Formulating a Research Question

#Define who and what the study concerns

A useful research question states who is being studied and what the study aims to learn. Describe the population in terms relevant to the question, such as the care setting, symptoms or stage of a condition. Avoid defining the group more broadly than recruitment can realistically support.

State the intended use of the measurement, test or model. Screening people without symptoms differs from investigating symptoms or monitoring an established condition. Specify where the method would fit in an existing process and what information would be available when it is used.

#Choose a comparison and meaningful outcome

The comparison should match the decision under study. It might be an existing method, usual care or another clearly defined approach. A weak comparison can make a new method appear useful without showing whether it adds value over a realistic alternative. Explain why the chosen comparison is appropriate.

Define the outcome and when it will be assessed. Measurement accuracy, symptom change and health outcomes answer different questions. If an outcome is only an indirect indicator of patient benefit, make that distinction explicit. Consider possible harms and burdens alongside the hoped-for benefit, rather than treating accuracy as the only relevant result.

#Let the question guide the design

Descriptive questions may need a survey or observation of a defined group. Prediction questions require evaluation using information available before the predicted outcome. Questions about whether an intervention causes a change often benefit from random assignment where feasible and ethical. No design is best for every purpose.

Check that the planned data and analysis can answer the stated question. Separate the main question from secondary or exploratory questions. Record important assumptions and practical constraints. If the question changes after data are examined, explain the change and present the resulting analysis with appropriate caution.

#Common misunderstandings

A research question is not the same as a broad topic or a prediction. “Does this treatment help?” leaves important details unresolved, including what “help” means and when it would be measured. A useful question makes those choices explicit without assuming the answer.

Another misunderstanding is that every question needs a randomised trial. Trials can help assess treatment effects, but other designs may better address experiences, uncommon harms or how care works in routine practice. The design should fit the question, while recognising what it cannot establish.

A statistically significant result does not automatically mean a benefit matters to patients. Equally, an inconclusive finding does not prove there is no effect; the study may lack precision.

Finally, questions developed after examining results can be valuable, but they should be labelled as exploratory. Presenting them as planned questions can make findings appear more convincing than the evidence supports.

#Questions worth asking a clinician

• How should we define the patient population and intended clinical use so our research question is focused enough to guide study design?
• Which comparison best reflects the relevant clinical alternative for our population, such as usual care, another intervention, or no intervention?
• Which outcomes should we specify in advance to answer our research question and reflect what matters to patients?
• When should we assess each outcome to capture meaningful benefits or harms for the intended clinical use?
• Is our question descriptive, predictive, or causal, and what study design would let us answer that type of question?