How to Choose a Primary Outcome Before You Calculate Sample Size
A clear primary outcome turns a broad research question into a study that can be measured, analyzed, and powered correctly. Here is a simple way to choose one.
Physician-data scientist at King Chulalongkorn Memorial Hospital, Bangkok. Research interests: Machine learning, causal inference, and AI in healthcare.
Before you calculate sample size, choose your primary outcome.
This sounds like a small step, but it controls almost everything that comes next: the study design, the statistical test, the required number of participants, and how reviewers judge whether your study answered its question.
A research question can be broad. A primary outcome must be specific. It is the main thing you will measure to decide whether your study was positive, negative, or inconclusive.
Think of it this way: the research question is the destination. The primary outcome is the address.
What Is a Primary Outcome?
A primary outcome is the most important result your study is designed to measure.
For example, if your research question is:
"Does a new discharge planning program reduce hospital readmissions?"
Your primary outcome might be:
"All-cause hospital readmission within 30 days after discharge."
That outcome is clear because it says:
- What is being measured: hospital readmission
- Who it applies to: discharged patients in the study
- When it is measured: within 30 days
- How it will be counted: yes or no
This level of detail matters. "Readmission" alone is too vague. A patient readmitted after 7 days is different from one readmitted after 1 year. A planned readmission is different from an emergency readmission. If the outcome is vague, the analysis becomes vague too.
Why the Primary Outcome Comes Before Sample Size
Sample size calculation is not a generic math exercise. It depends on the exact outcome you choose.
A continuous outcome, such as systolic blood pressure, uses a different calculation than a binary outcome, such as death or readmission. A time-to-event outcome, such as time to cancer recurrence, uses a different calculation again.
The primary outcome also determines the effect size you need to detect.
For blood pressure, the effect size might be a 5 mmHg difference. For readmission, it might be a drop from 20% to 15%. For survival, it might be a hazard ratio of 0.75.
If you change the primary outcome after calculating sample size, the original sample size may no longer be valid. That is why the order matters:
- Define the research question.
- Choose the primary outcome.
- Choose the statistical test.
- Calculate the sample size.
If you skip step 2, the rest is built on soft ground.
A Simple Test: Can One Sentence Define It?
A strong primary outcome should fit into one clear sentence:
"The primary outcome is [what you measure], assessed [when], using [how it is measured], in [which patients]."
Here are two examples.
Weak:
"The primary outcome is improvement in diabetes control."
Stronger:
"The primary outcome is change in HbA1c from baseline to 6 months among adults with type 2 diabetes."
Weak:
"The primary outcome is complications after surgery."
Stronger:
"The primary outcome is any Clavien-Dindo grade III or higher postoperative complication within 30 days after surgery."
The stronger versions remove guesswork. They tell the statistician what kind of variable this is, tell the clinical team what data to collect, and tell reviewers exactly what success means.
Five Questions to Ask Before You Finalize It
1. Is it clinically meaningful?
Do not choose an outcome only because it is easy to measure.
A biomarker can be useful, but only if it reflects something patients, clinicians, or health systems care about. A small laboratory change may be statistically significant but clinically unimportant.
Ask: if this outcome improves, would anyone change practice?
2. Can it be measured reliably?
An outcome must be collected the same way for every participant.
Blood pressure measured after 5 minutes of rest is not the same as blood pressure measured quickly in a busy clinic. Pain measured with a validated scale is easier to interpret than pain described loosely in chart notes.
If measurement is inconsistent, your study may look noisy even when the intervention works.
3. Is the time point clear?
Many outcomes change over time.
Weight loss at 1 month may look different from weight loss at 12 months. Infection within 7 days is different from infection within 90 days. If you do not define the time point before the study starts, you create room for biased choices later.
Pick the time point that best matches the biology, clinical workflow, and patient impact.
4. Does it match the study design?
The outcome should fit what your design can actually observe.
A short pilot study should not use 5-year mortality as its primary outcome. A retrospective chart review should not rely on a patient-reported symptom that was rarely recorded. A single-center study may struggle with rare events.
Good outcome selection is not just about importance. It is also about feasibility.
5. Can you power the study for it?
Some outcomes require very large sample sizes.
If the event rate is low, such as 1% mortality, detecting a meaningful reduction may require thousands of participants. That does not mean the outcome is wrong, but it may mean the design is not feasible.
This is where outcome choice and sample size planning meet. If the best outcome makes the study impossible, consider a different design, a higher-risk population, a longer follow-up period, or a validated intermediate outcome.
Primary Outcome vs. Secondary Outcomes
Most studies measure more than one outcome. That is fine.
The primary outcome is the main decision point. Secondary outcomes add context.
For a discharge planning study, the primary outcome might be 30-day readmission. Secondary outcomes might include emergency department visits, patient satisfaction, length of stay, and cost.
The danger is treating every outcome as equally important. If you test many outcomes and highlight only the positive ones, the study becomes hard to trust. This is one reason protocols and trial registries ask you to define outcomes in advance.
A useful rule:
- One primary outcome answers the main question.
- Secondary outcomes explain the result.
- Exploratory outcomes generate ideas for future research.
Common Mistakes
1. Choosing too many primary outcomes. Multiple primary outcomes can be valid, but they require careful planning and often a larger sample size. If every outcome is primary, none of them really is.
2. Using an outcome that is too broad. "Clinical improvement" needs a definition. Improvement by whose judgment? Measured with what tool? At what time point?
3. Picking an outcome after seeing the data. This is outcome switching. It makes the result look stronger than it is and is a common reason reviewers lose confidence in a study.
4. Choosing a surrogate outcome without justification. A surrogate, such as a lab value, should have a clear link to patient benefit. If that link is weak, say so and treat the study as preliminary.
5. Forgetting the analysis plan. The outcome and statistical test must match. A binary outcome, continuous outcome, ordinal scale, and time-to-event outcome each need different methods.
A Practical Template
Before you calculate sample size, write this in your protocol:
- Primary outcome name
- Exact definition
- Measurement tool or data source
- Assessment time point
- Variable type: continuous, binary, ordinal, count, or time-to-event
- Expected value in the control group
- Minimum clinically important difference
- Planned statistical test
Example:
"The primary outcome is 30-day all-cause hospital readmission, defined as any unplanned inpatient admission within 30 days after discharge, measured from hospital administrative records. This is a binary outcome. Based on prior hospital data, the expected control event rate is 20%. The study is powered to detect an absolute reduction to 15% using a two-sided test of proportions."
That paragraph gives enough information for a statistician, ethics board, and reviewer to understand the logic of the study.
Where ProtoCol Fits
ProtoCol is built around this sequence: define the gap, design the methodology, then calculate the sample size.
During planning, ProtoCol helps turn a broad research idea into a specific PICOTS question, checks whether the outcome fits the study design, and then uses that outcome to guide the biostatistical plan.
That structure matters because a sample size number is only as good as the outcome behind it.
If the primary outcome is clear, the statistics become much easier to defend.
If the primary outcome is unclear, even perfect arithmetic will not save the protocol.