Internal quality control (IQC) decides, run by run, whether patient results may leave the laboratory. Done well, it catches the errors that matter and stays quiet otherwise. Done by habit, it raises false alarms every day and still lets real errors through. This page covers what ISO 15189:2022 requires, how control rules and limits work, and what to do when a control fails.
Two things at once: detect analytical errors large enough to harm a patient, and reject as few good runs as possible. Every QC design is a trade-off between error detection and false rejection, and the right balance differs from test to test. ISO 15189:2022 (clause 7.3.7.2) sets the frame. In summary, a laboratory has to:
Rules sensitive enough to catch an error that would change a clinical decision.
Rules calm enough that staff trust every alarm instead of repeating controls.
The most recent large picture comes from the 2025 Great Global QC Survey by Westgard QC, answered by more than 1,280 laboratories. The results show progress in some habits and a clear rise in daily QC alarms. They describe the laboratories that chose to take part, not the market as a whole.
| Survey result (global) | 2021 | 2025 |
|---|---|---|
| Out of control every day or several times a day | about 23 % | 33 % |
| Simply repeat the control when it fails | 68 % | 75 % |
| Apply 2 SD limits to every test | 59 % | 52 % |
| Use Westgard multirules | 86 % | 83 % |
| Calculate limits from their own mean and SD | 58 % | 70 % |
| Use the manufacturer’s ranges | 56 % | 46 % |
| Use third-party liquid assayed controls | 43 % | 49 % |
| Release patient results despite a failed run | 10 % | 6.9 % |
| Took no action at all on QC cost | 64 % | 54 % |
Source: Westgard QC, 2025 Great Global QC Survey. Survey results of self-selected respondents (USA 440, Asia 289, Middle East 146, Europe 143, Africa 118, Latin America 114), not market statistics. Central Asia and the Caucasus are not reported separately.
Internal QC habits, 2021 and 2025
Share of responding laboratories
Source: Westgard QC, 2025 Great Global QC Survey. Survey results of self-selected respondents (1,280+ laboratories in 2025), not market statistics; Central Asia is not reported separately.
Europe (143 laboratories): out of control every day, up from 27 % in 2021, although 76 % of the respondents are accredited to ISO 15189.
Asia (289 laboratories, India to Australia): still release results when QC is out of control; use of lyophilised third-party controls rose from 35 % to 52 %.
Middle East (146 laboratories): out of control every day, nearly double the 13 % of 2021.
Sources: Westgard QC, 2025 Great Global QC Survey — Europe, Asia, Middle East. No separate figures exist for Central Asia; the regional groups above are the closest reference points.
Yes, but not in the expected direction. In the same survey, 56 % of very large laboratories reported being out of control every day, against 16 % of small ones. More tests and more control measurements per day multiply the false rejections of badly chosen rules (Westgard QC, 2025 Great Global QC Survey, laboratories big and small).
Because a stable method still produces a control value outside ±2 SD about once in twenty measurements. Used as a rejection rule, 1-2s therefore rejects good runs by chance alone, and the rate grows with every additional control measured in the run. The figures below are plain statistics for a method without any error:
| Control measurements per run | Expected false rejections with a 1-2s rule |
|---|---|
| 1 | ≈ 5 % |
| 2 | ≈ 9 % |
| 3 | ≈ 13 % |
| 4 | ≈ 17 % |
Expected false rejections of a 1-2s rule
Stable method, no analytical error — bar length relative to 25 %
A laboratory with 40 tests and two controls each meets several such alarms a day. Staff learn that most alarms mean nothing, start repeating controls until they pass, and the one alarm that is real goes unnoticed. That is the mechanism behind the survey figures above. The background is explained in our article Why your lab is “out of control” every day.
Multirule QC combines several rules so that each looks for a different kind of error while the false-rejection rate stays low. The common rules, named after the number of results and the limit they use:
| Rule | Triggers when | Points to |
|---|---|---|
| 1‑2s | one result beyond ±2 SD | warning only, inspect the other rules |
| 1‑3s | one result beyond ±3 SD | random error |
| 2‑2s | two results beyond the same 2 SD limit (two levels in one run, or one level in two runs) | systematic error |
| R‑4s | the difference between two results in one run exceeds 4 SD | random error |
| 4‑1s | four consecutive results beyond the same 1 SD limit | systematic error (shift) |
| 10x | ten consecutive results on the same side of the mean | systematic error (drift) |
What the rules see on a Levey-Jennings chart
Illustrative control results of one level over 20 runs; red points violate a rule
The type of error matters for troubleshooting: random errors point to pipetting, bubbles, sample or reagent handling; systematic errors point to calibration, reagent or calibrator lots, and instrument components.
By measuring how good each method is against the quality the test needs. The sigma metric combines three numbers: the allowable total error (TEa), the bias and the imprecision (CV), all in percent: sigma = (TEa − |bias|) / CV. The higher the sigma, the less QC a test needs.
Example with illustrative values: a test with an allowable error of 10 %, a bias of 2 % and a CV of 1.5 % reaches (10 − 2) / 1.5 ≈ 5.3 sigma. The same test with a CV of 3 % reaches only 2.7 sigma and needs far more control effort, or a better method.
| Sigma | Typical QC design |
|---|---|
| 6 or more | a single rule with wide limits (1-3s), two controls per run |
| 5 to 6 | 1-3s / 2-2s / R-4s, two controls per run |
| 4 to 5 | full multirule including 4-1s, four control measurements (for example two levels in two runs) |
| below 4 | maximum QC, more frequent runs, and a review of the method itself |
Which QC design fits which sigma?
The higher the sigma, the less QC a test needs
Rule selection by sigma follows the approach published by Westgard QC; the table is a simplified summary. Sources for TEa: the US CLIA acceptance limits (42 CFR 493, subpart I, for glucose ±8 % or ±6 mg/dL), specifications from biological variation (EFLM database) and national rules such as the German RiliBÄK. Bias comes from EQA or reference material, CV from your own internal QC data.
Limits calculated from the laboratory’s own data, not the manufacturer’s range on the package insert. Manufacturer ranges have to fit every laboratory using the product and are therefore wide; with them, real shifts stay inside the limits.
The trend is moving in this direction: 70 % of the laboratories in the 2025 survey calculate limits from their own mean and SD, up from 58 % in 2021 (Westgard QC, 2025 Great Global QC Survey).
Manufacturer controls are optimised for one analyser and reagent system. That makes them convenient, but they can move together with a calibration shift and hide it. Independent third-party controls are not matched to one system, so they show such shifts, and one control can serve several analysers, which supports consolidation.
49 % of the 2025 survey respondents use third-party liquid assayed controls (2021: 43 %), and 20 % have consolidated their controls (Westgard QC, 2025 Great Global QC Survey). More in Third-party vs manufacturer controls; control and calibration material is part of our catalogue.
A failed run is a fixed sequence of decisions, not a repeat button. The order below keeps both the patient and the method in view:
In the 2025 survey, 75 % of the laboratories reported that their first reaction to a failed control is to repeat it (Westgard QC, 2025 Great Global QC Survey).
Internal QC shows whether a run is stable compared with the laboratory’s own history; external quality assessment shows whether the results are right compared with everyone else. A calibration bias passes internal QC unnoticed as long as the limits move with it, and only EQA reveals it. Read EQA results together with the internal QC data of the same period, as described in Evaluating EQA results, and use the bias from EQA in the sigma calculation above.
