Is the laboratory result in control, and is it normal?

Start here

Why run a control sample when the analyser has already been calibrated?

A calibration is correct at the moment it is done. Reagents deteriorate, kit lots change, pipette calibration changes and operators change. A control sample with an expected value, run with the patient samples, shows whether the method is still giving the same answer today.

Internal quality control (IQC) is the sample material and procedures that verify that the intended quality of results has been reached, including monitoring the day-to-day reproducibility of results and detecting major errors in the analytical process. External quality assessment (EQA) is an external evaluation of a laboratory's performance using panel samples whose values are known to the organiser but not disclosed to the laboratory. Both definitions are from the NBTC operational guidelines for blood centres (2024).

Control materials

What makes a good control material?

A control material is a sample you run like a patient sample but whose expected range you have already established. It is useful only if it behaves like patient samples and tests the method where clinical decisions are made.

1. Stable for the property being measured, so that changes on the chart come from the method and not from the control.
2. Matrix as close as possible to patient samples, so that it reacts to the method in the same way.
3. Concentrations at or near decision limits, and where possible covering the measuring range. In clinical chemistry, controls are usually available in two or three levels covering the medical decision points.
4. Third-party material should be considered, instead of or in addition to controls from the reagent or instrument manufacturer.
5. Lot changes kept apart: do not change the IQC lot on the same day or run as a reagent or calibrator lot change, otherwise a shift cannot be traced to its cause.
6. For qualitative screening tests such as TTI ELISAs, a borderline-reactive control shows small changes that a strongly positive kit control does not.

Items 1 to 5 follow the IQC requirements in the NBTC operational guidelines (2024). Item 6 is from Dubey and Sonker (2021). Before plotting, establish the control's own mean and SD: Westgard QC recommends a minimum of 20 measurements over at least 10 days.

The chart

How do you draw a Levey-Jennings chart?

Measure the control at least 20 times to learn its usual mean and SD. Draw horizontal lines at the mean and at 1, 2 and 3 SDs above and below it. Then plot each new control result in time order.

Worked example 1: setting the limits

Made-up teaching data. Level 1 glucose control, 20 baseline results (mg/dL): 98, 101, 103, 99, 100, 97, 102, 104, 100, 96, 101, 99, 103, 98, 100, 102, 97, 101, 99, 100.

1. Mean = 2000 ÷ 20 = 100.0 mg/dL.
2. SD (dividing by n − 1 = 19) = 2.18 mg/dL. CV = 2.18 ÷ 100 × 100 = 2.2%.
3. ±1 SD: 97.8 to 102.2. ±2 SD: 95.6 to 104.4. ±3 SD: 93.5 to 106.5 mg/dL.

Why 2 SD limits alone reject good runs

If the method is stable and results are normally distributed, one control falls beyond ±2 SD in 4.55% of runs and beyond ±3 SD in 0.27%. With two controls in a run, at least one falls beyond ±2 SD in 1 − 0.9545² = 8.9% of perfectly good runs. Westgard QC gives the same figure, about 9%, for a Levey-Jennings chart with 2 SD limits and N = 2. Limits at 3 SD reject far fewer good runs (0.54% with two controls) but detect fewer real errors. Multirule QC was designed to keep false rejections low while keeping error detection high.

Westgard rules

Which Westgard rule was broken, and what does it tell you?

Westgard, Barry, Hunt and Groth (1981) combined several control rules on one Levey-Jennings chart. Each rule looks for a different pattern, and the pattern suggests the type of error.

• Manual use: 1-2s is a warning that triggers inspection with the rejection rules. Computer use: the warning step is not needed, because the software can apply every rejection rule to every run.
• When only three control materials are run, rules that fit multiples of 3 are used instead: 2of3-2s, 3-1s, 6x or 9x. Variants 8x and 12x fit N = 4.
• Opening problem: runs 3 (105) and 4 (106) are both beyond +2 SD (104.4), so the 2-2s rule rejects run 4. Systematic error is the likely cause, and the ward samples should not be released.

Interactive

Add a shift or extra scatter and watch which Westgard rules reject the runs

A shift moves every result the same way, so rules that look at several results on one side (2-2s, 4-1s, 10x) react first. Extra scatter makes single extreme results and results on opposite sides more likely, so 1-3s and R-4s react.

Decide

A control result is outside its limits: what do you do?

Work through the rules in order. Every rejected run needs a cause, a correction and repeated patient samples.

After a rejection, patient results tested since the last successful IQC event must be evaluated (NBTC 2024). Results from before the correction are not used to apply the across-run rules again.

Comparison

Random or systematic error: how do they differ?

The two errors have different causes and different fixes. The rule that was broken points you to one kind of cause first.

External quality assessment

What does external quality assessment show that your own controls cannot?

Your IQC chart compares your method with itself. If every result is 5% too high every day, the chart stays calm. EQA sends the same undisclosed sample to many laboratories and compares your result with theirs.

z = (x − μ) ÷ σ

x

your laboratory's result on the EQA sample

μ

mean of the EQA participants (the assigned value)

σ

SD of the EQA participants

Worked example 2 (made-up): your creatinine result is 1.32 mg/dL, the peer mean is 1.20 and the peer SD is 0.05. z = (1.32 − 1.20) ÷ 0.05 = 2.4. Peer-comparison reports use the same idea as the standard deviation index: SDI = (laboratory mean − group mean) ÷ group SD. An SDI within ±1.0 is acceptable, 1.0 to 1.9 means a problem may exist and should be investigated, and 2.0 or more calls for troubleshooting of bias and corrective action.

Proficiency testing is the structured evaluation of laboratory methods to assess the suitability of processes, procedures, equipment, materials and personnel. EQA providers do it by distributing material for testing. Interlaboratory comparison is the organisation, performance and evaluation of tests on the same or similar items by two or more laboratories under predetermined conditions.

India

What do India's operational guidelines for NEQA in blood centres require?

The Blood Transfusion Services division of the Directorate General of Health Services published Operational Guidelines for National External Quality Assessment Programme for Blood Centres in India, first edition 2024.

Sigma metrics

How many sigmas does your test have?

Decide how far a result may be from the true value (the allowable total error, TEa). Take away the bias. Count how many SDs of your method fit into what is left. That count is the sigma metric.

Sigma = (TEa% − |bias%|) ÷ CV%

TEa

allowable total error for the test, in percent

bias

systematic error, usually from EQA or peer-group data, as an absolute percentage

CV

imprecision from internal QC data, in percent

Worked example 3 (made-up): TEa 10%, bias 2%, CV 2%: sigma = (10 − 2) ÷ 2 = 4.0. With bias 1% and CV 1.5%: (10 − 1) ÷ 1.5 = 6.0. The quality goal index (QGI) = bias ÷ (1.5 × CV) shows why a method falls short: below 0.8 points to imprecision, above 1.2 to inaccuracy. For method A, QGI = 2 ÷ 3 = 0.67, so imprecision is the main problem.

The short-term scale assumes a later shift of 1.5 SD, so 6 sigma corresponds to 4.5 SDs and 3.4 defects per million. The same sigma can come out higher or lower depending on the TEa source chosen (CLIA, biological variation or RiliBÄK), so always state the source.

Interactive

Change TEa, bias and CV and watch the sigma metric and the suggested QC rules

Bias moves the curve towards one limit and CV widens it. Sigma counts how many SDs still fit between the bias and the TEa limit.

Reference intervals

How is a reference interval established?

A reference interval describes the central 95% of results in carefully selected healthy people. It is built in four stages: selecting the people, collecting the samples, measuring them, and calculating the limits.

1. Select reference individuals. Use a questionnaire with exclusion criteria (conditions that prevent inclusion) and partition criteria (age, sex). Obtain written informed consent. In a priori sampling, exclusions are applied before collection. In a posteriori sampling, after. Both are direct methods.
2. Control pre-analytical factors. Biological: sampling time, fasting, physical activity. Methodological: collection technique, additives, tourniquet use, handling, transport, centrifugation time and speed, storage.
3. Control analytical factors. Define the method, instrument, reagents and calibrators with traceability, and run a quality control programme with clear criteria.
4. Analyse the values. Remove outliers (for example Dixon's D/R ratio above 1/3, or Tukey fences after Box-Cox transformation). Decide on partitions (for example a standard deviation ratio above 0.3). Take the 2.5th and 97.5th percentiles, with 90% confidence intervals for both limits.

The parametric method assumes a normal distribution (after transformation if needed) and uses mean ± 1.96 SD. CLSI recommends the non-parametric method, although an IFCC multicentre study found the two methods gave very close results. Minimum sizes quoted in the guidance: 120 per partition, so 240 for separate male and female intervals. The IFCC multicentre protocol sets a minimum of 500 per country.

Interactive

Change the number of healthy people and watch how precise the reference limits become

Each limit depends on the few most extreme values. With few people, those values change a lot from one study to the next.

Decide

Can you use the reference interval from the kit insert or another laboratory?

Establishing an interval is expensive. CLSI EP28-A3c allows a laboratory to adopt an existing interval if it checks that the interval fits its own method and population.

• Transference: when the method changes, compare the new and old methods (similar imprecision, known interferences, the same or comparable calibrators, comparable values) and transfer the limits by linear regression. The reference populations must also be comparable.
• Verification: test 20 healthy people from your own population. If no more than 2 of 20 results (10%) fall outside the interval, adopt it, at least provisionally. If more than 2 fall outside, test a second 20. If 3 or more fall outside again, re-examine the analytical procedure and the biological differences between the populations.

Comparison

Is a result outside the reference interval abnormal?

Not necessarily. By definition, 5% of results from healthy people fall outside the reference interval. A clinical decision limit answers a different question.

Before and after the analyser

What else affects the quality of a test report?

IQC and EQA check the analytical phase. Most laboratory errors happen before the sample reaches the analyser or after the result leaves it.

In a real paper

Sigma metrics and reference intervals in Indian laboratories

How to run it

How do you draw control charts and calculate reference intervals in software?

For a thesis or an audit you may need to reproduce a control chart or calculate a reference interval yourself. Each step below was checked in the official documentation.

IBM SPSS Statistics (Statistics Base)

1. Analyze > Quality Control > Control Charts.
2. Select Individuals, Moving Range, choose the data organisation, and click Define.
3. Move the control results into Process Measurement, and a run or date variable into Subgroups Labeled By.
4. Click Control Rules to flag out-of-control points. The SPSS rule list (for example, greater than +3 sigma, 2 points out of the last 3 greater than +2 sigma, 8 consecutive points above the centre line) is not the Westgard multirule set, so check the rules you select.

R

library(qcc)                     # CRAN package qcc, version 2.7
baseline <- c(98, 101, 103, 99, 100, 97, 102, 104, 100, 96,
              101, 99, 103, 98, 100, 102, 97, 101, 99, 100)
new_runs <- c(101, 99, 105, 106, 100, 98, 93, 102, 104, 103, 103, 104, 97, 95)
# centre and SD fixed from the baseline, limits at 3 SD (nsigmas = 3 is the default)
qcc(new_runs, type = "xbar.one", center = mean(baseline), std.dev = sd(baseline))

library(referenceIntervals)      # version 1.3.1
# non-parametric interval (RI = "n") with non-parametric 90% CIs (CI = "n", needs n of 120 or more)
refLimit(values, out.method = "horn", out.rm = FALSE, RI = "n", CI = "n",
         refConf = 0.95, limitConf = 0.90)

library(refineR)                 # version 2.0.0: indirect method from routine results
fit <- findRI(Data = routine_results)
getRI(fit, RIperc = c(0.025, 0.975))

In referenceIntervals, out.rm = FALSE detects outliers without removing them. With fewer than 120 values, refLimit uses bootstrapping for the confidence intervals. jamovi: no control chart or reference interval analysis in the core program could be verified, so no jamovi steps are given.

How to report it

What should you write in a thesis or audit report?

Give enough detail for a reader to repeat the QC decision or the interval: the materials, the numbers, the rules and the sources of every target.

• IQC: control materials (manufacturer or third party, levels), how the mean and SD were set (number of measurements, days), the rules, N per run, and the frequency.
• Sigma: TEa and its source, where bias and CV came from, the period, and the level of control.
• EQA: the scheme, number of cycles, and z-score or SDI per cycle.
• Reference interval: selection criteria for reference individuals, pre-analytical conditions, method, n per partition, outlier method, parametric or non-parametric estimation, and 90% confidence intervals for each limit (recommended by IFCC since 1987).

Worked sentence (teaching numbers): “Glucose was monitored with a third-party level 1 control in every run. The control mean (100.0 mg/dL) and SD (2.18 mg/dL, CV 2.2%) were set from 20 baseline measurements. Runs were judged with 1-3s/2-2s/R-4s/4-1s/10x rules. With TEa 10% (source stated), bias 2.0% from EQA and CV 2.0% from IQC, the sigma metric was 4.0 (QGI 0.67). In the EQA cycle, creatinine gave z = 2.4 and corrective action was recorded. Reference limits were the 2.5th and 97.5th percentiles by the non-parametric rank method in 120 healthy adults per partition, with 90% confidence intervals.”

By specialty

Quality control and reference interval problems by specialty

Common myths

Common mistaken beliefs

• ✗Myth: If the controls pass, the patient results are correct.

  ✓IQC checks the analytical phase. In the Padova studies, most laboratory errors were pre-analytical (68.2% in 1996, 61.9% in 2006).

• ✗Myth: Any control beyond 2 SD means the run must be rejected.

  ✓In the multirule procedure, 1-2s is a warning. Used as a rejection rule with two controls, it rejects about 9% of good runs.

• ✗Myth: IQC detects only random error and EQA detects only systematic error.

  ✓IQC rules such as 2-2s, 4-1s and 10x detect systematic error. EQA mainly shows bias, but a single EQA result also contains that day's random error.

• ✗Myth: The reference interval on the kit insert can be used as it is.

  ✓CLSI EP28-A3c asks laboratories to verify it with 20 healthy people from their own population, or to establish their own.

• ✗Myth: A result outside the reference interval means disease.

  ✓5% of results from healthy people fall outside the interval. Diagnosis uses decision limits from clinical studies.

• ✗Myth: A test with a high sigma needs more QC.

  ✓The opposite. At 6 sigma, one 1-3s rule with two controls is enough. Below 4 sigma, more rules and more controls are needed.

Why this matters to you

Five consequences of getting laboratory QC wrong

1. Wrong results released

   Releasing results after a broken rejection rule sends patients' results with a known error to the ward. The NBTC guidelines require laboratories to prevent release when IQC fails.

2. Good runs thrown away

   Rejecting every run with a result beyond 2 SD wastes reagents, time and samples on about 9% of runs that had nothing wrong.

3. Failed EQA and accreditation

   Participation in EQA is often an accreditation requirement. An unexplained bias found in EQA must be investigated and corrected.

4. Misclassified patients

   Adult intervals applied to newborns, or non-pregnant intervals applied in pregnancy, label healthy patients abnormal and miss real abnormalities.

5. Questions you cannot answer

   Westgard rules, IQC, EQA and reference intervals have been set repeatedly in MD Biochemistry and Transfusion Medicine papers since 2011.

History

History of laboratory quality control and reference intervals

1. 1924

   Walter A. Shewhart

   In a memorandum of 16 May 1924 at Western Electric, he proposed the statistical control chart to separate chance-cause from assignable-cause variation.

2. 1931

   Walter A. Shewhart

   Published Economic Control of Quality of Manufactured Product.

3. 1950

   S. Levey and E. R. Jennings

   Published "The use of control charts in the clinical laboratory" in the American Journal of Clinical Pathology.

4. 1952

   R. J. Henry and M. Segalove

   Described the running of standards in clinical chemistry and the use of the control chart in the Journal of Clinical Pathology.

5. 1977

   J. O. Westgard and colleagues

   Published the probabilities of false rejection and error detection for internal quality control rules.

6. 1981

   Westgard, Barry, Hunt and Groth

   Published "A multi-rule Shewhart chart for quality control in clinical chemistry", the basis of the Westgard rules.

7. 1986

   Bill Smith, Motorola

   Introduced the Six Sigma quality improvement method at Motorola.

8. 1987–1991

   IFCC

   Published a series of six papers recommending defined procedures for each laboratory to produce its own reference values.

9. 2000

   Medical laboratories

   The first paper expressing laboratory processes on the sigma scale was published.

10. 2008

    CLSI and IFCC

    Published guideline C28-A3 on defining, establishing and verifying reference intervals, now EP28-A3c, reaffirmed without revision in April 2020.

11. 2024

    Dte.GHS, Blood Transfusion Services

    Issued the first edition of the Operational Guidelines for National External Quality Assessment Programme for Blood Centres in India.

Guidelines

Quality control and reference intervals in guidelines

Reading a paper

Five checks on a QC or reference interval study

• Is the source of TEa stated, and would another source change the sigma values?
• Were CV and bias taken from the same period and the same control level?
• For a reference interval: how many reference individuals per partition, and were they selected with stated exclusion criteria?
• Are the outlier method, the estimation method (parametric or non-parametric) and 90% confidence intervals reported?
• Is a reference interval being used as if it were a decision limit, or the other way round?

Why is it called that?

Word surgery

Exam answers

Model answer outlines

Past papers

TNMGRMU questions on laboratory quality control and reference intervals

Questions from MD, MS, DM, MCh and Ph.D. papers that this chapter helps you answer. Tap a sitting to open the university's own question paper.

1. ×3Westgard Multi QC rules and their interpretationMD Biochemistry · Oct 2022MD Biochemistry · Nov 2020MD Biochemistry · Apr 2015
2. ×2Westgard rulesMD Immuno Haematology and Blood Transfusion · May 2022MD Biochemistry · Oct 2016
3. ×2Role of external quality assessment in transfusion medicineMD Immuno Haematology and Blood Transfusion · Apr 2015MD Immuno Haematology and Blood Transfusion · Apr 2015
4. ×2Internal Quality ControlMD Biochemistry · Oct 2015MS OBSTETRICS AND GYNAECOLOGY · Oct 2015
5. ×1Enumerate the salient features of recent Guidelines for National External Quality Assessment Programme for Blood Centres in IndiaMD Immuno Haematology and Blood Transfusion · May 2025
6. ×1Internal quality control program in labMD Biochemistry · Oct 2025
7. ×1External quality assessment in laboratory haematologyDM Clinical Haematology · Aug 2022
8. ×1Levey-Jennings (LJ) chart in transfusion transmitted disease testingMD Immuno Haematology and Blood Transfusion · Aug 2020
9. ×1Requirements for implementation of internal quality control programme in a laboratory . How to monitor and interpret the IQC values ?MD Biochemistry · Aug 2020
10. ×1Write briefly on quality control in laboratory. What are the variables that affect quality of the test reports?MD Biochemistry · Nov 2020
11. ×1Quality control chartsMD Biochemistry · Oct 2018
12. ×1Internal quality control for plasma clotting timesDM Clinical Haematology · Aug 2018
13. ×1Describe in detail the strategies for establishing reference intervals with respect to selection of reference individuals, specimen collection, Quality control procedures and statistical treatment of valuesMD Biochemistry · May 2017
14. ×1What are the criteria for choosing an internal quality control material and how will you interpret and monitor IQC values in your clinical chemistry lab?MD Biochemistry · May 2017
15. ×1Six sigma principlesMD Biochemistry · Oct 2017
16. ×1Reference intervalMD Biochemistry · Apr 2016
17. ×1Internal Quality Control is a measure of Random Error whereas external Quality Control is a measure of Systematic error – JustifyMD Biochemistry · Apr 2015
18. ×1Name 2 Westgard rules that detect random errorMD Biochemistry · Apr 2014
19. ×1Establishment of reference intervalsMD Biochemistry · May 2011

Further reading

Related Stateazy articles