stored plasma may become turbid; chemical or serological analysis may be affected by instability of enzymes; added preservatives may interfere with immunological reactions. With the blood count, there are especially difficult problems due to the instability of blood cells and the need to ensure homogeneity in aliquot samples, while procedures that enhance the stability of blood samples may also affect the behaviour of the cells. The preparation of extended life materials for QA is described at the end of this chapter.

Internal quality control procedures

Internal quality control monitors the performance of the test procedures in the laboratory on daily or a batch-to-batch basis. It includes measurements on specially prepared control materials, repeated measurements on patients’ specimens and statistical analysis of patients’ test data. This ensures continual validation of the reliability of the results produced by the laboratory, before reports are released.

Control charts

These were first applied in clinical chemistry by Levey and Jennings as a means to monitor laboratory performance using control materials and are now widely used in haematology for both automated and manual procedures. Control material is included in every batch of patients’ specimens or at specified intervals during the shift, and the result is logged on a control chart. To check precision, it is not necessary to know the exact value of the control specimen; however, if its value has been determined reliably by a reference method, the same material can also be used to check accuracy or to calibrate an instrument. If possible, controls with high, normal and low values should be used to monitor the method across the range of its linearity. It is advisable to use at least one control sample per batch, even if the batch is small, or at set intervals during a large run or a continuous process. Because the controls are intended to simulate random sampling, they must be treated in the same way as the patients’ specimens as far as possible. The results obtained with the control samples can be plotted manually on a Levey–Jennings style chart, as described below, or can be plotted automatically by the instrument, which may also provide an alert if the analysis falls outside the preset, acceptable limits of performance.

When plotting a control chart manually, the mean value and standard deviation (SD) of the control specimen is first established in the laboratory. Using arithmetic graph paper, draw a horizontal line to represent the mean (as a base) and, on an appropriate scale of quantity and unit, draw lines representing + 2SD and − 2SD above and below the mean. Plot the results of successive control sample measurements. If the test performance is satisfactory, sequential results oscillate about the mean value and less than 5% of the results fall outside ± 2SD.

Figure 25-1 illustrates a control chart from an automated system; a similar principle can be used for simple methods where the data are plotted manually ( Fig. 25-2 ).

Any of the following indicates a fault in technique or in the instrument or reagent:

• 1.

  One deviant result outside ± 3SD may indicate a gross error or ‘blunder’.

• 2.

  One or two results on or beyond ± 2SD may indicate a random error in the system.

• 3.

  Several consecutive similar results on one side of the mean indicate a consistent bias, possibly due to a calibration fault or a systematic error.

• 4.

  Consecutive fluctuating values, rising and falling by ± 2SD, indicate imprecision.

• 5.

  A trend of sequentially increasing or decreasing values with the repeated measurements is indicative of drift.

The fault may be in the reagents or the laboratory-ware, caused by incorrect adjustment/calibration of the instrument or other equipment (e.g. pipettes), technical error or even clerical error in transcribing the results. Before an intensive investigation is carried out, the test should be repeated with another control sample, and the possibility that the inconsistency is the result of deterioration of the batch of control material or insufficient mixing of the sample should be excluded. This control process is unlikely to detect an error in an individual specimen, which can only be detected by correlation checks. For haemoglobinometry, it may be useful to use both whole blood and lysate in a quality control check because differences in results obtained with these two preparations help to identify errors resulting from incorrect dilution, inadequate mixing or failure of a reagent to bring about complete lysis. If the control specimen is included with each batch of tests or is repeated during the course of a day, the measurements should not differ by more than the established coefficient of variation (CV) (see p. 566).

Duplicate tests on patients’ specimens

Duplicate tests on patients’ specimens provide another way of checking the precision of routine work. To start the process, test 10 consecutive specimens in duplicate under careful conditions. Calculate the differences between the pairs of results and derive the SD (p. 566). Subsequent duplicate measurements on any specimen in the same batch of tests should not differ from each other by more than ± 2SD. This method will detect random errors but not incorrect calibration. It is also insensitive to gradual drift which may, however, be detected if duplicate measurement is carried out in a later batch, provided that the specimen has not altered on standing. If the test is always badly done or has an inherent fault, the SD will be wide and the procedure may be ineffective.

This procedure is suitable for both manual and automated methods; it is impractical for routine blood counts in a busy laboratory, where three or four specimens in a batch can be tested in duplicate from time to time as a rough check of consistency.

Use of patient data for internal quality control

Where a minimum of 100 test requests are processed each day, there should be no significant day-to-day variability in the means of the red cell indices obtained by an automated blood counter, provided that the population of patients remains stable and that samples from a particular clinical source (e.g. renal or oncology patients) are not processed all in the same set, disproportionately influencing the mean. Assuming that the sample population is stable, any significant change in the means of the red cell indices will indicate a change in instrument calibration or a drift owing to a fault in its function. The procedure was developed by Bull ^, using a computerised algorithm to estimate the daily patient means of absolute values for mean cell volume (MCV), mean cell haemoglobin (MCH) and mean cell haemoglobin concentration (MCHC). In laboratories using manual methods, a simple adaptation of the same principle can be applied, confined to MCHC and excluding results from any specific clinic that are likely to be biased. From the daily means for all measurements on 10 consecutive working days, an overall daily mean and SD are established. The mean MCHC is then calculated at the end of each day. If the test does not vary by more than ± 2SD, it is considered to be satisfactory, but may be misleading if there is a simultaneous error in the same direction in both Hb and Hct. The results may be displayed graphically as illustrated in Figure 25-3 .

To start this programme for an automated laboratory, it is first necessary to assay samples from at least 300 to 500 patients using an automated blood counter over several days and to establish the means of the MCV, MCH and MCHC. Then, using the algorithm, it is possible to analyse the means in successive batches of 20 specimens. Plotting these means will allow any drift of the three indices to be recognised and used to identify instrument faults; an increased SD will signify loss of precision. To ensure that each batch is representative, the samples should be randomised before analysis and, if possible, within any batch of 20, no more than seven should come from one clinical source or have the same clinical condition. The method is now incorporated in automated blood counters ( Fig. 25-4 ).

A similar instrument-specific procedure is used by some manufacturers who gather the data submitted through network links by users of their instruments. This enables them to maintain a constant check of performance of these instruments overall and to detect any that require recalibration or investigation of faults.

Correlation check

A correlation check implies that any unexpected patient’s result must be checked to see whether it can be explained on clinical grounds or whether it correlates with other tests. For example, an unexpectedly higher or lower Hb might be explained by blood transfusion or haemorrhage, respectively. A low MCHC may be confirmed by demonstrating hypochromic red cells on a stained blood film; a high MCV must correlate with macrocytosis. Similarly, the laboratory should have a policy on the examination of a blood film to confirm marked leucocytosis or leucopenia, thrombocytosis or thrombocytopenia, to distinguish between platelets and red cell fragments or conversely between giant platelets and normal-sized red cells and to check features flagged by the automated analyser. The examination of a stained blood film as a correlation check will only be effective if the film is correctly made, stained and examined systematically by an experienced member of staff.

Recording or reviewing cumulative laboratory data on an individual patient is good clinical practice and provides an inbuilt quality control system. This is especially useful in detecting pre-analytical phase errors (e.g. patient and sample identification errors, incorrect sample collection, inadequate suspension of the blood before testing, partial clotting or deterioration on storage).

A formal way of detecting aberrant results is the ‘delta check’. The blood count on any patient should not differ from counts obtained in the previous 2–3 weeks by more than an amount that takes into account both the test CV and physiological variation, providing that the patient’s clinical condition has not altered significantly. A discrepant result without apparent clinical reason must be suspect until confirmed by a repeat test on a fresh specimen. The occurrence of a contrasting discrepancy in two different specimens on the same day would suggest that two specimens have been transposed. The test can, of course, also be carried out on the blood of healthy individuals whose blood count remains virtually constant on a day-to-day basis, subject only to physiological change (see p. 13).

Target values

The absence of metrological standards in haematology (with the exception of Hb) makes it difficult to establish the ‘true’ value for a test in haematology EQA. The EQA target value for a quantitative haematology assay can usually be assumed to be the result obtained by best performance of selected participants in the survey, by experts using reference methods or the consensus mean or median of participants’ results after trimming to remove outliers. In practice, the consensus of participants’ results is the usual method for determination of the target value in automated cell counting EQA. Consensus values for quantitative tests should only be calculated when the number of participants’ results available is sufficient to allow statistically meaningful analysis of results, and this generally will require a minimum of 20 participants per analysis group. Where a consensus all methods mean is used, it may be skewed by the instrument or method that has the greatest number in the data set; therefore, it is desirable to provide both all methods and method or submethod statistical analysis to elucidate performance problems and ensure fair treatment of participants. The use of stabilised EQA survey material (e.g. in automated counting) may also require analysis by matrix, instrument or method group as the performance of the stabilised material may not be analogous to fresh blood.

Before the consensus summary statistics (mean or median, SD and CV) are calculated, it is necessary to remove outliers using appropriate statistical methods. Major blunders (e.g. resulting from transposition of the EQA test specimens or reporting in the incorrect units) should be removed, either by the removal of any result that is more than ± 3SD from the mean or by the exclusion of the top and bottom 5% of participants’ results. The resulting ‘trimmed’ consensus mean and SD are then recalculated. The median is used when there is a non-Gaussian distribution of data with a wide scatter of results. If the median is used the estimated SD is calculated as:

The target value for a quantitative test should be accompanied by an assessment of the uncertainty of the result, which gives a measure of its reliability.

Although outlier results are removed for the calculation of the target value and summary statistics, they should be returned to the data set for assessment of the individual participant’s performance.