Building and reading a run chart
A before and after average can tell you that two periods differ. It cannot show whether a change began before your intervention, lasted only one week, or continued over time. A run chart keeps that sequence visible.
A run chart plots a measure in time order and usually adds a median reference line. IHI’s introductory guidance recommends calculating the median when you have at least 10 data points, labeling the axes, and annotating changes. [1] That is a practical starting point, not a guarantee that every short series can support a reliable conclusion.
Build a comparable series
Choose the measure and the time interval before plotting. Keep the outcome definition, eligible population, and data collection method consistent. Use consecutive periods and show missing periods rather than compressing the timeline.
Hypothetical example. A clinic tracks the percentage of new referrals whose receipt is confirmed within its locally chosen five-workday target. Each week represents referrals submitted that week, assessed after the full five-workday window. The target is fictional, not a clinical standard.
The 12 baseline percentages are 45, 55, 50, 40, 60, 45, 55, 50, 40, 55, 45, and 50. Their median is 50%. After a workflow change, six weekly values are 65, 70, 60, 75, 70, and 80.
Keep the baseline median at 50% while examining the new observations. Mark when the workflow began. Recalculating a median every time a new result arrives can move the reference you are trying to compare against. Any later phase median should have a stated rationale and clearly identified data period.
Weekly referrals confirmed within five workdays
Show the data
| Week | Percent confirmed | Signal |
|---|---|---|
| Week 1 | 45 | |
| Week 2 | 55 | |
| Week 3 | 50 | |
| Week 4 | 40 | |
| Week 5 | 60 | |
| Week 6 | 45 | |
| Week 7 | 55 | |
| Week 8 | 50 | |
| Week 9 | 40 | |
| Week 10 | 55 | |
| Week 11 | 45 | |
| Week 12 | 50 | |
| Week 13 | 65 | Shift |
| Week 14 | 70 | Shift |
| Week 15 | 60 | Shift |
| Week 16 | 75 | Shift |
| Week 17 | 70 | Shift |
| Week 18 | 80 | Shift |
Use one stated rule set
The following four rules follow the introductory NHS run chart guide. Do not combine them with different control chart thresholds. [2]
| Pattern | What to look for |
|---|---|
| Shift | Six or more consecutive useful points on the same side of the median. Points on the median do not count toward or break the shift. |
| Trend | Five or more consecutive points that increase or decrease. Ignore repeated equal values when counting the trend. |
| Too few or too many runs | An unusually small or large number of sequences on either side of the median, judged using a runs table for the number of useful points. |
| Astronomical point | A point that is conspicuously different from the rest and warrants investigation, not simply the highest or lowest value. |
These counts are for charts with fewer than 20 data points. With 20 or more points, use eight or more points for a shift and six or more for a trend. [3]
For the runs rule, exclude points exactly on the median when counting useful points. A run is a sequence on one side of the median. There is no single acceptable number of runs for every chart length. [2]
Read the example without overclaiming
The six postchange values are all above 50%, so the series meets the stated shift rule. It does not show a continuous upward trend: 70% is followed by 60%, and 75% is followed by 70%. A shift and a trend are different patterns.
The pattern supports investigating a sustained change in the process. It does not establish that the workflow alone caused it. Staffing, case mix, and measurement changes might also matter. In this fictional series, each weekly denominator is 20; a real project should retain and review its actual denominators.
A second fictional series starts from the same baseline but has postchange values of 55, 45, 60, 50, 45, and 55. It has no six-point shift or five-point trend. Do not label it an improvement merely because its last point is higher than its first.
Comparison series with the same baseline
Show the data
| Week | Percent confirmed | Signal |
|---|---|---|
| Week 1 | 45 | |
| Week 2 | 55 | |
| Week 3 | 50 | |
| Week 4 | 40 | |
| Week 5 | 60 | |
| Week 6 | 45 | |
| Week 7 | 55 | |
| Week 8 | 50 | |
| Week 9 | 40 | |
| Week 10 | 55 | |
| Week 11 | 45 | |
| Week 12 | 50 | |
| Week 13 | 55 | |
| Week 14 | 45 | |
| Week 15 | 60 | |
| Week 16 | 50 | |
| Week 17 | 45 | |
| Week 18 | 55 |
Your next decision
Decide the interval, denominator, baseline period, and rule set before interpreting your chart. A zero is a result; a missing value is not. Keep a short event log alongside the series and review the outcome with process and balancing measures from Outcome, process, and balancing measures.
Sources
[1] Institute for Healthcare Improvement. (2017). Quality Improvement Essentials Toolkit: Run chart.
[3] Provost, L. P., & Murray, S. K. (2011). The health care data guide: Learning from data for improvement. Jossey-Bass.