The Model for Improvement is the most widely used framework for improvement in healthcare. It asks three questions about aim, measurement, and change, and then tests changes in short Plan-Do-Study-Act cycles on a very small scale before committing to wider rollout.
This guide explains how to write an aim, choose outcome, process, and balancing measures, run small tests, and read a run chart to tell whether a change is helping. The worked example follows a unit raising the share of timely discharge summaries.
Before You Start
Why the Model for Improvement Matters
Small Tests Protect Patients
Testing a change with one clinician for one day limits risk while still producing real learning.
Keeps Improvement Practical
Three questions and a repeating cycle are simple enough for a busy unit to use without a statistician.
Uses Data, Not Opinion
Run charts give teams a quick, honest way to see whether a change is producing improvement.
Builds Toward Spread
Once a change works on a small scale, the same cycle expands it to more patients, units, or sites.
What the Model Is
The Model for Improvement was developed by Associates in Process Improvement and popularized in healthcare by the Institute for Healthcare Improvement (IHI). It has two parts: three fundamental questions, and Plan-Do-Study-Act (PDSA) cycles used to test changes. It is described in detail in The Improvement Guide by Langley and colleagues.
| Question | What the team writes down |
|---|---|
| 1. What are we trying to accomplish? | A specific aim: what, for whom, by how much, by when. |
| 2. How will we know that a change is an improvement? | A small set of measures: outcome, process, and balancing. |
| 3. What changes can we make that will result in improvement? | Change ideas, tested one small cycle at a time. |
The PDSA Cycle
| Phase | What happens | Example |
|---|---|---|
| Plan | State the change, the prediction, who does what, and what data will be collected. | One nurse will use a discharge checklist for the first two discharges tomorrow. Prediction: both summaries will be done before the patient leaves. |
| Do | Run the test and collect data. Record surprises and problems. | The nurse uses the checklist; one item is unclear. |
| Study | Compare results with the prediction and discuss what was learned. | Both summaries were done, but the unclear item slowed one of them. |
| Act | Adopt, adapt, or abandon the change, then plan the next cycle. | Reword the unclear item and test with three nurses next. |
The cycle is deliberately small at first. IHI describes starting with one patient, one clinician, or one day, then increasing scale as confidence grows. This is different from the DMAIC project style; compare with the PDCA Cycle Guide.
Choosing Measures
| Type | Question it answers | Example |
|---|---|---|
| Outcome | Is the result we care about getting better? | Percent of discharge summaries completed within 24 hours |
| Process | Are the steps we changed being done? | Percent of discharges using the checklist |
| Balancing | Did we make something else worse? | Time nurses spend at discharge, patient wait for discharge, and error reports |
Reading a Run Chart
A run chart plots a measure over time with a center line at the median of the baseline. Because it is not a control chart, it uses simple rules for spotting non-random change. Two of the most useful, following the guidance by Perla, Provost, and Murray in BMJ Quality & Safety (2011):
- Shift: six or more consecutive points all above or all below the median.
- Trend: five or more consecutive points all going up or all going down.
Points that fall exactly on the median are skipped when counting a shift. With fewer than about ten data points the chance of a false signal is higher, so collect enough baseline before interpreting.
Worked Example: Discharge Summaries Within 24 Hours
A medical unit aims to raise the share of discharge summaries completed within 24 hours of discharge from about 60% to at least 75% within four months. The data are illustrative.
- Aim: increase timely discharge summaries from 60% to 75% or more by the end of week 16.
- Measures: outcome is the weekly percent completed within 24 hours; process is checklist use; balancing is nurse and physician time spent per discharge.
- Baseline: eight weeks of data range from 58% to 65%, with a median of 61.5%.
- Test 1 (week 9): one physician uses a template for one day. It works, but one field is confusing.
- Test 2 (week 10): the fixed template goes to three physicians for a week. Completion rises.
- Test 3 (week 12 onward): the whole unit adopts the template and a daily huddle check.
The average of the last eight weeks is 74.25%, compared with 61.5% in the baseline, a gain of 12.75 percentage points. The eight-point run above the median is well past the six-point shift rule, so the improvement is unlikely to be random variation. It is still just below the 75% aim, so the team runs another cycle on the remaining delays. The balancing measures showed physician time per discharge unchanged, so nothing was traded away.
From a Small Test to a Permanent Change
The strength of PDSA is that the first test is small. A change is tried with one patient, one clinician, one day. If it works, the scale is increased step by step, and each step tests the change under new conditions.
Make a prediction. Before each cycle, write what you expect to happen and how confident you are. Comparing the prediction with the result is how teams learn, and it guards against rationalizing whatever happened.
Keep cycles short and sequential. A first cycle might have one nurse try a new discharge checklist with one patient. The second might have three nurses try a revised version. The fourth might cover the whole unit for a week. Teams that start with a unit-wide launch usually find problems too late.
Plan to fail, cheaply. A test that shows a change does not work is a success: it cost a day and saved a month. Record what was learned and adjust, adopt, or abandon the idea.
Test under different conditions. Try the change with different staff, at different times of day, and with different types of patients before calling it reliable.
Good Practice and Common Mistakes
- Keep the aim specific and owned. “By June, 90% of eligible patients will have a discharge summary sent within 24 hours” is testable. “Improve discharge communication” is not.
- Choose a few measures, including one outcome, one process, and one balancing measure that shows whether the change causes problems elsewhere.
- Collect data as part of the work. A small sample, such as five patients a day, collected simply, beats a large audit done months later.
- Plot data over time on a run chart, annotate it with the changes tested, and review it with the team.
- Record each cycle on a one-page form: the question, prediction, plan, result, and what was learned.
Common mistakes. Jumping to a large pilot; testing without a prediction; using before-and-after averages that hide the trend; changing several things at once so that nothing is learned; and stopping after a good result without building the change into standard work, so that it disappears when attention moves on. Also watch for the opposite error, which is endless testing without a decision to adopt.
Teams and sponsors. A small team that includes someone who does the work, someone who can change the process, and someone who understands the data is usually more effective than a large committee. Sponsors should ask about what was learned in the last cycle, not whether the target has been met this week. For related methods, see the PDCA Cycle Guide, the SPC Control Charts Guide, and the PDCA Learning Workbook. The figures in the examples are illustrative and do not replace your own measurements. This guide is not clinical advice.
Self-Assessment Questions
- Is the aim specific, with a numeric target and date?
- Do we have at least one outcome, one process, and one balancing measure?
- Is each test small enough to be safe and to complete quickly?
- Did we write a prediction before each test so we can learn from being wrong?
- Do we have enough baseline data to read the run chart?
Common Mistakes
Starting Too Big
A unit-wide rollout as the first test hides what is working and risks patients. Start with one clinician and one day.
No Prediction
Without a prediction the Study step becomes a review of what happened instead of learning whether the theory was right.
Ignoring Balancing Measures
A change that improves one measure by pushing work onto another part of the system is not an improvement.
Calling a Trend After Three Points
Use the run chart rules and enough baseline data. Reacting to every wiggle wastes effort and erodes trust in the data.
Model for Improvement and PDSA: Frequently Asked Questions
What is the Model for Improvement?
It is an improvement framework built around three questions: what are we trying to accomplish, how will we know that a change is an improvement, and what changes can we make that will result in improvement. Changes are then tested through PDSA cycles. It was developed by Associates in Process Improvement and widely used in healthcare through the Institute for Healthcare Improvement.
What is the difference between PDCA and PDSA?
They describe the same cycle. PDSA replaces the word Check with Study to emphasize learning from the results and comparing them with the prediction, rather than only checking whether the plan was followed. Many healthcare improvement teams use PDSA for that reason.
What is a balancing measure?
A balancing measure looks at the other side of a change: whether improving one part of the system causes a problem elsewhere. For example, if a team speeds up discharge, a balancing measure might track readmissions, staff overtime, or patient complaints.
Sources and Further Reading
- Gerald J. Langley and colleagues, The Improvement Guide: A Practical Approach to Enhancing Organizational Performance.
- Institute for Healthcare Improvement, Model for Improvement and PDSA resources.
- Jennifer Perla, Lloyd Provost, and Sandra Murray, "The run chart: a simple analytical tool for learning from variation in healthcare processes," BMJ Quality & Safety, 2011.
- W. Edwards Deming, The New Economics for Industry, Government, Education.