1. Define the quantity, controls, and ordered sequence
Name the stable control or reference quantity, unit, transformation, method population, baseline, and records eligible for evaluation. Define whether multiple control observations occur within one run or one observation occurs across consecutive runs.
Bioassay potency, nonlinear parameters, changing control lots, unequal uncertainty, and censored estimates do not automatically behave like the clinical-chemistry measurements for which the original multirule procedure was developed. Justify the transfer before choosing notation.
2. Translate every rule name into executable logic
Specify the number of observations, SD boundary, side of the mean, within-run or across-run window, ordering, overlapping windows, reset behavior, treatment of missing and invalid records, and whether the outcome is a warning or action signal.
| Common notation | Pattern to define exactly | Possible role to evaluate |
|---|---|---|
| 1_2s | One observation beyond the same defined ±2 SD boundary | Warning or gate to another rule |
| 1_3s | One observation beyond ±3 SD | Large isolated departure |
| 2_2s | Two eligible observations beyond the same ±2 SD side | Possible systematic shift |
| R_4s | Two defined within-run controls separated by more than 4 SD | Possible increased random error |
| 4_1s | Four eligible observations beyond the same ±1 SD side | Persistent shift |
| 10_x | Ten eligible observations on one side of the mean | Sustained bias or shift |
3. Apply the rules to one ordered sequence
Suppose an approved baseline has mean 100% and SD 4 percentage points, so +1 SD is 104%, +2 SD is 108%, and +3 SD is 112%. Under a rule set where 1_2s is a warning and 2_2s is an action signal, evaluate the observations in order without rounding or skipping a point.
Run 1 at exactly 104% does not exceed +1 SD. Run 2 at 109% exceeds +2 SD but not +3 SD, so it creates a 1_2s warning. Run 3 at 110% also exceeds +2 SD on the same side; Runs 2 and 3 together trigger 2_2s. Neither point triggers 1_3s.
| Run | Observed recovery | Rule evaluation |
|---|---|---|
| 1 | 104% | At +1 SD; no beyond-limit rule |
| 2 | 109% | 1_2s warning: above +2 SD and below +3 SD |
| 3 | 110% | 2_2s signal with Run 2: two consecutive points above +2 SD |
4. Approve the baseline and error modes
Retain the baseline population, center, SD estimator, assumptions, effective date, and approval. Identify the random-error increase, systematic shift, drift, or other failure each proposed rule is intended to detect.
Separate monitoring rules from per-run assay suitability and product specifications. A rule can supply investigation evidence without being authorized to reject the current potency result.
5. Back-test stable behavior and designed changes
Apply the exact logic to representative historical or simulated stable sequences and to plausible shifts or variance increases. Estimate individual and combined false-signal frequency, detection probability or delay, and operational investigation burden.
Use independent evaluation where practical. Do not tune rules on the same incidents later presented as proof that the selected multirule detects them.
6. Approve the rule set and response before routine use
Document the rule versions, charted population, baseline, evaluation evidence, interaction and precedence, notification, triage, investigation, disposition authority, and change control. Decide which signals are informational, warning, or action-level.
Display the exact rule, involved points, values, limits, sequence, context, and calculation time. The system must not silently alter the run, exclusion state, or report disposition.
7. Investigate the signal in assay context
Review control and reagent lots, analyst, instrument, maintenance, preparation, timing, plate effects, method and software versions, suitability, exclusions, deviations, and nearby runs. Record evidence for and against the error mode the rule was designed to detect.
Preserve the original points and flag. A signal with no confirmed cause remains a retained investigation outcome rather than disappearing through rebaselining.
8. Retain action and reassess rule performance
Record disposition, affected records, correction, corrective or preventive action, approval, follow-up, and any baseline decision. Periodically review false-signal burden, missed patterns, detection delay, and whether process changes invalidate the evaluation.
Version changes to notation, windows, baseline, response, or charted population. Historical outcomes remain reproducible under the rule set active when they were generated.
Limits and current product support
No Westgard-style rule set is universal for bioassays. The selected quantity, error model, sequence, baseline, operating frequency, and acceptable false-signal burden determine whether any rule transfers usefully.
Current Provenarium trending does not provide a customer-configurable Westgard multirule library. This page is method guidance for evaluation and future or separately scoped monitoring workflows.
Frequently asked questions
Which Westgard rules should a bioassay use?
There is no universal set. Select and justify rules based on the charted quantity, baseline, error modes, false-signal rate, and procedural response.
Does a Westgard signal invalidate the current potency result?
Not by itself. The method and quality procedure define investigation and disposition; the signal supplies monitoring evidence.