13 Research methods, statistics, critical review and evidence-based practice13.11 Clinical governance
13.11 Audit, change planning, feedback and PDSA
4 min read3 figures
Written by Resident Doctor Committee · Published
Quality improvement, audit and research
Quality improvement is the systematic, continuous, systems-focused and blame-free effort to make care safer and better, joining analysis of system performance to tested changes.
Clinical audit compares care with an explicit standard, acts on the gaps and measures again, so it covers only one part of quality improvement.
Service evaluation asks how an existing service performs.
Research tests interventions whose benefit is not yet established, whereas quality improvement tests changes that stay within accepted standards of care.
Randomising treatment generally makes a project research needing Research Ethics Committee review, which audit and service evaluation do not need on those grounds alone.
A local project that finds affected patients, reviews them, educates staff and changes the prescribing system to close a known safety gap is quality improvement, not observational research.
SQUIRE is the consensus reporting guideline for formal studies of healthcare quality improvement.
The audit cycle
Define the criterion, target standard, eligible denominator and legitimate exceptions, such as refusals, before collecting data.
Closing the cycle means acting on the gaps, then measuring again on comparable definitions.
Clinical governance and incident reporting
Clinical governance holds organisations accountable for the quality and safety of care through incident learning, complaints, mortality review, audit, appraisal and regulation.
A patient safety event is formally recorded through the organisation's incident reporting system before any Plan-Do-Study-Act (PDSA) work begins.
Under the statutory duty of candour, being open or apologising is not an admission of liability.
The Model for Improvement
The Model for Improvement (Langley, Nolan and colleagues) opens with three fixed questions, then tests changes in PDSA cycles.
Question | What it defines |
|---|---|
What are we trying to accomplish? | The aim |
How will we know a change is an improvement? | The measure |
What changes might produce improvement? | The changes to test |
The questions are revisited as PDSA cycles generate learning, not settled once as a rigid prerequisite.
Plan-Do-Study-Act cycles
PDSA cycles test a change on a small scale against a prediction made beforehand, refining it before wider implementation.
Stage | What happens |
|---|---|
Plan | Set out the change, roles, timing, measures and prediction |
Do | Run the change and record what happens |
Study | Learn from the data by comparing results with the prediction |
Act | Adopt, adapt or abandon the change for the next cycle |
A reminder tested on five patients predicts four complete checks; when broken equipment leaves two complete, the team fixes the process instead of sending more reminders.
PDCA (Plan-Do-Check-Act) is the older wording suited to error correction and fault detection, where Check means inspection against a standard.
Clinical audit also cycles, but it measures compliance with a standard rather than testing a change against a prediction.
Other improvement approaches
Approach | What it does |
|---|---|
FOCUS-PDCA | Five preparatory steps (find a process, organise a team, clarify what is known, understand why it varies, select the improvement), then PDCA |
DMADV (Six Sigma) | Define, measure, analyse, design, verify or validate, to build a new process or service to a set standard; DMAIC improves an existing one |
Lean Thinking | Removes waste: defines value from the user's view, maps the value stream and strips out steps that add no value |
Root cause analysis | Looks back at a single incident, used only after harm has occurred |
Failure mode and effects analysis | Looks ahead for failures proactively |
PDSA is prospective and iterative, while Lean and statistical process control look at whole ongoing processes.
Tools for understanding a problem
Tool | Layout | What it shows |
|---|---|---|
Flow chart (process map) | Stages of a pathway in order, with standard symbols for actions, decisions and hand-offs | Waits, loops and duplicated steps |
Fishbone (Ishikawa, cause and effect) diagram | Problem at the head, possible causes grouped under category headings off a central spine | Root causes brainstormed before data are gathered |
Pareto chart | Counted causes as bars from most to least frequent, with a cumulative percentage line | The vital few causes behind most of the problem (80/20 rule) |
Driver diagram | Aim, then broad primary drivers linked directly to it, narrower secondary drivers and specific testable change ideas | Change ideas, each testable in a PDSA cycle |
The links in a driver diagram are hypotheses to test, not established causes.
Measuring change over time
Measure | Example | What it shows |
|---|---|---|
Outcome | Median length of stay | Whether the targeted result changes |
Process | Discharge date recorded within 72 hours | Whether the intended process changes, not causation |
Balancing | 28-day readmission | Consequences elsewhere |
Chart | Centre line | Limits | Signal |
|---|---|---|---|
Run chart | Median | None calculated | Patterns such as six points on one side of the median |
Control chart (Shewhart or statistical process control chart) | Mean | Upper and lower limits at three standard deviations | A point beyond a limit, meaning special cause variation rather than routine noise |
Both charts plot points in time order, and a signal does not reveal its own cause.
Feedback and interpreting results
A 2025 Cochrane review found audit and feedback gave a median absolute improvement of 2.7 percentage points (interquartile range 0.0 to 8.6).
Feedback works better from a trusted source, with relevant benchmarks and a clear statement of what to do.
A before-and-after improvement may reflect a secular trend, regression to the mean, an observation effect or a changing case mix.