13.1 Statistical Quality Control, Westgard Multi-rules & Levey-Jennings Chart Analysis
Key Takeaways
- Statistical quality control relies on Gaussian distributions where 68.2% of values fall within ±1 SD, 95.5% within ±2 SD, and 99.7% within ±3 SD; the Coefficient of Variation (CV% = [SD / Mean] * 100) is a unitless index of relative imprecision enabling direct comparisons across instruments and methods.
- Systematic error reflects loss of accuracy and manifests on Levey-Jennings charts as shifts (abrupt step changes persisting on one side of the mean) or trends (gradual progressive drift over ≥6 days), caused by calibration degradation, reagent lot changes, or failing light sources.
- Random error reflects loss of precision and manifests as increased chart scatter or isolated outliers, caused by electrical noise, air bubbles in hydraulic lines, pipetting inconsistencies, or sample clots.
- Under the CLSI C24 Westgard multi-rule algorithm, 1_2s is strictly a warning rule and must NEVER trigger run rejection alone; 1_3s and R_4s detect random error, whereas 2_2s, 4_1s, and 10_x detect systematic error and mandate run rejection.
- When an analytical run is rejected, patient testing must halt immediately, results must be held, and a protocolized root-cause troubleshooting sequence must be executed and documented prior to recalibration and resuming testing.
Last updated: September 2026
13.1 Statistical Quality Control, Westgard Multi-rules & Levey-Jennings Chart Analysis
[!NOTE] Clinical Laboratory Quality Imperative: In clinical chemistry, analytical errors directly compromise patient safety and clinical decision-making. Statistical process control provides an objective mathematical framework to verify that analytical methods maintain acceptable accuracy and precision before patient specimens are analyzed and reported. Mastering Westgard rules, Levey-Jennings chart patterns, and error remediation is a core requirement for the C(ASCP) technologist.
Foundations of Statistical Quality Control
Statistical Quality Control (SQC) in the clinical chemistry laboratory involves the daily analysis of stable control materials with known analyte concentrations alongside patient specimens. By comparing measured control values against established target ranges, the laboratory monitors analytical performance and detects assay deterioration.
Statistical Parameters: Mean, Standard Deviation, and Coefficient of Variation
- Mean (Central Tendency, $\bar{x}$):
- The arithmetic average of a set of replicate measurements:
- In quality control, the target mean represents the set point or expected value for a given control material lot. It reflects the accuracy (trueness) of the analytical system when calibrated correctly.
- Standard Deviation (Dispersion, $s$ or SD):
- A mathematical measure of the spread or dispersion of individual observations around the mean:
- Standard deviation quantifies the imprecision (random variation) of the analytical method. A smaller standard deviation indicates tighter clustering of data points around the mean and superior analytical repeatability.
- Coefficient of Variation ($CV%$):
- The standard deviation expressed as a percentage of the mean:
- Clinical Utility: While standard deviation is expressed in the concentration units of the analyte (e.g., mg/dL, mmol/L) and varies directly with analyte concentration, the Coefficient of Variation is a unitless relative metric.
- $CV%$ enables direct, standardized comparison of analytical precision across different methodologies, different instruments, different laboratories, and across widely disparate concentration ranges (e.g., comparing the precision of serum glucose at 75 mg/dL versus 350 mg/dL, or comparing automated chemistry imprecision against manual methods).
+-----------------------------------------------------------------------------------------+
| Core Statistical Metrics in Clinical Chemistry |
+-----------------------------------------------------------------------------------------+
| Metric Mathematical Formula Clinical Interpretation|
| ───────────────────────────────────────────────────────────────────────────────────── |
| Mean (x̄) x̄ = (Σ xi) / n Central tendency; |
| monitors assay accuracy|
| |
| Standard Deviation (s) s = sqrt[ Σ(xi - x̄)² / (n - 1) ] Dispersion / scatter; |
| monitors imprecision |
| |
| Coefficient of CV% = (s / x̄) * 100% Unitless relative |
| Variation (CV%) imprecision; compares |
| methods & levels |
+-----------------------------------------------------------------------------------------+
The Gaussian (Normal) Distribution & Control Limits
When a stable analytical system performs repetitive measurements on a homogeneous control material, the observed data points assume a symmetrical, bell-shaped Gaussian distribution. The mathematical properties of this distribution govern quality control limit establishment:
- $\bar{x} \pm 1s$: Encompasses 68.27% of all measurements.
- $\bar{x} \pm 2s$: Encompasses 95.45% of all measurements (approximately 19 out of 20 runs).
- $\bar{x} \pm 3s$: Encompasses 99.73% of all measurements (approximately 369 out of 370 runs).
+-----------------------------------------------------------------------------------------+
| The Gaussian Distribution & Quality Control Confidence Limits |
+-----------------------------------------------------------------------------------------+
| |
| Relative Frequency |
| ^ |
| | | |
| | / | \ |
| | / | \ |
| | / | \ |
| | / | \ |
| | / | \ |
| | / | \ |
| | / | \ |
| | / | \ |
| +---+------------+------------+------+------+------------+------------+---> Value |
| -3s -2s -1s Mean(x̄) +1s +2s +3s |
| | | | | | | | |
| | <─────── 68.27% ────> | | |
| <──────────────────── 95.45% ────────────────────────> |
| <──────────────────── 99.73% ─────────────────────────────────────> |
| |
| Inherent Statistical Outliers: |
| - Outside ±2s: 4.55% of all acceptable runs (1 in 20) occur naturally by chance! |
| - Outside ±3s: 0.27% of all acceptable runs (1 in 370) occur naturally by chance. |
+-----------------------------------------------------------------------------------------+
Establishing Control Limits: Protocol & Specimen Selection
- Material Characteristics: Control materials must be stable, matrix-matched (e.g., human serum base for serum assays to mimic viscosity and protein binding), available in large batches spanning at least one year of testing, and distributed across at least two clinically relevant concentration levels (e.g., normal and pathological/abnormal).
- Data Collection: To establish definitive control limits, the laboratory must analyze the new lot over a minimum of 20 operating days, running controls across multiple shifts, calibrations, reagent vials, and operators. This captures natural between-run and day-to-day analytical variability.
- Calculation: The cumulative mean and standard deviation are calculated after excluding documented statistical outliers. The warning limit is set at $\bar{x} \pm 2s$ and the action/rejection limit is standardly set at $\bar{x} \pm 3s$.
Levey-Jennings (L-J) Control Charts
A Levey-Jennings chart is a visual time-series graph on which daily control results are plotted chronologically. The horizontal axis (x-axis) represents the date, run number, or time. The vertical axis (y-axis) displays the target mean (center line) bounded by horizontal reference lines drawn at $\pm 1s$, $\pm 2s$, and $\pm 3s$.
+-----------------------------------------------------------------------------------------+
| Anatomy of a Levey-Jennings (L-J) Chart |
+-----------------------------------------------------------------------------------------+
| |
| Measured |
| Concentration |
| ^ |
| +3s | - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Action Limit |
| | * (1_3s violation) |
| +2s | ----------------------------------------------------------- Warning Limit |
| | * * |
| +1s | · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · |
| | * * * |
| Mean | ════════════════════════════════════════════════════════════ Target Mean (x̄) |
| | * * |
| -1s | · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · |
| | * |
| -2s | ----------------------------------------------------------- Warning Limit |
| | |
| -3s | - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Action Limit |
| +----+----+----+----+----+----+----+----+----+----+----+----+---> Day / Run |
| 1 2 3 4 5 6 7 8 9 10 11 12 |
+-----------------------------------------------------------------------------------------+
Systematic Error vs. Random Error
Analytical errors in the clinical laboratory fall into two distinct categories: systematic error and random error. Recognizing the difference on Levey-Jennings charts is essential for identifying the root cause and applying proper corrective action.
+-----------------------------------------------------------------------------------------+
| Systematic Error vs. Random Error Comparison |
+-----------------------------------------------------------------------------------------+
| Parameter Systematic Error (Bias) Random Error (Imprecision) |
| ───────────────────────────────────────────────────────────────────────────────────── |
| Primary Effect Loss of Accuracy (Trueness) Loss of Precision |
| Directionality Unidirectional; predictable Bidirectional; unpredictable |
| L-J Chart Pattern Shift or Trend Widened scatter; outlier spike |
| Common Causes • Calibration drift or error • Air bubbles in sample/reagent|
| • New reagent lot variation • Electrical voltage surge |
| • Failing photometer lamp • Pipetting imprecision |
| • Dirty optical cuvette or filter • Particulate matter or clot |
| • Deteriorating calibrator • Inadequate reagent mixing |
| • Incubator temperature drift • Operator technique variation |
| Westgard Detectors 2_2s, 4_1s, 10_x 1_3s, R_4s |
+-----------------------------------------------------------------------------------------+
Systematic Error Patterns: Shifts vs. Trends
- Shift (Abrupt Step Change):
- Definition: A sudden, permanent transition of control values from one level to another, where data points cluster persistently on one side of the target mean.
- Typical Root Causes: Sudden change in reagent lot without proper calibration; incorrect calibrator set point entered into analyzer software; new calibration using an improperly reconstituted or expired calibrator; sudden replacement of a photometer lamp; sudden pipette seal failure; dirty sample probe or optical flow cell.
+-----------------------------------------------------------------------------------------+
| L-J Pattern: Systematic Shift |
+-----------------------------------------------------------------------------------------+
| +2s | ---------------------------------------------------------------- |
| | * * * * * <--- Abrupt step |
| +1s | · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · clustering above |
| | the mean |
| Mean | ════════════════════════════════════════════════════════════ |
| | * * * |
| -1s | · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · |
| | * * |
| -2s | ---------------------------------------------------------------- |
| +----+----+----+----+----+----+----+----+----+----+----+-----> Day |
| 1 2 3 4 5 6 7 8 9 10 11 |
| ^ |
| Reagent lot changed here! |
+-----------------------------------------------------------------------------------------+
- Trend (Gradual Drift):
- Definition: A continuous, progressive displacement of control values in one direction away from the target mean over $\ge 6$ consecutive operating days.
- Typical Root Causes: Gradual photometer lamp aging and loss of radiant intensity; progressive degradation or evaporation of onboard reagents; gradual accumulation of protein residue on optical cuvettes or sample probes; gradual deterioration of calibration standards stored improperly; slow filter degradation or tubing wear in peristaltic pumps.
+-----------------------------------------------------------------------------------------+
| L-J Pattern: Systematic Trend |
+-----------------------------------------------------------------------------------------+
| +2s | ---------------------------------------------------------*------ |
| | * |
| +1s | · · · · · · · · · · · · · · · · · · · · · · · · ·*· · · · · · · |
| | * |
| Mean | ═════════════════════════════════════*══════════════════════ <--- Progressive |
| | * drift upward |
| -1s | · · · · · · · · · · · · ·*· · · · · · · · · · · · · · · · · · · over 6+ days |
| | * |
| -2s | -------------*-------------------------------------------------- |
| +----+----+----+----+----+----+----+----+----+----+----+-----> Day |
| 1 2 3 4 5 6 7 8 9 10 11 |
| <--- Aging photometer lamp / progressive reagent evaporation ---> |
+-----------------------------------------------------------------------------------------+
The Westgard Multi-rule System (CLSI C24)
Historically, clinical laboratories relied on a single rule: rejecting any analytical run where a control fell outside $\pm 2s$ ($1_{2s}$ rule). However, because 4.55% of all measurements naturally exceed $\pm 2s$ in a Gaussian distribution, using $1_{2s}$ as a rejection rule causes an unacceptable false rejection rate (~5% for 1 control, ~9.75% for 2 controls). Technologists would waste time and expensive reagents recalibrating perfectly stable analyzers.
To solve this, Dr. James Westgard formulated a combination of multi-rule statistical algorithms (codified in CLSI C24) that maximize the probability of error detection ($P_{ed}$) while maintaining a very low probability of false rejection ($P_{fr} < 1%$).
+-----------------------------------------------------------------------------------------+
| CLSI C24 Westgard Multi-rule Decision Algorithm |
+-----------------------------------------------------------------------------------------+
| |
| [ Run Quality Control ] |
| │ |
| ▼ |
| Does any control exceed |
| Mean ± 2s? |
| (1_2s Rule) |
| / \ |
| NO / \ YES (WARNING ONLY!) |
| ▼ ▼ |
| [ ACCEPT ] Inspect Other Rules: |
| [ RUN ] ─────────────────────────────────────────────── |
| • 1_3s ? ──> Exceeds ± 3s? ──> REJECT (Random) |
| • 2_2s ? ──> 2 in a row > 2s? ─> REJECT (System)|
| • R_4s ? ──> Range >= 4s? ───> REJECT (Random) |
| • 4_1s ? ──> 4 in a row > 1s? ─> REJECT (System)|
| • 10_x ? ──> 10 on one side? ──> REJECT (System)|
| │ |
| ALL NO ▼ ANY YES |
| │ [ REJECT RUN ] |
| ▼ [ HOLD PATIENT RESULTS ] |
| [ ACCEPT RUN] [ EXECUTE TROUBLESHOOTING ] |
+-----------------------------------------------------------------------------------------+
Comprehensive Westgard Multi-rules Breakdown
-
$1_{2s}$ Rule (Warning Rule Only):
- Definition: One control observation exceeds either the $+2s$ or $-2s$ limit.
- Action: WARNING ONLY. It triggers the application and inspection of the remaining rejection rules.
- CRITICAL ASCP RULE: NEVER reject an analytical run based on a $1_{2s}$ violation alone! In the absence of other rule violations, the run is accepted and patient testing proceeds.
-
$1_{3s}$ Rule (Rejection Rule - Random Error):
- Definition: One control observation exceeds either the $+3s$ or $-3s$ limit.
- Error Detected: Primarily random error, though a massive systematic shift can also trigger it.
- Action: Reject the analytical run. The probability of this occurring by chance in a stable system is only 0.27% (1 in 370 runs).
-
$2_{2s}$ Rule (Rejection Rule - Systematic Error):
- Definition: Two consecutive control observations exceed the same $+2s$ limit OR two consecutive observations exceed the same $-2s$ limit.
- Application: Can be violated in two ways:
- Across runs: The same control level exceeds $+2s$ in two consecutive analytical runs.
- Within run: Both Level 1 and Level 2 controls exceed $+2s$ (or both exceed $-2s$) within the same analytical run.
- Error Detected: Systematic error (bias).
- Action: Reject the analytical run.
-
$R_{4s}$ Rule (Rejection Rule - Random Error):
- Definition: One control observation exceeds the $+2s$ limit and another control observation exceeds the $-2s$ limit within the same analytical run. The total span (range) between the two controls is $\ge 4s$.
- Application: Applied strictly within a single run across different control levels (e.g., Level 1 is at $+2.2s$ and Level 2 is at $-2.1s$; range $= 4.3s$). It is not applied across consecutive runs of the same level.
- Error Detected: Random error (extreme dispersion across levels).
- Action: Reject the analytical run.
-
$4_{1s}$ Rule (Rejection Rule - Systematic Error):
- Definition: Four consecutive control observations exceed the same $+1s$ limit OR four consecutive observations exceed the same $-1s$ limit.
- Application: Can occur across four consecutive runs for one control level, or across two consecutive runs for two control levels.
- Error Detected: Systematic error (early shift or trend).
- Action: Reject the analytical run.
-
$10_x$ Rule (Rejection Rule - Systematic Error):
- Definition: Ten consecutive control observations fall on the same side of the mean, regardless of their magnitude (even if all are within $\pm 1s$).
- Application: Can occur across 10 consecutive runs for one control level, or across 5 consecutive runs for two control levels ($5 \times 2 = 10$).
- Error Detected: Systematic error (persistent analytical bias or small calibration shift).
- Action: Reject the analytical run.
+-----------------------------------------------------------------------------------------+
| Summary Table of Westgard Multi-rules |
+-----------------------------------------------------------------------------------------+
| Rule Name Full Definition Error Detected Action / Disposition |
| ───────────────────────────────────────────────────────────────────────────────────── |
| 1_2s 1 control exceeds ±2s Warning Flag ACCEPT run; screen |
| remaining rules |
| |
| 1_3s 1 control exceeds ±3s Random Error REJECT run; halt |
| (or large systematic)patient testing |
| |
| 2_2s 2 consecutive controls > +2s Systematic Error REJECT run; halt |
| OR 2 consecutive < -2s (Bias) patient testing |
| |
| R_4s 1 control > +2s AND Random Error REJECT run; halt |
| 1 control < -2s (range >= 4s) patient testing |
| |
| 4_1s 4 consecutive controls > +1s Systematic Error REJECT run; halt |
| OR 4 consecutive < -1s (Shift / Trend) patient testing |
| |
| 10_x 10 consecutive controls on Systematic Error REJECT run; halt |
| same side of mean (any size) (Shift / Bias) patient testing |
+-----------------------------------------------------------------------------------------+
Systematic Troubleshooting of Out-of-Control QC
When a Westgard rejection rule is violated, the laboratory must follow an established, standardized troubleshooting sequence. Testing patient samples while an assay is out of control is a direct violation of CLIA regulations.
+-----------------------------------------------------------------------------------------+
| Protocolized QC Out-of-Control Remediation Workflow |
+-----------------------------------------------------------------------------------------+
| |
| STEP 1: HALT TESTING ──> Immediately stop patient testing on the affected channel. |
| Do not release or validate pending patient results! |
| │ |
| ▼ |
| STEP 2: IDENTIFY ERROR ─> Inspect L-J chart to classify failure as RANDOM (1_3s, R_4s) |
| or SYSTEMATIC (2_2s, 4_1s, 10_x). |
| │ |
| ▼ |
| STEP 3: PHYSICAL CHECK ─> Inspect reagent volumes, lot numbers, expiration dates, |
| onboard bubbles, sample probe, wash fluid, lamp hours. |
| │ |
| ▼ |
| STEP 4: TEST FRESH QC ──> Thaw/reconstitute a fresh aliquot of control material to |
| rule out control degradation, evaporation, or handling error. |
| │ |
| ▼ |
| STEP 5: RECALIBRATE ───> If fresh QC still fails, prepare fresh calibrators and |
| perform a full channel calibration. |
| │ |
| ▼ |
| STEP 6: RE-RUN QC ───> Test controls post-calibration. If in control (within ±2s), |
| verify patient samples tested since last valid QC. |
| │ |
| ▼ |
| STEP 7: DOCUMENTATION ──> Record problem, root cause, corrective action, and |
| supervisory review in corrective action log. |
+-----------------------------------------------------------------------------------------+
Critical Practical Rules in QC Troubleshooting
- Never Repeat QC Mindlessly: Simply repeating the same out-of-control vial until it passes by statistical chance is known as "hunting for a pass" and is strictly prohibited by CLIA and CAP inspection checklists.
- Verify Previous Patient Results: If a systematic error rule ($2_{2s}, 4_{1s}, 10_x$) is violated, all patient specimens analyzed since the last acceptable QC run must be identified and evaluated. If the analytical shift exceeds total allowable error ($TE_a$), previous patient samples must be retested and corrected reports issued if clinical interpretation is altered.
- Mandatory Documentation: Every out-of-control event requires complete documentation: instrument identifier, date and time, analyte channel, rule violated, root cause identified, corrective actions taken, post-correction control data, and technologist and supervisor signatures.
Test Your Knowledge
A clinical chemistry technologist runs two levels of quality control for serum potassium on an automated analyzer. Level 1 yields a result of 4.1 mmol/L (mean = 4.0 mmol/L, SD = 0.1 mmol/L). Level 2 yields a result of 6.3 mmol/L (mean = 5.8 mmol/L, SD = 0.2 mmol/L). All prior control runs over the past week were within ±1 SD of their respective means. How should the technologist interpret this run under Westgard multi-rule criteria?
A
B
C
D
Test Your Knowledge
An automated chemistry analyzer displays a quality control failure for serum total calcium. Review of the Levey-Jennings chart reveals that for the past seven consecutive days, Level 1 control values have steadily dropped: Day 1 (-0.2s), Day 2 (-0.6s), Day 3 (-1.1s), Day 4 (-1.4s), Day 5 (-1.7s), Day 6 (-2.1s), and Day 7 (-2.4s). What type of analytical error is occurring, and what is its most probable cause?
A
B
C
D
Test Your Knowledge
During a morning analytical run on a multi-channel chemistry analyzer, the Level 1 control for glucose is at +2.2 SD, and the Level 2 control in the same run is at -2.1 SD. No previous rule violations were observed. What Westgard rule has been violated, what error type does it indicate, and what is the proper laboratory action?
A
B
C
D