Clinical Quality Improvement Project Pack
Three documents and three sheets that freeze the baseline median before any change is tested, then check every new point against that same frozen line.
Free download · No account needed
Run Chart Data, screening view
Six straight points below the frozen median
Illustrative weeks for a fictional four-provider primary care practice, Wrenfield Family Health, testing a two-touch reminder workflow against its weekly no-show rate.
| Week | No-Shows / 240 | Rate | vs. Frozen Median (13.75%) | Run |
|---|---|---|---|---|
| 16 | 27 | 11.25% | Below | 1 |
| 17 | 25 | 10.42% | Below | 2 |
| 18 | 24 | 10.00% | Below | 3 |
| 19 | 22 | 9.17% | Below | 4 |
| 20 | 23 | 9.58% | Below | 5 |
| 21 | 21 | 8.75% | Below | 6 (shift confirmed) |
Six consecutive weeks below a median a 12-week baseline froze before this reminder workflow started is a shift, per the run chart's own rule. Weeks 13 through 15 do not appear here: the second of those three came back above the median, so cycle one alone had not yet cleared the threshold.
Full 24-week chart, the other three rules, and the arithmetic: below.
Most quality improvement templates hand a team a grid to plot data on and a PDSA worksheet, then leave the actual proof to a glance at whether the line looks better afterward. Per a peer-reviewed primer on interpreting QI data, it is "crucial to have baseline data before initiating an improvement project," since without it there is no way to confirm a problem existed, let alone that anything afterward improved it. An eyeballed before-and-after average skips that requirement and calls the difference an improvement regardless.
This pack freezes that baseline first: at least 12 consecutive points, collected before any change, with their median locked as the line every later point is checked against. Per Robert Lloyd's chapter on run charts, a shift of 6 or more consecutive points on one side of that line, or a trend of 5 or more moving the same direction, is what counts as a real signal. Wrenfield Family Health, a fictional four-provider practice, froze its weekly no-show-rate baseline at 33 no-shows, 13.75 percent of 240 scheduled appointments, before testing anything.
Cycle one (a call alone) did not hold: two weeks fell below 33 no-shows, then a third came back above it, so no rule fired. Cycle two added a text reminder, and weeks 16 through 21 ran six straight below the frozen median, confirming a shift worth 494 recovered visit-slots a year. A plain before-and-after average across the same 24 weeks would show a similar-looking drop, from 13.68 to 10.28 percent, with no way to tell whether that gap is the reminder working or the ordinary variation the practice already had.
What's in the pack
Project Charter
The aim statement, the measure's operational definition precise enough that two different counters agree on the same number, and the baseline collection plan, filled in before any intervention gets named.
Baseline and Follow-up Measurement
Every measurement period's raw count and rate in one place, split cleanly into the frozen baseline window and everything measured afterward, so the median governing every later rule check is always traceable back to the exact points it came from. The same respect for a measure's exact definition governs which six measures a group reports under a MIPS reporting checklist.
PDSA Cycle Log
One row per cycle: what was tested, at what scale, the prediction made before results came in, and what actually happened. Each row closes with the adapt, adopt or abandon decision, so a later reader can see exactly which test was running when a rule fired.
Run Chart Data
Every point checked against the frozen median with the running consecutive-point count and whichever of the four rules fired, if any, so the chart records a specific, falsifiable signal instead of a line a reader has to interpret alone.
Intervention Description
What changed, on what scale, who owned it, and the prediction made in advance, recorded per cycle rather than once for the whole project. A second cycle that adds a new element is a different test from the one before it, not an edit to it.
Results Report
Names the rule that fired, the point it fired on, and the PDSA cycle running at the time. States the size of the change in the aim statement's own units, and reports plainly when no rule fired within the planned time frame rather than calling a trend a result. A confirmed change's own median becomes the next frozen baseline, the same version discipline a clinical registry's specification applies to its own required fields.
How to use it
- 1
Open the pack, or download it
Open the pack in River and describe the process you want to improve, or download the blank Word and CSV files and run the project on your own.
- 2
Send your baseline, or start collecting it
If fewer than 12 consecutive pre-intervention points exist, the agent sets the measurement frequency and says plainly how many periods away a usable baseline is. If they exist, it computes and freezes the median immediately.
- 3
Log each PDSA cycle as you test it
After every new measurement period, the agent checks the point against the frozen median and reports the running consecutive-point count and whether a rule just fired, rather than leaving that read to a glance at the chart.
- 4
Report the result once a rule fires or the timeline ends
Either outcome writes the Results Report. A confirmed shift's own median becomes the frozen baseline for whatever the team targets next.
Frequently asked questions
Is this template free?
Yes. Download the three documents and three sheets as Word and CSV files with no signup and no card. "Edit with AI" is a separate, optional path for teams that want the agent to run the project with them. The rest of the library is at the template library.
What format are the downloaded files?
Word documents (.docx) for the Project Charter, the Intervention Description and the Results Report, and CSV (.csv) for the three sheets, zipped into one download. They open natively in Word, Pages, Google Docs, Excel, Numbers and Sheets.
Why freeze the baseline median instead of recalculating it as new data comes in?
Because a moving median chases the data it is supposed to judge. If it recalculates every time a point is added, a genuinely working intervention pulls the average down with it, and the line a later point gets compared against keeps sliding toward whatever that point is about to show. Freezing it first is what lets six consecutive points below the line mean something specific.
Does a confirmed shift mean a payer contract improves too?
No, and it answers a different question. This pack confirms whether the underlying rate itself changed in a statistically defensible way. Whether that change moves a specific payer's reconciliation payment depends on that contract's own measure definitions and thresholds, which is what a value-based contract performance review reads once a statement actually arrives.
What happens if no rule ever fires?
The Results Report says so directly instead of describing an encouraging-looking trend as a win. The team can extend data collection, test a different change, or close the project without a confirmed result. All three are legitimate outcomes; the only outcome this pack will not produce is a report that calls a slope a signal.
Does this pack help pick which measure to improve?
No, it assumes the measure is already chosen and turns it into a baseline, a test, and a rule-based read on the result. Picking which measure is worth a project's time, such as a gap a care gap and population health report already flags, happens before this pack's first document gets filled in.
Freeze the baseline before you test the change
Download the blank pack as Word and CSV files, or open this exact pack in River and let the agent freeze your baseline median and check every new point against it.
Edit with AI