PDSA Cycle and the Model for Improvement: Procedure, Template, and Worked Example

The PDSA cycle — Plan, Do, Study, Act — is a method for testing a change on a small scale, learning from what happens, and adjusting before spreading it. It is the engine inside the Model for Improvement, which wraps three questions around it: what are we trying to accomplish, how will we know that a change is an improvement, and what change can we make that will result in improvement.

The distinguishing feature of PDSA is not the four steps — almost any improvement method has an equivalent — but the deliberate smallness of each test. AHRQ, in its own PDSA guidance, is explicit about this: each cycle “should be as brief as possible for you to gain knowledge that it is working or not (some can be as short as 1 hour)”, each typically “contains only a segment or single step” of a larger implementation, and each “will likely involve only a portion of the practice (maybe 1 or 2 doctors)”. A PDSA cycle that runs for three months across an entire department is not a PDSA cycle; it is a project with a review meeting at the end.

The Model for Improvement: three questions, then the cycle

PDSA on its own tests changes. It does not tell you which change to test or how you would recognise success. That is what the three questions do, and skipping them is why so many improvement efforts consist of energetic cycles that never converge on anything.

1. What are we trying to accomplish?

The aim statement. It should be specific, numerical, and time-bound — not “improve discharge communication” but “increase the proportion of patients receiving a completed discharge medication list to 90% within six months on unit 4W”. A vague aim cannot fail, which sounds comfortable and is the problem: it also cannot tell you when to stop.

2. How will we know that a change is an improvement?

The measures. Improvement work conventionally uses three types, and using only the first is the most common measurement error:

• Outcome measures — what happens to the patient or the system. Closest to what you care about, slowest to move, and often the noisiest.
• Process measures — whether the parts of the system are performing as intended. These move first and are what tell you whether the change was actually implemented, as distinct from whether it worked.
• Balancing measures — what might be getting worse elsewhere as a consequence. Reducing length of stay while readmissions climb is not an improvement, and only a balancing measure will show it.

3. What change can we make that will result in improvement?

The change ideas. These come from process analysis, from the people doing the work, from published change packages, and from theory about why the current process produces the current result. The Model for Improvement is agnostic about where the idea comes from; PDSA is how you find out whether it holds here.

The four steps

Plan

The planning step has three components, and AHRQ’s worksheet asks for each explicitly.

What you plan to do. A concise statement of the test — “much more focused and smaller than the implementation” of the overall change. This is where most cycles fail: the plan describes the whole intervention rather than the next testable slice of it.

What you predict will happen. AHRQ’s worksheet asks “I hope this produces:” and invites either quantitative or qualitative expectations — “a certain number of doctors performed teach-back”, or “nurses noticed less congestion in the lobby”. The prediction is what makes the cycle a test rather than an activity. If you do not write down what you expect, you cannot be surprised, and being surprised is the entire mechanism by which PDSA generates knowledge.

The steps to execute. AHRQ specifies two things this must include:

• The population you are working with — “are you going to study the doctors’ behavior or the patients’ or the nurses’?”
• The time limit — and AHRQ adds a permission that matters: “you may set a time limit of 1 week but find out after 4 hours that it doesn’t work. You can terminate the cycle at that point because you got your results.”

Do

Run the test, and — this is the part that separates Do from simply implementing — observe. AHRQ’s prompt is “What did you observe?”, and the guidance is to record how patients, doctors and nurses reacted, how the change fitted with the existing system or patient flow, and to ask “Did everything go as planned?” and “Did I have to modify the plan?”

Observations during the Do step are frequently more valuable than the measured result, because they are where you learn why something did or did not work. A change that produced no measurable effect because nobody actually did it is a completely different finding from one that was implemented faithfully and had no effect — and only the observations distinguish them.

Study

Compare what happened against what you predicted. AHRQ’s prompts are “What did you learn?” and “Did you meet your measurement goal?”

The word is Study, not Check, and the change from Shewhart’s and Deming’s earlier “Check” was deliberate: checking asks whether the target was met, studying asks what the result tells you about your theory of the process. A cycle that met its target but for an unexpected reason has produced important knowledge that a pass/fail check would discard.

Act

Decide what happens next. AHRQ’s prompt: “What did you conclude from this cycle?” — “whether it worked or not. And if it did not work, what you can do differently in your next cycle to address that. If it did work, are you ready to spread it across your entire practice?”

The conventional decision set is:

• Adopt — the change worked at this scale; move to a larger test, or to implementation.
• Adapt — the change has promise but needs modification; run another cycle with the modification.
• Abandon — the change did not work and modification will not save it; test a different change idea.

Abandoning is a success of the method, not a failure of the team. The value of testing on one patient with one nurse for one hour is precisely that abandoning costs almost nothing.

A worked template

The following is the worksheet structure from AHRQ’s PDSA tool, with the fields it specifies. Reproduce it as a single page; if it does not fit on one page, the test is too big.

Field	What goes in it
Tool / change	The overall change or tool being implemented.
Step	The smaller step within that change being tested in this cycle.
Cycle	The cycle number. “As you work through a strategy for implementation, you will often go back and adjust something and want to test whether the change you made is better or not. Each time you make an adjustment and test it again, you will do another cycle.”
PLAN
I plan to…	A concise statement of this specific test — a small portion of the overall implementation.
I hope this produces…	The prediction. A measurement or an outcome, quantitative or qualitative.
Steps to execute	What you will do, who the population is (staff group or patients), and the time limit.
DO
What did you observe?	Reactions of patients and staff, fit with existing workflow, whether the plan was followed or modified.
STUDY
What did you learn? Did you meet your measurement goal?	Result against prediction, and what that says about the underlying theory.
ACT
What did you conclude from this cycle?	Adopt, adapt or abandon — and if adapting, what changes in the next cycle.

A worked example, from AHRQ’s own material

AHRQ’s tool supplies complete worked examples; the following is one cycle from the patient-feedback example, reproduced to show the level of specificity a usable PDSA record has.

Tool: Patient Feedback  |  Step: Dissemination of surveys  |  Cycle: 1st Try

Plan — I plan to: test a process of giving out satisfaction surveys and getting them filled out and back to us.
I hope this produces: at least 25 completed surveys per week during this campaign.
Steps to execute: We will display the surveys at the checkout desk. The checkout attendant will encourage the patient to fill out a survey and put it in the box next to the surveys. We will try this for 1 week.

Do — What did you observe? We noticed that patients often had other things to attend to at this time, like making an appointment or paying for services and did not feel they could take on another task at this time. The checkout area can get busy and backed up at times. The checkout attendant often remembered to ask the patient if they would like to fill out a survey.

Study — What did you learn? We only had 8 surveys returned at the end of the week. This process did not work well.

Act — What did you conclude? Patients did not want to stay to fill out the survey once their visit was over. We need to give patients a way to fill out the survey when they have time.

Two things about that record are worth copying. The prediction was numerical and was wrong by a factor of three, which is exactly the kind of result that produces learning. And the Act conclusion is a causal claim about why it failed — patients would not stay after the visit — which directly specifies the next cycle. A conclusion reading “this did not work, we will try something else” would have generated nothing.

Sequencing cycles: ramps, not repetitions

A single PDSA cycle proves almost nothing. The method works through a sequence in which each cycle increases scale or difficulty — commonly described as a ramp. A typical progression:

1. Test with one. One clinician, one patient, one shift. The objective is feasibility, not effect: does this even work once?
2. Test with a few, under favourable conditions. Willing volunteers, day shift, a straightforward patient population. Refine the mechanics.
3. Test under unfavourable conditions. This step is skipped more often than any other, and skipping it is why so many changes collapse on spread. Night shift, agency staff, the busiest day, the most complex patients, the unit that objected. If a change only works when the enthusiasts run it in daylight, it is not ready.
4. Test at scale. A whole unit, a whole clinic session.
5. Implement. This is a different activity from testing: it means changing the standard work — documentation, training, job descriptions, order sets, EHR configuration — so the change persists without the project team.
6. Spread. Move to other units, with each receiving unit typically running its own short adaptation cycles rather than adopting verbatim.

The distinction between step 4 and step 5 is where improvement projects most often fail permanently. A change that is being sustained by attention will regress the moment attention moves. Implementation means making the new way easier than the old way when nobody is watching.

Measurement in PDSA: annotated run charts, not before-and-after

PDSA measurement is time-series measurement. You plot the measure repeatedly over time and annotate the chart with the cycles, because the question is not “was the after-value higher than the before-value” but “did the pattern of variation change, and did it change when we intervened?”

A two-point before-and-after comparison cannot distinguish a real effect from ordinary variation, which is the central lesson of statistical process control and the reason improvement work uses run charts and control charts rather than a paired test on two means. Non-random signals in a run chart — a shift of consecutive points on one side of the median, a sustained trend, too few or too many runs — are what indicate that something other than chance has occurred.

Two consequences follow for practice. First, collect the measure frequently and in small samples rather than infrequently in large ones: twenty points of five observations each are far more informative for improvement than two points of fifty. Second, start collecting before the first cycle, because a run chart with no baseline cannot show a shift.

PDSA and research: a different epistemic contract

PDSA is often described as “rapid-cycle research”, which is misleading. Its logic is genuinely different from that of a designed study, and confusing the two produces both bad improvement work and bad research.

	PDSA cycle	Designed study
Question	Will this change work here?	Does this intervention work, in general?
Design changes mid-course	Expected — that is the point	Threat to validity
Sample	Deliberately tiny, then ramped	Powered in advance
Confounding	Managed by repeated testing under varied conditions	Managed by design — randomisation, control, blinding
Generalisability	Not claimed; local knowledge is the product	The primary objective
Analysis	Annotated time series; qualitative observation	Pre-specified statistical analysis

The practical question this raises — whether a given project needs IRB review — is a real determination with real consequences and is not answered by calling the work “QI”. It turns on whether the activity is designed to develop or contribute to generalisable knowledge. See the guide to quality improvement versus human subjects research and the IRB determination, which covers the criteria and the grey zone where a QI project intended for publication sits.

PDSA is also frequently confused with action research. They share a cyclical structure and participatory ethos and are historically distinct traditions: action research is a social-science methodology producing knowledge claims and academic outputs, with its own plan-act-observe-reflect cycle; PDSA is an operational improvement method whose product is a working local process.

Common failure modes

• The cycle that is really a project. A “PDSA” running for months across a department with no interim study point. AHRQ’s guidance is the corrective: single step, short duration, small sample.
• No prediction. Without a written prediction there is nothing to be surprised by, and Study degenerates into describing what happened.
• Plan, Do, Do, Do. Cycles run repeatedly without a genuine Study step. The tell is that nobody can say what was learned from cycle 2 that changed cycle 3.
• Testing only under favourable conditions. The change works for the volunteers and collapses on spread.
• No balancing measure. Improvement in the target measure achieved by displacing the problem.
• No baseline. Measurement started when the intervention started, so no shift can be demonstrated.
• Confusing implementation with testing. The team declares success at the end of a successful test and moves on without changing standard work, and the improvement decays.
• Documenting only successes. Abandoned cycles carry the highest information density in the whole record. A PDSA log with no failures is not a log of a real improvement effort.

Frequently asked questions

What does PDSA stand for?

Plan, Do, Study, Act. Plan the test and predict its result; Do it on a small scale while observing; Study the result against the prediction; Act by adopting, adapting or abandoning the change.

What is the difference between PDSA and the Model for Improvement?

The Model for Improvement is the wider framework: three questions — what are we trying to accomplish, how will we know a change is an improvement, and what change can we make that will result in improvement — followed by PDSA cycles to test the changes. PDSA is the testing engine inside it; the Model is what gives the cycles an aim and a measurement system.

What is the difference between PDSA and PDCA?

PDCA — Plan, Do, Check, Act — is the earlier formulation. The substitution of “Study” for “Check” was deliberate: checking asks whether the target was hit, while studying asks what the result reveals about the theory being tested. In practice the terms are often used interchangeably, but the distinction is the point of the method.

How long should a PDSA cycle be?

As short as it can be while still producing knowledge. AHRQ’s guidance says each cycle “should be as brief as possible for you to gain knowledge that it is working or not (some can be as short as 1 hour)”, and that a cycle may be terminated early once the result is clear — a planned one-week test that has plainly failed after four hours has produced its result and should stop.

How many patients should a PDSA cycle involve?

At the start, one. AHRQ notes a cycle “will likely involve only a portion of the practice (maybe 1 or 2 doctors)”, broadening only once the feedback is obtained and the process refined. The scale increases across a ramp of cycles rather than being set at the outset.

Does a PDSA project need IRB approval?

It depends on whether the activity is designed to develop or contribute to generalisable knowledge, not on what it is called. Many QI projects are not human subjects research; some are, particularly where the design is randomised, where participants are exposed to risk they would not otherwise face, or where generalisable publication is a primary purpose. Make the determination formally rather than assuming — see the dedicated guide on the QI versus human subjects research determination.

What measures should a PDSA project use?

Outcome measures for what you ultimately care about, process measures to show whether the change was actually implemented, and balancing measures to detect harm displaced elsewhere. Plot all of them over time, annotated with the cycles, rather than comparing a before value with an after value.

What is a PDSA ramp?

A planned sequence of cycles that increases in scale and difficulty: test with one, then a few under favourable conditions, then under deliberately unfavourable conditions (night shift, agency staff, complex patients), then at unit scale, then implement into standard work, then spread. The unfavourable-conditions step is the one most often skipped and the one that most often explains failure on spread.

Can a PDSA cycle fail?

Yes, and it should sometimes. A cycle that disconfirms the prediction has produced knowledge at very low cost, which is the entire economic argument for testing small. Abandoning a change idea after a failed cycle is a correct use of the method.

Who invented the PDSA cycle?

Its lineage runs from Walter Shewhart’s plan-do-see cycle at Bell Laboratories through W. Edwards Deming, who formulated and popularised the PDSA form. The Model for Improvement that wraps the three questions around it was developed by Associates in Process Improvement and is widely disseminated through the Institute for Healthcare Improvement, to which AHRQ’s own PDSA tool directs readers for further material.

Related reading

• Quality Improvement vs. Human Subjects Research: Do You Need IRB Review?
• Action Research: Method, Cycle, and When to Use It
• Just Culture Algorithm: How Hospitals Classify Behavior After an Adverse Event
• Morbidity and Mortality (M&M) Conference: Structure and Peer Review Protection
• Response Bias: The Main Types and How to Design Against Them
• Research Methods & Statistics