Laboratory Quality Control: A Practical Westgard Rules Guide

Control limits are only as meaningful as the mean and standard deviation behind them. Manufacturer-assigned values on the control insert are a starting point for a new lot, not a permanent target — they are derived across many instruments and methods and are typically wider than your own performance. Using them indefinitely produces charts that almost never flag anything.

The laboratory should establish its own mean and SD from a run of its own data on the specific instrument and method, collected across multiple days so that between-day variation is represented. Twenty independent points is a common minimum, and the values should be reviewed when reagent lot, calibrator lot, or instrument service changes materially.

Lot changeover deserves its own routine. Overlapping the new control lot with the old one for a period lets you establish the new mean against known-stable performance instead of assuming a shift was caused by the lot. Skipping the overlap is how laboratories end up unable to distinguish a control problem from an analytical one.

A Levey-Jennings chart plots each control value against the mean with lines at one, two, and three standard deviations. Its usefulness is pattern recognition over time, not the judgement of any single point. A run of values drifting steadily upward tells you something quite different from a single point beyond two SD, even though the single point looks more alarming.

The two patterns worth naming are shift and trend. A shift is an abrupt move to a new level that then holds — typically a new reagent lot, a recalibration, or a maintenance event. A trend is a gradual movement across successive runs — typically deteriorating reagent, a drifting light source, an ageing electrode, or a slowly failing temperature control.

Charts are also how you notice the least dramatic failure mode: control values that hug the mean too tightly. Unnaturally low scatter usually means the control is not being treated like a patient sample, or that someone is repeating until the number looks acceptable. That behaviour destroys the entire purpose of the exercise.

Westgard multirules exist because a single rule cannot balance error detection against false rejection across tests with different analytical performance. Applying one rule set to every analyte is the most common misconfiguration in hospital laboratories. A test with wide allowable error and tight precision needs far less rule machinery than a test running close to its clinical limit.

The practical approach is to compare each test's imprecision against the total allowable error for that analyte, then select rules accordingly. Where the method is comfortably better than the requirement, a simple rule with two control levels is sufficient and will rarely produce false alarms. Where the method is marginal, multirules with more control observations are justified because the process genuinely needs closer watching.

Document the selection per analyte with the reasoning attached. Assessors ask why a rule set was chosen, and the answer should be a performance comparison rather than a vendor default. This also gives the laboratory a defensible basis for relaxing rules on tests that keep generating investigations without ever finding a real cause.

A rejection rule has meaning only if the laboratory has decided in advance what happens next. The policy should state that patient results from the affected run are held, that the cause is investigated before repeating, and that a repeat is permitted only after a defined corrective action. Repeating the control first and thinking later is the habit that turns quality control into theatre.

Where the investigation identifies a cause — expired reagent, a missed maintenance step, a calibration due — the corrective action is obvious and the run is repeated. Where no cause is found, the laboratory needs a stated limit on how many repeats are allowed before escalation. Two is a common ceiling, and escalation means a section head decision recorded with the run.

Held results have clinical consequences, so the policy needs a communication route. Where a run affects urgent or critical care samples, the treating team should be told that results are delayed rather than left waiting. A laboratory system such as HealUDoc can hold results in an unreleased state and surface the delay to the ordering clinician, which is preferable to a silent gap in the record.

Internal control tells you whether today matches your own history. External quality assessment tells you whether your history is correct. Both are required, and neither substitutes for the other — a laboratory can be beautifully precise around a wrong mean, and only an external comparison will reveal it.

Participation should cover the laboratory's scope, including tests with low volume, because rarely performed tests carry higher risk rather than lower. Where no scheme exists for a particular analyte, interlaboratory comparison with a peer laboratory using a documented protocol is the accepted alternative, and it needs the same record trail as formal participation.

The most important operational rule is that PT samples must be handled exactly like patient samples — same staff, same instrument, same procedure, no repeat testing, no consultation with other laboratories before submission. Special handling of PT material invalidates the whole exercise and is treated seriously by accreditation bodies.

An unacceptable external result is a documented non-conformity and should be treated with the same investigation discipline as a clinical incident. Start by confirming that it is real: check for transcription errors, unit mismatches, sample mix-up, and whether the reported value was submitted for the correct analyte. Clerical causes are common and are still causes worth recording.

If the value was genuinely produced, the investigation moves to method, calibration, reagent lot, and instrument state on the date the sample was run. Reviewing the internal QC charts around that date usually narrows the question quickly — a shift visible in-house that nobody actioned tells a different story from clean internal control alongside an external failure, which points toward calibration bias.

The closure step that laboratories skip is patient impact assessment. If a bias existed, the laboratory has to decide whether previously released results in that window need review or reissue, and record the decision either way. That assessment, more than the corrective action itself, is what an assessor will look for.