Westgard Rules Explained

Each name is a count and a limit. 1-3s is one result beyond 3 SD; 2-2s two results beyond the same 2 SD line; 4-1s four beyond the same 1 SD line; 10x ten on one side of the mean. R-4s is the odd one: a range, one result above +2s and one below −2s. Westgard writes the limits as subscripts; you will also see 1:2s and 12s for the same rule.

Two more letters run through every QC manual. N is the number of control results available when you decide on a run: two levels run once is N = 2. R is the number of runs a rule looks back over. The rules were built for Ns of 2 or 4, which is why they fit a chemistry bench running two levels.

To read two levels with one set of rules, each result becomes an SD index: (result − mean) ÷ SD. Westgard's multirule FAQ describes plotting those values on one chart whose limits are ±1, ±2 and ±3. A level 1 glucose at +1.4 and a level 2 at +1.2 are then “the same side, beyond 1s”, whatever their units.

“Westgard rules” is the bench's name, not his. He prefers “multirule QC”, and the familiar 1-3s/2-2s/R-4s/4-1s/10x set was published as an example, not a prescription for every test.

One control result beyond the mean ± 2 SD. In Westgard's original multirule it is a warning: it tells you to inspect the run with the rejection rules, “like a yield or warning sign”. It does not mean stop.

Used as a rejection rule on its own, which is what a chart with only 2s limits does, it throws away good runs. Westgard puts the false rejections at:

Reject the run when a single control result exceeds the mean + 3 SD or the mean − 3 SD. It is the rule a Levey-Jennings chart with 3s limits applies, and it mostly catches random error: the method's scatter has grown. On its own it rarely cries wolf, about 1% false rejections with 2 to 4 controls a run, but it also misses smaller errors, which is why the other rules exist.

Reject when 2 consecutive control measurements exceed the same mean + 2s or the same mean − 2s. Both on the same side: one high and one low is a different rule. It catches systematic error, and where you see it says something about where the error sits:

• Both levels in one run (across materials): the shift reaches across the measuring range. Westgard's lesson 18 reads it as error “affecting the results throughout the critical analytical range”.
• One level in two runs (within a material, across runs): the shift is at one end. The same lesson reads a high control out twice as consistent with a loss of linearity at the high end.

Reject when one control measurement in a group exceeds the mean + 2s and another exceeds the mean − 2s. It is a counting rule, not a measured range: Westgard's answer is that +2.6 SD and −1.8 SD is not a violation, even though the two are 4.4 SD apart. It catches random error.

It is read within a run only. Westgard's note on the rule is plain: it “should only be interpreted within-run, not between-run”. Across runs it would also fire on a shift between runs and make a systematic error look random. With two control levels it is applied across the two levels of the same run.

Reject when 4 consecutive control measurements exceed the same mean + 1s or the same mean − 1s. With two levels a run, the four can be the last two runs on both levels, or the last four runs on one level. None of them has to be far out. It catches systematic error that persists from run to run, and Westgard's power curves show the look-back over two runs adding detection that one run alone cannot.

Under the classic protocol this chart is accepted: no result passed 2s, so nothing triggered a look. See warning or trigger below.

Reject when 10 consecutive control measurements fall on one side of the mean. It usually has to be applied across runs, and often across materials: with two levels a run, that is the last five runs. Variants fit other control counts: 8x and 12x for Ns of 4, and 6x or 9x with three levels. It catches systematic error too small for any limit line to show.

In the original paper-chart protocol, 1-2s is the gate: if no result passes 2s, you accept the run without reading the other rules. That saved time when every point was plotted by hand. The cost is that 4-1s and 10x can only fire on a run that also has a 2s result.

Lesson 18 shows the gap. In its run 12, four results in a row sit above +1s and none passes 2s; under the protocol the run is accepted, though “a computer program that inspects the data without use of the 1-2s warning would flag this run as being out-of-control”. Westgard's own advice for software is that the warning rule is not needed: a computer can check every rule on every run.

1. Hold the run. Control results must be acceptable “before reporting patient test results” (§ 493.1256(f)).
2. Name the error from the rule. 1-3s or R-4s: random. 2-2s, 4-1s or 10x: systematic. Which levels broke it tells you which end of the range to look at.
3. Find and fix the cause with the analyzer's troubleshooting guide. Start with what changed: CLIA already makes you rerun controls after a complete change of reagents, major preventive maintenance or a critical part replaced (§ 493.1256(d)(6)).
4. Rerun both control levels and confirm the method is back in control.
5. Repeat or verify the patient samples from the failed run.
6. Look back further. Every patient result since the last good run is in question, not just this run's.
7. Write it down. The corrective action is a record an inspector reads. Westgard adds: consult a supervisor for any decision to report patient results when a run is out of control.

Often not. Westgard's best practices start from the test's Sigma metric: (allowable total error − bias) ÷ CV, all in percent. Bias comes from proficiency testing or a peer comparison program; CV from your own control data (the CV calculator gives it, and the sigma metric calculator does the sum). The better the method, the less QC it needs:

These are the single-rule designs from that lesson. Westgard's later Westgard Sigma Rules give multirule choices for the same Sigma bands (1-3s alone at 6 Sigma; 1-3s/2-2s/R-4s at 5), and the sigma metric calculator follows those.

Peer data is where most small labs get the bias figure: see the QC peer data programs compared. Proficiency testing gives a second estimate: CMS describes PT samples arriving three times a year ; the CAP and API comparison covers the main providers.

Hematology, coagulation and many immunoassays run three levels, and the two-level rules do not divide into threes. Westgard's rules page gives the set that fits:

• 2of3 2s: reject when 2 out of 3 control measurements exceed the same mean + 2s or mean − 2s.
• 3-1s: reject when 3 consecutive control measurements exceed the same mean + 1s or mean − 1s.
• 6x: reject when 6 consecutive control measurements fall on one side of the mean; 9x looks back further.

His best-practice list writes the three-level procedure as 1-3s/2of3 2s/R-4s/3-1s/12x. Using 8x or 10x with three levels makes no sense: they would look back over a fraction of a run.

No section of 42 CFR 493 names the Westgard rules. CLIA asks for the parts the rules sit on, and leaves the criteria to the lab:

The lab must “establish or verify the criteria for acceptability of all control materials” (§ 493.1256(d)(10)) and “document all control procedures performed” ((g)). The rules you choose are those criteria, and the runs that broke them, with what you did, are the record. How often the controls run is its own rule: see how often to run QC.