How to collect data for a healthcare QI project

To collect data for a quality improvement project, start with agreed measures, use a reliable source, record observations consistently and check them before review. Keep collection manageable, but do not sacrifice the definitions or coverage needed to answer your improvement question.

This article covers the evidence stage of our guide to running a successful QI project. If you have not yet decided what to measure, start with a QI measurement plan. Here, the focus is putting that plan into daily practice.

What data should you collect for a QI project?

Collect the observations needed for your outcome, process and balancing measures, plus enough context to interpret them. Numerical data can show how a process changes over time; patient and staff feedback can help explain what those changes mean.

For an outpatient waiting-time project, this might include arrival and consultation-start times, whether preparation was completed before the clinic, staff overtime and a brief patient-experience question. Avoid collecting extra fields simply because they may be interesting later.

1. Check whether suitable data already exists

Speak with the people who record and manage the information. A familiar field name does not guarantee that the data matches your definition: “appointment time” may mean the booked slot, the patient’s arrival or when the consultation actually began.

Before reusing an extract, check:

• Whether it covers the patients, setting and dates in your project.
• How each field is recorded and whether that practice has changed.
• How quickly data becomes available after the event.
• Whether missing records, duplicates or corrections are visible.
• Whether you can access the detail needed for your calculation.

Compare a small set of extracted records with the source and the actual workflow. Existing data can reduce effort, but it is not automatically more accurate than a well-designed local collection process.

2. Choose a collection method that fits the question

Routine records and system extracts

Use these for information already captured consistently, such as activity counts or timestamps. Agree the extraction rules with the data owner and keep a record of filters and calculations so the next extract is comparable.

Observation or a simple tally sheet

Use a short form to capture an event or process step that is not recorded elsewhere. For example, the person preparing a clinic could record whether each room-readiness item was complete by the agreed time. Define what “complete” means and avoid relying on memory at the end of a busy week.

Brief surveys and patient or staff conversations

Use feedback to understand experience, effort and unintended consequences. Ask clear, consistent questions and offer a practical way to respond. Record how people were invited and how many responded; a small voluntary sample may miss people with different experiences.

Keep qualitative notes separate from numerical scores. Summarise recurring themes and exceptions without turning a few comments into a claim about everyone.

3. Make the collection instructions easy to follow

Give each collector the same short instructions: who or what is eligible, when to record, which fields are required, what each response means and where to put the information. Include a worked example and explain how to flag uncertainty.

Name an owner and a backup. If collection depends on one enthusiastic person being present, it is unlikely to survive leave or a rota change.

Where sampling is needed, agree the method before collection. Spread observations across the conditions relevant to your project rather than choosing only convenient cases. Keep the approach consistent and record changes. The right sample depends on the question and the process; a small sample is not automatically sufficient for every conclusion.

4. Pilot the collection process

Test the form or extract during one session or with a small set of records. Ask two collectors to apply the instructions to the same examples and compare their results.

Look for ambiguous fields, missing categories, unnecessary work and information that becomes available too late. Record how long collection takes. Fix the method, update its version date and explain the changes to everyone involved before expanding it.

Healthcare example: collecting outpatient waiting-time data

This is an illustrative collection method, not a report of patient results. A team is working on adult Tuesday morning general outpatient clinics. Its outcome measure is median arrival-to-consultation waiting time for each session.

The team first confirms how arrival and consultation-start timestamps are recorded. Its working sheet contains a session date, a locally approved anonymous or coded record reference, the two timestamps, calculated wait in minutes and a data-quality flag. Identifiable source information remains in the organisation’s approved system.

After each clinic, the data lead checks eligible records, calculates waiting times and records the number of usable and missing observations. Non-attenders and cancelled appointments are handled according to the agreed measurement definition rather than assigned a zero wait.

Alongside the waiting-time sheet, the team records whether the preparation checklist was completed, staff overtime and responses to its patient-experience question. These observations answer different questions and should not be blended into one score.

What if a timestamp is missing or looks wrong?

Flag the record and check the original source where possible. For example, a consultation-start time earlier than arrival may indicate an entry error or a misunderstanding of the field. Correct it only when there is evidence for the correction, and retain a record of the change.

If the value cannot be verified, apply the agreed missing-data rule. Do not guess it, replace it with zero or quietly remove it. Report how much data is missing and consider whether missingness is concentrated in busy sessions or particular groups.

5. Check data quality before each review

• Completeness: are expected sessions or records missing?
• Validity: are dates, units and values plausible?
• Consistency: have definitions, sources and collection methods stayed the same?
• Duplicates: has the same observation been counted twice?
• Coverage: does the dataset reflect the intended population and conditions?
• Traceability: can someone understand how the reported measure was calculated?

For percentages, retain the numerator and denominator rather than storing only the final percentage. A result based on five responses and one based on fifty responses should not be treated as interchangeable.

6. Turn the data into learning

Display measures in time order using a chart suited to the data. Mark when tests occurred and add relevant context, such as staffing changes or disrupted sessions. Review outcome, process and balancing measures together.

Ask what the pattern suggests, what remains uncertain and what you will test next. Do not treat one good week as proof of improvement. The IHI’s guidance on establishing measures provides further context on using measurement to support improvement.

Keep a short note of the team’s interpretation and resulting decision. Data collection has value when it informs action, including a decision to adapt or stop a change.

Frequently asked questions

Can we use a spreadsheet to collect QI data?

Yes, where it meets your organisation’s requirements and the team can manage access, definitions and versions reliably. Agree one working source and protect formulas from accidental edits. More complex collaboration may benefit from a shared improvement platform.

Should we collect data before testing a change?

Comparable baseline data helps the team understand the starting process. You may also need small, focused observations during early tests to learn whether the change is workable. Keep the purpose of each dataset clear.

What if data collection becomes too time-consuming?

Review whether every field supports a decision, whether a reliable existing source is available and whether an agreed sampling approach would help. Pilot the revised method and document any effect on comparability.

Use your data in the next test

Continue with how to use PDSA cycles to test changes. For a shared way to display and discuss measures, explore Simana’s charting tools.