Category: Basic QC Concepts | Estimated reading time: 8 minutes
When Charts Alone Are Not Enough
Levey-Jennings charts provide a rich visual overview. But visual overviews, no matter how rich, still leave room for subjectivity — two medical laboratory technologists (MLTs) looking at the same chart could make different decisions about whether the QC for that day should be accepted or rejected.
This is the problem James Westgard and his team sought to solve in 1981. The solution was not to replace the LJ chart, but rather to add a set of objective decision rules — clear, unambiguous criteria that could be applied consistently by anyone viewing the same chart. This is what we know as the Westgard Multirule.
To put it simply: The Westgard Rules are like a security alarm — they don't just alert you that something is wrong, but detect patterns of when an instrument is starting to become "unhealthy."
Why the Westgard Multirule Was Created
Before 1981, most laboratories used a single, simple rule: if a QC value fell outside ±2SD, reject; if it stayed within, accept.
The problem is, the ±2SD rule has a very high false rejection rate. Statistically, even under normal conditions, about 4.5% of all measurements will fall outside ±2SD simply due to normal random variation. This means nearly 1 in 20 QC runs will be rejected not because there is a real problem, but merely due to statistical chance.
In laboratories running two-level QC, this false rejection rate is even higher — creating unnecessary workloads, wasting reagents and time, and, more seriously, eroding trust in the QC system itself ("ah, rejected again, it’s probably just a false alarm").
Westgard and his team designed the multirule system with two balancing goals in mind:
- Maximize the error detection rate — the ability to detect truly real errors.
- Minimize the false rejection rate — reducing unnecessary rejections.
Balancing these two factors is a core challenge in the design of QC systems — and the Westgard Multirule has become the solution that remains the gold standard in medical laboratories worldwide to this day.
Enter target and SD, then input sequential control values to visualize the distribution pattern and detect rule violations (1 level control). Data is not stored.
Enter Target and SD (SD must be greater than 0) to display the chart.
This tool supports 1 level control and does not store data. R-4s and 2-2s inter-level rules require 2 levels and are not checked here. For multi-level monitoring, saved history, and in-depth analysis, use MyQCLab.
Six Rules, Two Functions
The Westgard Multirule consists of six rules that work in a hierarchy, rather than being applied separately. The first rule functions as a screening tool, and the subsequent rules serve as diagnostics to determine the type of error that has occurred.
1₂s — Warning Rule (Warning, Not Reject)
One point falls outside ±2SD
This is the only rule in the Westgard system that does not cause a rejection. It is simply a warning signal — an invitation to check subsequent rules more carefully.
Why not reject immediately? Because as explained above, about 4.5% of normal measurements will randomly fall outside ±2SD. A single point alone is not enough to conclude that there is a problem, but it is enough to trigger further inspection.
In practice: when 1₂s is triggered, do not immediately reject — check if any other rules were also triggered in the same run.
1₃s — Reject (Random Error)
One point falls outside ±3SD
The most powerful and clear rejection rule. Statistically, only 0.27% of normal measurements are expected to fall outside ±3SD — if this occurs, there is almost certainly a real error, not just random variation.
Type of error: random error — a sudden and dramatic event in one measurement. Possible causes: air bubbles in the sample, clots in the probe, incorrect dilution, or contamination of a newly opened control vial.
2₂s — Reject (Systematic Error)
Two consecutive points fall outside ±2SD in the same direction
Two consecutive points outside +2SD, or two consecutive points outside −2SD. Same direction is the keyword — two points outside ±2SD in opposite directions do not violate this rule.
Type of error: systematic error — something is consistently pushing the values in one direction. Possible causes: calibration starting to drift, a new reagent lot with different characteristics, or degradation of the control material.
This rule can be applied within one control level (two consecutive measurements at the same level) or across two control levels (one measurement at the normal level, one at the abnormal level, both outside ±2SD in the same direction).
R₄s — Reject (Random Error)
The range between two consecutive points exceeds 4SD
For example, one point is at +2.2SD and the next point is at −1.9SD — the total range is 4.1SD, thus R₄s is triggered.
Type of error: random error — a very large variation between two consecutive measurements indicates dramatic instability in the process. Possible causes: large variations in pipetting technique, extreme temperature fluctuations, or problems with the instrument's sample delivery system.
4₁s — Reject (Systematic Error)
Four consecutive points fall outside ±1SD in the same direction
Four consecutive points above +1SD, or four consecutive points below −1SD. Nothing passes ±2SD, but the pattern of consistency in one direction remains a significant signal.
Type of error: systematic error that develops more slowly than those detected by 2₂s. Possible causes: gradual calibration drift, slowly degrading reagents, or gradual changes in environmental conditions.
This rule can also be applied across two control levels: two consecutive points at the normal level and two consecutive points at the abnormal level, all on the same side of their respective means.
10x — Reject (Systematic Error)
Ten consecutive points fall on one side of the mean
Ten consecutive points above the mean, or ten consecutive points below the mean — even if none of them exceed ±1SD.
By probability, the chance of getting 10 consecutive points on one side of the mean under normal conditions is (0.5)¹⁰ = 0.1% — very small. If this occurs, it is almost certain that something systematic is taking place.
Type of error: systematic error that is very subtle and slow-developing — a small but very consistent shift. This is the most "patient" rule in the Westgard system: it takes the longest to be triggered, but once it is triggered, the signal is very strong.
How to Apply: Hierarchy, Not Parallel
A common mistake in applying the Westgard Multirule is trying to check all rules simultaneously — this is confusing. The correct way is hierarchical, step-by-step:
- Is there a point outside ±2SD? → If no, accept. If yes, proceed.
- Is there a point outside ±3SD? → If yes, reject (1₃s). If no, proceed.
- Are there two consecutive points outside ±2SD in the same direction? → If yes, reject (2₂s). If no, proceed.
- Does the range between two consecutive points exceed 4SD? → If yes, reject (R₄s). If no, proceed.
- Are there four consecutive points outside ±1SD in the same direction? → If yes, reject (4₁s). If no, proceed.
- Are there ten consecutive points on one side of the mean? → If yes, reject (10x). If no, accept.
The Relationship Between Westgard Rules and Sigma Metrics
There is one important relationship that is often overlooked: the higher the sigma metric of a method, the fewer Westgard rules need to be applied.
Laboratories whose methods have reached a sigma ≥ 6 can use only the 1₃s rule, because a highly stable process does not require such strict oversight. Conversely, methods with a sigma < 4 require a stricter combination of rules to ensure all significant errors are truly detected.
Westgard & Westgard (2016) developed the Normalized OPSpecs Chart concept, which allows laboratories to select the optimal combination of QC rules based on the sigma metrics of their methods — an approach that is far more rational than applying all rules by default to every parameter.
(Read more about how to calculate and interpret Sigma Metrics in the related article.)
Conclusion
The Westgard Multirule is not just about memorizing six rules. It is a carefully designed decision system to maximize error detection capabilities while minimizing false rejections — and if understood correctly, it provides diagnostic information about the type of error occurring, rather than just whether an error exists or not.
Like a good security alarm: it does not sound for every movement, only for those that are truly suspicious. And when it sounds, it tells you where and what is happening — not just that something is wrong.