Quality improvement studies
Why this level?
Specify the research question: intervention, diagnostic accuracy, prognosis, aetiology, or screening intervention. Other question types are outside this hierarchy. NHMRC 2009 sourceQuality improvement describes the purpose and context of the work, not a single level of causal evidence. The actual evaluation may be descriptive, a controlled time series or a randomized comparison; its design and assumptions determine the strength of the claim.
What question does this help answer?
Did the intended care process improve, were patient outcomes or harms affected, and what explains the observed change in this setting? Distinguish documenting local improvement from estimating the intervention's causal effect.
When to use it
Use an improvement approach when a service has a defined performance gap and a plausible change to test. Record the proposed mechanism, local context and implementation adaptations so readers can understand what was actually delivered.
When not to use it
Do not use the label quality improvement as a substitute for a study design or an oversight determination. If the claim concerns effectiveness beyond local monitoring, select a comparison capable of supporting that claim and obtain the appropriate institutional determination.
Choose measures that can disagree
- An outcome measure asks whether the result that matters improved.
- A process measure asks whether the intended change in care was delivered.
- A balancing measure looks for unintended consequences or displaced work.
- Define each numerator, denominator, data source and observation interval consistently; report missing observations and any changes in ascertainment.
Choose the evaluation, then the analysis
An uncontrolled before-and-after comparison describes a change but has limited ability to attribute it. Repeated measurements can support an interrupted time series when there is a defensible interruption and enough information to estimate trends. A suitable concurrent comparison or randomized rollout may strengthen attribution. Iterative implementation is not automatically an interrupted time series with one intervention date.
Three stories for the same change
Key assumptions and design requirements
- Prespecify the expected timing and shape of the change, while recording later adaptations honestly.
- Investigate concurrent initiatives, changes in patient mix and regression to the mean.
- Account for repeated observations, clustering, autocorrelation and seasonality when the data require them.
- Match sample-size or precision planning to clusters and time points; large patient counts alone do not make a small ward-level comparison precise.
- State a missing-data assumption and evaluate whether incomplete measurement differs across periods or settings.
Clinical example
A ward tests a discharge-medication checklist. It measures discrepancies after discharge, checklist completion and discharge delays. The team plots repeated observations, documents a concurrent pharmacy staffing change and considers an appropriate comparison ward. Checklist completion can improve while discrepancies remain unchanged or delays increase, so those findings must be reported separately.
What to watch out for
An improvement after implementation does not by itself show that the intervention caused it. Secular trends, regression to the mean, changing denominators and concurrent changes need an explicit evaluation strategy.
Before using this in your workspace
Attach the improvement aim, proposed mechanism, evaluation design and measures to the study. Keep implementation changes dated and distinguish planned comparisons from analyses chosen after seeing results.
- SQUIRE 2.0 (Standards for QUality Improvement Reporting Excellence): revised publication guidelines from a detailed consensus process (opens in a new tab) · BMJ Qual Saf · doi 10.1136/bmjqs-2015-004411
- Interrupted time series regression for the evaluation of public health interventions: a tutorial (opens in a new tab) · Int J Epidemiol · doi 10.1093/ije/dyw098