“Machine learning in healthcare” is an area. “Does an automated triage model reduce time to specialist referral in a district hospital emergency department?” is a topic. The second can be answered; the first cannot.
Start from what you can actually access
Feasibility should shape the topic before ambition does. Ask what data, patients, samples, equipment, permissions and time you genuinely have. A brilliant question you cannot resource produces nothing.
Read the limitations sections
The most reliable source of topics is the final paragraph of other people's papers. Authors state what they could not do and what should be studied next. Read twenty limitations sections in your field and the recurring gaps become obvious.
Our guide to finding research gaps systematically covers this in more depth.
Narrow along four axes
- Population. Which group specifically, and why that one?
- Variable. Which exposure, intervention or factor?
- Outcome. Measured how, with what instrument?
- Context. Which setting, region or time period?
Fixing all four turns an area into a question. If any is still vague, the study design will be vague too.
Confirm it has not been answered
Search before committing. If you find the study already done, read it carefully — the useful move is often to replicate it in a different population, or to test whether its conclusion holds under different conditions.
Test it against three questions
- Can I collect or access the data within my timeframe?
- If the result is negative or null, is the paper still publishable?
- Who would cite this, and why?
The second question matters most. A study whose only publishable outcome is a positive finding creates pressure toward exactly the practices that ruin careers.
Write the abstract first
Before collecting anything, write the abstract you hope to publish. If you cannot write a coherent one from the plan, the design has a hole in it. This takes an hour and routinely saves months.
When the topic is settled, our guide to writing the paper itself covers the route from question to submission.