Westgard rules for clinical QC, explained
Westgard rules are the statistical patterns a clinical laboratory uses to decide whether a quality-control (QC) run is in-control or out-of-control. First published by James Westgard in Clinical Chemistry in 1981, the six base rules — 1-2s, 1-3s, 2-2s, R-4s, 4-1s, 10-x — combine into multi-rule sets that catch both random and systematic analytical errors with a low false-rejection rate. This post is a working reference: the precise trigger condition for each rule, the clinical interpretation, why 1-2s is a warning rather than a rejection, the canonical multi-rule set, and how to read a Levey-Jennings chart that visualises everything together. This post is a standalone primer for anyone who needs the rules, independent of any product.
:::note LabFlow does not implement quality control today Quality control is planned and unbuilt — see the roadmap. Nothing below describes a LabFlow feature. :::
The vocabulary first
A QC measurement is a single reading of a control material — a manufactured substance whose target concentration is known. The lab runs the control through its analyser the same way it runs a patient sample, plots the measured value on a Levey-Jennings chart, and checks the rules. The vocabulary:
| Term | Meaning |
|---|---|
| Control material | A purchased liquid (or freeze-dried) with a known target concentration of an analyte (e.g. glucose at 60 mg/dL) |
| Control level | A specific control at a specific concentration. A lab typically runs two or three levels per analyte (low, normal, high) |
| Control lot | One manufactured batch of a control material with its own lot number, expiry, and manufacturer-assayed mean / SD |
| Mean | The average measured value for this lot on this analyser over many QC runs |
| SD (standard deviation) | A measure of how spread-out the measurements are around the mean |
| z-score | (measured − mean) / SD — how many standard deviations the current measurement is from the mean |
| In-control | The current QC run is within statistical expectation; patient reporting continues |
| Out-of-control | The current QC run violates a rejection rule; patient reporting locks until corrective action |
| Levey-Jennings chart | A time-series scatter plot of the QC measurements with horizontal bands at ±1 SD, ±2 SD, ±3 SD |
The mean and SD come from one of two sources. The manufacturer ships an assayed value with each lot (the population mean across many labs); the lab also computes an in-house mean / SD after 20 in-house measurements. The in-house values are typically tighter than the manufacturer's broader-population assay; the lab uses the in-house values once available.
The six base rules
Each rule's name encodes its trigger: "how many consecutive measurements, how many SDs from the mean, on which side".
1-2s — the warning, not the rejection
Trigger: 1 measurement is more than 2 SD from the mean (either side).
Verdict: Warning. Not a rejection rule on its own.
Why a warning: About 5% of in-control measurements naturally land outside ±2 SD by chance (the area under a normal distribution beyond ±2 SD is approximately 4.55%). Treating every 1-2s as a rejection produces a 5% false-rejection rate — far too high for a routine workflow. Westgard's original 1981 paper specifically warned against using 1-2s standalone.
How to use it: Use 1-2s as the trigger that escalates evaluation of the stricter rules. If 1-2s trips on a measurement, the system evaluates 1-3s, 2-2s, R-4s, 4-1s, 10-x in sequence. If none of those trip, the run is in-control with a 1-2s warning logged. If any of those trip, the run is out-of-control.
1-3s — the random-error catcher
Trigger: 1 measurement is more than 3 SD from the mean (either side).
Verdict: Reject.
Clinical interpretation: A measurement at >3 SD has a natural-occurrence probability of about 0.27% (twice the area beyond ±3 SD on a normal distribution). A 1-3s trip on a true in-control run is rare; it usually signals a single random error — a sample mishandling, a pipetting mistake, a one-off analyser glitch. Look at the most recent maintenance or operator change; re-run the QC.
2-2s — the systematic-shift catcher
Trigger: 2 consecutive measurements (on the same level or across the lab's run grid, see below) are more than 2 SD from the mean on the same side.
Verdict: Reject.
Clinical interpretation: Two consecutive measurements both >2 SD on the same side rarely happens by chance (probability about 0.05%). It usually signals a systematic shift — a new reagent lot needs recalibration, the analyser's calibration has drifted, an instrument temperature is off. The fix is a recalibration or reagent-lot change followed by re-QC.
The "same side" condition is key. Two measurements at +2.1 SD and -2.1 SD do not trip 2-2s; those would trip R-4s instead (random-error increased imprecision). The directional consistency is what makes 2-2s a systematic signal.
R-4s — the random-imprecision catcher
Trigger: Within the same QC run (or across consecutive runs on the same analyser + analyte), 2 measurements differ by more than 4 SD — one above +2 SD and one below −2 SD.
Verdict: Reject.
Clinical interpretation: Increased random error — the analyser is now imprecise. Common causes: a pipette is jamming, a reagent is partially expired, the analyser's sampling probe is contaminated. The fix is a hardware-side intervention (clean / replace / recalibrate) followed by re-QC.
4-1s — the smaller-systematic-shift catcher
Trigger: 4 consecutive measurements are more than 1 SD from the mean on the same side.
Verdict: Reject.
Clinical interpretation: Like 2-2s but with smaller individual deviations and a longer pattern. Systematic shift that's persistent but mild — often a recalibration or a reagent-lot change picked up before it grows into a 2-2s. Same fix as 2-2s.
10-x — the long-drift catcher
Trigger: 10 consecutive measurements are on the same side of the mean (regardless of how close to the mean they are).
Verdict: Reject.
Clinical interpretation: Long-running systematic drift — the analyser has slowly moved off-target. Common cause: a slow degradation in reagent (an opened vial sitting too long), a calibration that's slowly drifting, a temperature trending. The fix is a fresh calibration followed by re-QC. 10-x is sensitive to mild drifts that the other rules miss.
Some labs use 8-x or 12-x instead of 10-x; the exact number is a configurable parameter in most LIMS. The principle is the same.
The canonical multi-rule set
Westgard's canonical recommendation, still the most-used in 2026:
1-2sas warning → if trip, evaluate1-3s/2-2s/R-4s/4-1s/10-xin sequence. Reject on the first hit. If none, in-control with a 1-2s warning logged.
The evaluation order matters. The strictest rules (1-3s) run first; the weakest catchers (10-x) run last. A 1-3s trip on the current measurement is a rejection regardless of whether 4-1s would also trip — the system stops at the first rule that fires.
In a flow chart:
new measurement
↓
| is 1-2s tripped? |
↓ no ↓ yes
in-control check 1-3s
↓ no
check 2-2s
↓ no
check R-4s
↓ no
check 4-1s
↓ no
check 10-x
↓ no
in-control with 1-2s warning logged
Rules that need >1 measurement (2-2s, R-4s, 4-1s, 10-x) consider the current run plus the most-recent N−1 in-control runs on the same analyte + same analyzer + same level. The "same analyzer" qualification matters in a multi-instrument lab — a 10-x trend on Cobas-A should not lock reporting on Cobas-B, because the instruments are independent.
How to read a Levey-Jennings chart
A Levey-Jennings (LJ) chart plots QC measurements on a single analyte × level over time. The X-axis is the run number or timestamp; the Y-axis is the measured value. Three horizontal bands mark the SD lines:
value
|
+3 SD ─┼─────────────────────── reject band (1-3s)
|
+2 SD ─┼─────────────────────── warning band (1-2s)
|
+1 SD ─┼───────────────────────
|
mean ┼━━━━━━━━━━━━━━━━━━━━━━━ target
|
-1 SD ─┼───────────────────────
|
-2 SD ─┼─────────────────────── warning band (1-2s)
|
-3 SD ─┼─────────────────────── reject band (1-3s)
|
└───────────────────────────────────── time / run
Each QC measurement is a dot on the chart. Visual patterns to spot:
- Single outlier above +3 SD or below -3 SD — 1-3s rejection.
- Two consecutive dots both above +2 SD (or both below -2 SD) — 2-2s rejection.
- One dot above +2 SD and the next below -2 SD — R-4s rejection.
- Four consecutive dots above +1 SD (or below) — 4-1s rejection.
- Ten consecutive dots all above the mean (or all below) — 10-x rejection.
- A consistent slow climb or descent — drift; 10-x will catch it eventually but a human eye spots it sooner.
A good LJ chart also marks lot transitions (vertical dashed lines) because each new lot resets the in-house mean / SD computation, and it surfaces the violated rule code per point rather than only colouring the point.
What happens when a rule trips
When a rule rejects, the QC run is marked out-of-control and the system locks patient-result reporting for the (analyte + analyzer) pair until the lab acknowledges and acts. The action options:
| Action | When to use |
|---|---|
| Recalibrate and re-QC | The analyser's calibration has drifted; recalibrate and run the QC again |
| New reagent lot | The reagent is suspected (e.g. 2-2s after a lot change) — switch lots and re-QC |
| Operator error — re-run | An obvious one-off (a pipetting error) — re-run with attention |
| Instrument fault — service call | A hardware issue; the analyser needs a service technician |
| Override (with a documented clinical reason) | A lab manager overrides the lock for a documented reason; the override is audit-logged and surfaces on the affected patient reports as a footnote |
Patient reporting unlocks automatically when the next QC pass is in-control. The override path requires a Lab Manager with the explicit qc.override permission, and the audit row carries the manager's name and the reason.
Beyond the six base rules
The six base rules are the canonical set. Extensions exist for specific scenarios:
| Extension | Trigger | Use case |
|---|---|---|
2-3s | 2 consecutive >3 SD same side | Stricter version of 2-2s for highly-controlled methods |
3-1s | 3 consecutive >1 SD same side | Tighter version of 4-1s for shorter run grids |
7-T | 7 consecutive trending in the same direction | Detects gradual drift earlier than 10-x |
6-1s | 6 consecutive >1 SD same side | Intermediate between 4-1s and 10-x |
Most modern LIMS expose these as configurable rule combinations on a per-analyte basis. The default ship is usually the canonical Westgard multi-rule; the custom rules are opt-in.
Sigma metrics — the modern complement
Sigma metrics quantify a method's analytical performance as a "sigma value" — roughly, how many SDs of tolerance fit between the method's bias and its allowable error. The formula:
sigma = (TEa - |bias|) / CV
Where TEa is the total allowable error (often from CLIA limits), |bias| is the absolute systematic bias the method shows vs. a reference, and CV is the coefficient of variation (SD / mean × 100).
A method at sigma ≥ 6 is "world-class" — almost any QC rule set will catch errors before they reach patients. A method at sigma ≤ 3 is shaky and needs the strictest possible multi-rule set just to stay safe. The Westgard team published charts ("Westgard Sigma Rules") that prescribe which rule combination to use based on the method's sigma value:
| Sigma | Recommended multi-rule |
|---|---|
| ≥ 6 | 1-3s only (very simple) |
| 5 | 1-3s + 2-2s + R-4s |
| 4 | 1-3s + 2-2s + R-4s + 4-1s |
| ≤ 3 | Full canonical 1-2s/1-3s/2-2s/R-4s/4-1s/10-x |
Sigma metrics aren't enforced in rule evaluation; they're an evidence artefact for performance reviews and an input to choosing the multi-rule set. A lab tracks its sigma metrics per analyte and reviews them quarterly.
Frequently asked questions
Why isn't 1-2s a rejection rule?
Because about 5% of in-control runs naturally trip 1-2s by chance (the area beyond ±2 SD on a normal distribution). Treating every 1-2s as a rejection produces a 5% false-rejection rate — far too high to run a routine clinical workflow. Westgard's 1981 paper specifically warned against this; CLSI EP18 and EP23 echo the warning. Use 1-2s as the trigger that escalates evaluation of the stricter rules.
How many in-house measurements do I need before I can switch from manufacturer's mean / SD to in-house?
Twenty is the canonical answer — twenty in-house measurements on the same (level + lot + analyzer) lets the in-house mean and SD stabilise enough to use. Some labs use 25-30 for highly-variable analytes; tightening the threshold to 15 is sometimes done but produces noisier in-house statistics. Most modern LIMS default to 20 and let the lab override per analyte.
How does this work in a multi-instrument lab?
QC is tracked per (analyte + analyzer + level + lot). A lab with two Cobas analysers running the same test runs QC independently on each; the rule evaluation respects the instrument boundary so a 10-x trend on Cobas-A does not lock reporting on Cobas-B. This is the canonical pattern; any LIMS that mixes QC across instruments has a defect.
What's the difference between Westgard and CLSI rules?
CLSI EP23 and EP18 are the formal standards-body documents that codify and extend Westgard's work for the clinical laboratory. The CLSI documents are normative for accreditation purposes (CAP, CLIA, ISO 15189) and add the procedural overlay — how to run QC, how often, how to document overrides. The Westgard rules themselves remain unchanged; the CLSI documents are the canonical place to find them quoted in a regulator-accepted form.
Can I run Westgard rules on qualitative tests (positive / negative results)?
Not directly — Westgard rules are designed for quantitative measurements that follow a roughly normal distribution. Qualitative tests use a different QC scheme: typically, a positive control and a negative control are run with each batch, and the QC pass/fail is binary (did the positive control read positive and the negative control read negative?). The CLSI EP12 standard covers qualitative QC.
What's the false-rejection rate of the canonical multi-rule set?
About 0.03 to 0.05 percent per run — i.e. one false rejection in every 2000-3000 in-control runs. This is well below the 1% threshold most labs target. The false-rejection rate of a poorly-chosen rule set can climb much higher; the canonical multi-rule was specifically designed to keep this low while maintaining good error detection.
Do I need to learn the rules myself, or does the LIMS handle them?
A LIMS that implements QC handles the rule evaluation and surfaces the violated-rule code on the run; the lab manager needs to understand what each rejection means in order to choose the right corrective action. The clinical knowledge of "1-3s usually means random error → look at the operator or the most recent maintenance" is what the lab manager brings. (LabFlow does not evaluate QC rules today — see the roadmap.)
Are Westgard rules used outside clinical chemistry?
Yes — the same rules apply to any quantitative QC, including haematology, immunoassay, molecular diagnostics, environmental testing, and food safety. The "in-control / out-of-control" decision is method-agnostic; the rules are method-agnostic. The reference values (mean, SD) are method-specific; the rule logic is universal.
Where to read more
- What LabFlow has not built — including quality control.
- The original Westgard paper: J.O. Westgard, P.L. Barry, M.R. Hunt, T. Groth, "A Multi-Rule Shewhart Chart for Quality Control in Clinical Chemistry", Clinical Chemistry 1981.
- CLSI EP23 (statistical quality control for quantitative measurement procedures) and CLSI EP18 (laboratory quality control) — the accreditation-relevant documents.
- WestgardQC.com — the family's reference site, which has detailed worked examples and additional rule combinations.
About the author: Ahsan Mahmood is the engineer behind LabFlow. Contact at aoneahsan@gmail.com or via aoneahsan.com.