2.1 Statistical Quality Control, Levey-Jennings Charts & Westgard Multirules

Statistical Quality Control, Levey-Jennings Charts & Westgard Multirules

Foundational Clinical Laboratory Statistics

In the modern clinical laboratory, statistical quality control (SQC) forms the objective backbone of analytical reliability. Quality control specimens—materials with established target values and known analytical matrices—are assayed alongside patient samples to verify that the analytical measurement system remains stable and error-free. Every quantitative laboratory decision rests on standard Gaussian (normal) distribution theory, which describes the predictable dispersion of repeated measurements around a central tendency.

Mean, Standard Deviation, and Gaussian Distribution

The mean (x̄) represents the arithmetic average of a dataset of control measurements, calculated as the sum of all individual observations divided by the total number of observations (n). In clinical chemistry and hematology, an initial target mean is established by analyzing a minimum of 20 replicates of the control material across at least 20 separate analytical runs over 20 distinct operating days, capturing intermediate day-to-day instrument variability.

The standard deviation (s or SD) quantifies the dispersion or spread of values around the mean, reflecting the inherent analytical imprecision of the test system:

s = √[ Σ(x_i - x̄)² / (n - 1) ]

The denominator utilizes n - 1 degrees of freedom to provide an unbiased estimate of population variance. Under standard Gaussian distribution rules:

• 68.27% of all control measurements fall within ±1s of the target mean.
• 95.45% of all measurements fall within ±2s of the target mean.
• 99.73% of all measurements fall within ±3s of the target mean.

Because 95.45% of valid results naturally fall within ±2s, approximately 4.55% (roughly 1 in 20) of fully acceptable control measurements will fall between 2s and 3s purely due to random Gaussian variation. This fundamental statistical reality explains why a single result between 2s and 3s cannot serve as an automatic run rejection criterion without generating excessive false alarms.

Coefficient of Variation (CV)

Because the standard deviation is expressed in the identical units of measurement as the analyte (e.g., mg/dL, mmol/L, or U/L), it cannot be used directly to compare analytical imprecision across different concentration levels or diverse analytical methodologies. The Coefficient of Variation (CV) normalizes the standard deviation against the mean as a percentage:

CV (%) = (s / x̄) × 100

For example, if a high-level glucose control yields a mean of 300 mg/dL with an SD of 6.0 mg/dL, its CV is (6.0 / 300) × 100 = 2.0%. If a low-level glucose control yields a mean of 60 mg/dL with an SD of 2.4 mg/dL, its CV is (2.4 / 60) × 100 = 4.0%. Despite having a smaller absolute SD (2.4 vs. 6.0), the low-level assay exhibits twice the relative imprecision. Laboratory managers use CV to set method performance specifications, monitor analyzer consistency over time, and compare competitive diagnostic platforms.

Standard Deviation Index (SDI / Z-Score)

To evaluate inter-laboratory accuracy and peer-group consensus in external quality assessment (EQA) and proficiency testing (PT), laboratories calculate the Standard Deviation Index (SDI), also known as the Z-score:

SDI = (x̄_lab - x̄_peer) / s_peer

where x̄_lab is the laboratory's reported result or control mean, x̄_peer is the peer group consensus mean, and s_peer is the peer group standard deviation. The clinical laboratory interprets SDI values as follows:

• |SDI| ≤ 1.0: Excellent performance; the method aligns closely with consensus.
• 1.0 < |SDI| ≤ 1.5: Acceptable performance; minimal systematic bias.
• 1.5 < |SDI| ≤ 2.0: Borderline performance; warrants proactive review.
• |SDI| > 2.0: Unacceptable bias; requires immediate investigation and corrective action before proficiency testing failure occurs.
• |SDI| > 3.0: Critical systematic error; indicates severe calibration or analytical failure.

Total Allowable Error (TEa) and Biological Variation

In diagnostic testing, clinical utility requires that analytical error remain small enough not to alter clinical interpretation or patient management. Total Allowable Error (TEa) defines the maximum permissible error threshold combining both systematic error (bias) and random error (imprecision):

TEa ≥ |Bias| + 1.65 × s or TEa (%) ≥ |Bias (%)| + 1.65 × CV (%)

The multiplier 1.65 represents a 95% one-sided confidence interval, ensuring that no more than 5% of patient samples exceed the allowable error budget. TEa goals are derived from established hierarchical models:

1. Clinical Outcome Studies: Impact of analytical error on direct clinical decision-making.
2. Biological Variation Databases: Based on intra-individual (CV_I) and inter-individual (CV_G) biological variability (e.g., the EFLM biological variation database). Desirable analytical performance goals dictate:
   • Bias_desirable ≤ 0.375 × √(CV_I² + CV_G²)
   • CV_desirable ≤ 0.50 × CV_I
   • TEa ≤ Bias_desirable + 1.65 × CV_desirable
3. Regulatory and PT Limits: Mandated performance standards established under CLIA '88 (e.g., CLIA acceptable limit for serum potassium is target ± 0.5 mmol/L, or calcium target ± 1.0 mg/dL).

Six Sigma Quality Management in the Clinical Laboratory

Six Sigma metrics bridge laboratory analytical statistics with industrial engineering quality control. The Sigma metric quantifies the capability of an analytical process to deliver results within defined tolerance limits (TEa):

Sigma Metric (σ) = [ TEa (%) - |Bias (%)| ] / CV (%)

The resulting Sigma value dictates the complexity of quality control monitoring required:

• ≥ 6σ (World-Class Quality): The defect rate is < 3.4 defects per million opportunities. Analytical performance is exceptionally robust. The laboratory can maintain high error detection with a simple 1_3s rejection rule using N=2 controls per day, eliminating unnecessary false rejections.
• 5σ (Excellent Quality): Controlled effectively using 1_3s or 1_2.5s rules with N=2.
• 4σ (Good Quality): Requires multirule combinations (1_3s / 2_2s / R_4s / 4_1s) with N=2 to N=4 controls per run to detect clinically relevant errors.
• 3σ (Marginal / Minimum Acceptable): Requires extensive multirule SQC (1_3s / 2_2s / R_4s / 4_1s / 10_x), increased control frequency (N ≥ 4 to 6), and frequent recalibration.
• < 3σ (Unacceptable Quality): Defect rates exceed clinically acceptable standards. The method cannot be reliably controlled by statistical QC alone; it requires immediate redesign, vendor replacement, or alternative methodology.

---

Levey-Jennings Control Charts

Developed by Stanley Levey and E.R. Jennings in 1950 and adapted for clinical laboratories by J.O. Westgard, the Levey-Jennings (L-J) control chart is a visual graphical representation of analytical process performance over time.

Chart Architecture

• X-axis: Displays chronological sequence, time, or analytical run numbers.
• Y-axis: Displays the concentration of the control material, or normalized SD units (0, ±1s, ±2s, ±3s).
• Center Line: Represents the established target mean (x̄).
• Control Limit Lines: Horizontal lines drawn at ±1s (informational), ±2s (warning), and ±3s (action/rejection).

Analytical Error Signatures: Shifts vs. Trends

Evaluating the temporal pattern of plotted control points allows laboratory personnel to distinguish between two distinct forms of analytical error: systematic error and random error.

1. Shift (Abrupt Systematic Error)

A shift is defined as an abrupt, sustained displacement of control values to one side of the mean, establishing a new continuous baseline. Shifts reflect an immediate, systemic change in the measurement system. Common etiologies include:

• Transition to a new lot of reagents or calibrators with a discordant assigned value or titer.
• Sudden change in instrument calibration factor or erroneous recalibration.
• Severe contamination of system water, diluent, or wash solution.
• Replacement of critical hardware components (e.g., photometer lamp, sampling probe, flow cell) without proper realignment.
• Inaccurate reconstitution of lyophilized control materials (e.g., volumetric pipetting error).
• Sudden shift in analytical incubation temperature or optics block temperature.

2. Trend (Gradual Systematic Error)

A trend is defined as a gradual, continuous progression of control values in one consistent direction (upward or downward) across six or more consecutive analytical runs. Trends indicate progressive degradation of a system component. Common etiologies include:

• Aging and declining luminous intensity of photometer lamps, lasers, or excitation sources.
• Gradual evaporation of reagents, wash buffers, or controls left exposed on an open reagent carousel.
• Thermal deterioration of reconstituted control materials or enzyme reagents over extended on-board storage.
• Progressive build-up of protein deposits, biofilm, or particulate debris in hydraulic tubing, cuvettes, or aspiration probes.
• Gradual deterioration of chromatographic columns (HPLC) or capillary electrophoresis coatings.
• Slow sensor or electrode drift in ion-selective electrode (ISE) modules.

3. Random Error

In contrast to systematic error, random error is unpredictable and unsystematic, characterized by an isolated control value deviating widely from the mean in either direction without an underlying pattern. Random error affects precision rather than trueness. Common etiologies include:

• Transient air bubbles or micro-particulates in reagent or sample fluidics.
• Unnoticed fibrin clots or particulate matter during sample aspiration.
• Electrical voltage surges or fluctuating power supply.
• Incomplete mixing or particulate sedimentation in the control vial.
• Environmental temperature or humidity spike during a specific run.

---

The Westgard Multirule Algorithm

Single-rule QC systems (such as rejecting any run exceeding ±2s) suffer from excessive false rejection rates (~4.5% with N=1, and ~9% with N=2). To balance high probability of error detection with low probability of false rejection (P_fr < 1%), Dr. James O. Westgard developed a decision-tree algorithm combining multiple statistical rules.

Classical Westgard Multirules Defined

Contemporary Rule Variations

Modern automated laboratory information systems (LIS) and middleware incorporate extended multirule combinations tailored to specific assay sigma metrics:

• 2 of 3_2s: Two out of three consecutive controls exceed ±2s on the same side of the mean (systematic error).
• 3_1s: Three consecutive controls exceed ±1s on the same side of the mean (systematic error).
• 6_x, 8_x, 9_x, 12_x: Variations of the run-length rule where 6, 8, 9, or 12 consecutive points fall on one side of the mean, used depending on the number of control levels assayed per run (N=2 vs. N=3).

---

Troubleshooting Out-of-Control Runs: Managerial Protocols

When an analytical run violates a rejection multirule, laboratory management must enforce a rigorous, standardized corrective action protocol. The common technologist temptation to 're-run the control until it passes' represents an unacceptable regulatory violation under CLIA '88 and CAP standards. By pure chance, a flawed system will eventually generate a control value within ±2s, releasing erroneous patient results into the medical record.

Mandatory 5-Step Out-of-Control Workflow

1. Quarantine Patient Results: Immediately place a software hold on all patient results processed in the analytical run since the last acceptable QC evaluation. Under no circumstances may patient results be verified or transmitted to the electronic health record (EHR).
2. Diagnose the Error Type: Review the Levey-Jennings chart to classify the failure mode:
   • Random Error (1_3s, R_4s): Direct attention to sample aspiration issues, bubbles in reagent tubing, clotted probes, power fluctuations, or incomplete mixing of the control vial.
   • Systematic Error (2_2s, 4_1s, 10_x): Direct attention to reagent expiration, lot-to-lot differences, calibration expiration, onboard reagent degradation, or detector temperature drift.
3. Execute Root Cause Investigation:
   • Inspect the instrument status: verify operating temperatures, hydraulic pressures, flow cell cleanliness, and error logs.
   • Inspect reagents: check lot numbers, open-vial stability, expiration dates, onboard volume, and physical appearance (color change, precipitate).
   • Inspect controls: verify reconstitution date, storage temperature, number of freeze-thaw cycles, and evaporation.
4. Implement and Document Corrective Action:
   • If a root cause is established (e.g., exhausted reagent cartridge or degraded calibrator), replace the defective component, recalibrate if indicated, and assay fresh controls across all operating levels.
   • All control levels must demonstrate acceptable statistical performance before releasing the testing hold.
   • Document the complete failure, root cause, corrective intervention, and operator ID in the laboratory's electronic quality management system.
5. Evaluate Patient Specimen Impact (Look-Back Analysis):
   • For systematic error rejections, the laboratory must audit patient specimens analyzed between the last in-control run and the out-of-control run.
   • Re-assay a statistically representative subset of patient specimens (minimum 5-10 samples) across the measurement range. If clinical results show significant discordance exceeding total allowable error (TEa), re-test all affected patient specimens and issue corrected reports per laboratory policy.