Interpreting Levey-Jennings Chart Patterns: Trends, Shifts, and Drifts — What They Mean and When to Take Action

Category: Basic QC Concepts | Estimated reading time: 9 minutes

Two Most Frequently Asked Questions

After learning to create Levey-Jennings charts, almost all medical laboratory technologists (MLTs) encounter these two situations:

1. "Why are my QC points always above the mean?"
2. "My chart keeps going up. Is that normal or not?"

Not a single point falls outside ±2SD, there is no 1₃s violation, but the chart "feels wrong." That feeling is often correct. Charts that contain systematic patterns are signals that the process mean is changing. This article discusses how to read those patterns objectively.

Three Terms That Are Often Confused

Patterns on an LJ chart generally indicate systematic error, which is a change in the mean of control results (this concept is discussed in the article Two Types of Error). The shape depends on the speed of the change:

• Shift is a sudden change. The points "jump" to a new level and stay there. This pattern usually has a clear starting point.
• Trend is a gradual change. The points move progressively up or down from day to day.
• Drift in daily conversation is often used for slow shifts, so it is almost synonymous with trend. Classic QC literature distinguishes between trend and shift, while "drift" appears more often as a practical term. In this article, both are treated as gradual shifts.

This distinction is important because the direction of the investigation is different: for a shift, the cause is sought in an event, while for a trend, it is sought in a slowly deteriorating process.

Question 1: "Why Are My Points Always Above the Mean?"

The Logic

If the process is truly centered on the mean, each point has an equal chance of being above or below the mean. The probability of N consecutive points being on one specific side is (0.5)ᴺ:

• 7 points: about 0.78% per side (about 1.6% for either side)
• 9 points: about 0.20% per side (about 0.4% for either side)
• 10 points: about 0.10% per side (about 0.2% for either side)

These numbers represent the probability for a specific sequence of points, assuming the points are independent and the process is centered. Because rules are checked every time a new point is added, false alarms accumulate in the long run.

How Many Points Are Considered Enough?

The answer depends on the guidelines used:

• Westgard 10x uses 10 consecutive points on one side of the mean. There are 8x, 9x, and 12x variants. Bio-Rad material warns that the most sensitive variants (7x, 8x, 9x) should be used very selectively.
• Nelson Rule 2 uses 9 points on one side.
• Western Electric Rules (predecessor to Nelson rules) use 8 points.

All these guidelines are valid. What is dangerous is one laboratory using different numbers without a written agreement. Determine one rule, write it in the SOP, then apply it consistently.

If the Points Are Indeed Always on One Side, What Are the Possible Causes?

Here are initial hypotheses that need to be verified, not conclusions:

• The mean being used is not representative. This is a frequently overlooked cause. The mean could be from the kit insert, from the previous lot of controls, or from preliminary testing performed right after service or calibration (see the article Preliminary Testing and New Control Materials). As a result, the process is actually stable, but the reference line is wrong.
• Shift due to an event, such as a change in reagent or calibrator lot, recalibration, or maintenance.
• Changes in the control material itself, for example, a vial left open too long so the value shifts.
• Systematic error that develops slowly and only becomes visible after several days.

One quick question that helps: does the first "off" point coincide with an event in the lab record? If so, it is likely a shift with a traceable cause.

Question 2: "My Chart Keeps Going Up, Is That Normal?"

Short Answer

It is not automatically abnormal, but it also should not be ignored. Even fully random data produces short upward or downward sequences of 3 or 4 consecutive points. What should be suspected are long and consistent sequences.

Rules for Trends

• Westgard 7T: 7 consecutive points moving in the same direction, all progressively increasing or all progressively decreasing. This rule applies across runs and is available as an additional option in many QC software packages.
• Nelson Rule 3: 6 consecutive points that are constantly increasing or decreasing.

The difference of one point is significant. With a simple calculation (assuming independent, continuous data without duplicate values), the probability of 6 consecutive points forming a pure upward or downward sequence by pure chance is about 2/720, or 0.28% (roughly 1 in 360). For 7 points, the probability is about 2/5,040, or 0.04% (roughly 1 in 2,500). This means 7T is stricter: its false alarms are much rarer, but it requires one additional point before triggering. Because real QC data is often rounded so that identical values occur, the SOP should state how identical values are treated.

Trends Can Occur Within Limits

Trends do not have to cross ±2SD or ±3SD to be significant. A chart can rise continuously within the ±1SD zone and still indicate something is changing, exactly like the "signal before official violation" scenario in the Levey-Jennings Chart article.

If you often encounter such subtle patterns, visual rules have their limits. Methods designed for small and gradual shifts, such as CUSUM and EWMA, are much more sensitive (see their respective articles).

Two "Too Good" Patterns That Should Also Be Suspected

Suspicious patterns are not always deviations. There are patterns that actually look too neat:

• Points sticking around the mean. Nelson Rule 7 marks 15 consecutive points that are all within ±1SD as a signal that the apparent variation is smaller than it should be. If ±1SD truly represents about 68% of the data, the probability of 15 consecutive points all being within it is only about 0.3%. The cause to investigate: the SD being used is too large (e.g., taken from the manufacturer's range, not from preliminary testing), so the control limits are too loose. This aligns with the warning in the article "Always Green" QC.
• Points alternating up and down regularly (Nelson Rule 4: 14 consecutive points). This pattern suggests two alternating sources of variation, such as two operators, two shifts, or two reagent bottles used alternately.

Practical Steps When Finding a Pattern

1. Calculate objectively. How many consecutive points, on which side, increasing or decreasing? Match it with the rules written in your lab's SOP, not your feelings.
2. Find the starting point of the pattern, then match it with event records: new reagent or calibrator lot, calibration, maintenance, new control vial, operator change, room temperature change.
3. Distinguish between shift and trend. A shift points to a specific event. A trend points to something slowly deteriorating: reagents or calibrators nearing expiration, aging instrument components, unstable storage temperatures, control materials left open for too long.
4. If a rejection rule is met (e.g., 10x), follow the workflow in the QC Reject article: hold patient results, investigate, correct, re-verify QC, document.
5. If the rule is not yet met but the pattern is suspicious, do not ignore it and do not immediately reject. Increase vigilance, observe whether the pattern continues, record your observations, and consider more sensitive methods (CUSUM or EWMA).
6. Document what was seen, what was checked, and the decision. This is tangible evidence of interpretive competence during an audit.

A matter of balance: the more and more sensitive the rules used, the faster patterns are detected, but the more often alarms sound without any real problems. Therefore, sensitive rules such as 7x or 8x should be used sparingly, and the final choice should consider the method's performance (see the articles Sigma Metrics and Monte Carlo Simulation).

Summary

• Points always on one side of the mean: first check if the mean itself is representative, then look for a shift or change in control material.
• Chart keeps going up: count the sequence. 3–4 points are normal, 6–7 consecutive points already meet trend rules (Nelson 6, Westgard 7T).
• There is no single "correct" number for all labs. The important thing is one rule, written, and consistent.
• Trends that are still within limits are still worth investigating, and CUSUM/EWMA are more sensitive for this case.
• "Too neat" patterns (sticking to the mean) are just as important as patterns that deviate.

Closing

The Levey-Jennings chart is not just a tool to answer "reject or not," but also "what is happening to my process." MLTs who are used to reading its patterns catch problems early, before formal rules sound and before patient results are affected. The difference lies in the habit of looking at the sequence of points, not just the last point.