If you work in a lab, you’ll see Levey-Jennings (LJ) charts every day. They’re simple to draw, but reading one well is one of the most useful skills a young scientist can learn.
What’s on the chart
An LJ chart is one control material, plotted over time.
- The x-axis is time: the day or run number.
- The y-axis is the measured control value.
- The centre line is the mean for that control lot.
- The limit lines sit at the mean ±1s, ±2s and ±3s, where s is the standard deviation. In Westgard’s teaching example the mean line is green, ±2s is yellow and ±3s is red. The ±1s lines are added when you use multirules.
Each control level gets its own chart, because a low control and a high control have different means and SDs. Charts should also be labelled with the test, analyser, control lot number, and the mean and SD in use.
Where the mean and SD come from
The lines are only as good as the data behind them. The usual starting point is to analyse the control material at least 20 times over at least 10 days, and then calculate the mean and SD from that data.
Ideally that’s your data, from your analyser. Westgard notes that the ranges printed by manufacturers often cover performance across many systems in the field, so they can be much wider than is appropriate for QC in a single laboratory. Wide limits look reassuring but let real problems through.
What “in control” looks like
For a stable method, new control results should look like the old ones. Points bounce randomly either side of the mean. Most sit inside ±1s, a few stray past ±2s, and results beyond ±3s are very rare. With two controls per run, a 2s limit will flag about one good run in ten just by chance. That’s why a single 2s excursion is a warning, not a verdict.
Shifts vs trends
Two patterns tell you something systematic is going on.
A shift is a sudden jump to a new level that then stays there. The points cluster on one side of the mean, often starting on a specific day. When you see one, ask what changed that day. A new reagent lot, a new calibrator lot, a recalibration or a maintenance procedure are the usual suspects.
A trend is a gradual drift in one direction over several runs, with each point creeping a little higher (or lower) than the last. It suggests something slowly deteriorating. Recognised causes of systematic error include degrading reagents, detector drift, degrading instrument components and incorrect temperature settings.
The 2-2s, 4-1s and 10x multirules are the most sensitive to both patterns (see the companion article on Westgard rules).
Lot changes: the sneaky shift
Lot changes deserve special respect.
Reagent and calibrator lots. No two manufacturing lots are identical. A review in Clinical Biochemist Reviews describes real cases where an unrecognised lot change caused harm, including falsely raised PSA results at an Australian laboratory. The same review makes a key point: QC and external quality assurance (EQA) materials often aren’t commutable, meaning they don’t behave like real patient samples. Your QC might shift when patients don’t, or stay flat when patients do. That’s why new reagent lots are best evaluated with native patient samples. Small shifts can also add up across successive lots without any single change looking alarming. Moving patient averages can help catch that slow creep.
Control lots. When a new lot of control material arrives, its mean will differ from the old lot’s. The usual practice is to run the new lot alongside the current one for a while to establish its own mean before the old lot runs out. Accreditation checklists based on ISO 15189:2022 also expect labs to be able to detect reagent or calibrator lot changes, and to avoid changing the QC lot at the same time.
What to do when a rule fails
Westgard’s example QC protocol gives a sensible order of operations:
- Stop. Don’t report patient results from an out-of-control run.
- Identify the likely error type from the rule that failed. 1-3s and R-4s suggest random error. 2-2s, 4-1s and 10x suggest systematic error.
- Troubleshoot. Look at what’s changed since the last acceptable QC: lots, calibration, maintenance, reagent volumes, error flags.
- Correct the problem, then re-run controls to confirm the method is back in control.
- Deal with patient results. Repeat or verify the patient samples from the affected run. Accreditation requirements also expect labs to evaluate the patient results released since the last acceptable QC.
- Escalate. Consult a supervisor before any decision to release results from an out-of-control run.
Two common bad habits to avoid:
- Repeating the control until it passes. If the first result was a genuine signal, you’ve just ignored it.
- Recalibrating as a reflex. In a Westgard QC guest post, Nils Person explains that recalibration can temporarily “cover up” a developing problem. QC passes again, but the root cause is still there. Recalibrate when the investigation points to calibration, not as a first response.
One more detail: when a run is rejected, leave all of its control values out of your QC statistics, not just the one that broke the rule. And after you’ve fixed a problem, start fresh rather than applying across-run rules to data from before the fix.
Key takeaways
- Build your limits from your own stable data, not the manufacturer’s range.
- A shift is a sudden step. A trend is a slow drift. Both mean systematic error.
- Treat every reagent, calibrator and control lot change as a potential shift.
- When a rule fails: stop, investigate, fix, re-verify, then review the patient results affected.