How to pick a research question that actually gets published

Almost every unfinished student research project we see failed for the same reason, and it was not statistics, motivation or supervision. It was the question. It was too big, too vague, or it required data that was never going to exist.

The good news is that this is the cheapest failure to prevent, because a bad question is identifiable in an afternoon. The habit worth building is to interrogate a question hard before you commit ten weeks to it.

Admiration is not feasibility

Students tend to propose the question they most admire: whether a novel biomarker predicts outcome in sepsis, whether a new technique changes survival. These are excellent questions. They are also questions that require a funded cohort, a multi-year follow-up, or access to a specialist lab, and you have a laptop and ten weeks between rotations.

The alternative is not to lower your ambition. It is to change the unit of ambition from the size of the topic to the quality of the execution. A tightly executed analysis of a modest question gets published. A sprawling, half-finished attempt at an important one does not get published at all — and an unpublished project teaches you very little, because the teaching in research is concentrated in revision and peer review, which only happen after submission.

Four tests to run before you commit

The data test. Where do the numbers come from, and do they already exist? If the answer involves recruiting patients, get realistic about how many you can consent per week and multiply honestly. If the answer is an existing database — a public mortality file, a registry, a hospital record set you already have permission to access, or the published literature itself in a systematic review — your project has a floor under it.

The finish test. Can you produce a complete first draft in ten weeks working part-time? If the honest answer is no, cut scope until it is yes. A question narrowed to one outcome in one population over one clearly bounded period is not a lesser question; it is a finishable one.

The novelty test. Search PubMed properly — not one query, but four or five with synonyms and MeSH terms — and read what already exists. If you find nothing at all, be suspicious: either your search is wrong or the question is uninteresting to the field. What you want is a small, defensible gap: the same question in a different population, an updated time window, a stratification nobody applied, a synthesis of trials that have never been pooled.

The so-what test. Write one sentence describing what a clinician, a policymaker or the next researcher does differently if your result comes out as expected. If you cannot write that sentence, the question is not ready — and the sentence you eventually write becomes the last line of your abstract.

Where good student questions come from

Three sources produce most of the publishable student work we mentor.

Public secondary data. Mortality and natality files, cancer registries, national surveys and open trial repositories all permit descriptive and analytical work with no ethics submission and no recruitment. The novelty comes from the combination you choose, and the craft comes from the trend modelling.

Systematic reviews and meta-analyses. These are the highest-value first projects for a student who cannot access patient data, because the raw material is the published literature. The design is prespecified, the reporting standard is PRISMA, the protocol can be registered on PROSPERO before you begin, and a well-conducted synthesis sits high in the evidence hierarchy. They are not easy — screening several thousand abstracts in duplicate is genuine labour — but they are entirely doable by a disciplined pair of students, and they teach critical appraisal better than any course.

Clinical audit and local practice questions. If you have access to a department's records, the gap between guideline and practice is an inexhaustible and genuinely useful source of questions. These publish well in specialty and regional journals, and they occasionally change something.

Sharpen it into a structured statement

Once you have a candidate, force it into a structure. PICO for comparative and interventional questions: population, intervention or exposure, comparator, outcome. PEO for descriptive and qualitative ones: population, exposure, outcome. The structure is not bureaucracy — it is a diagnostic, because whichever element you cannot fill in is precisely the element that will sink you later.

Compare a vague question — does obesity affect heart disease — with a structured one: among US adults aged 45 to 64, how did age-adjusted mortality from ischaemic heart disease with obesity recorded as a contributing cause change between 1999 and the most recent finalised year, and did the trend differ by sex and urbanisation? The second is answerable, its data source is implied, its analysis is implied, and its search strategy writes itself.

Write that structured statement down and put it at the top of your protocol. Every subsequent decision — inclusion criteria, variables, the statistical test, the first table — is a consequence of it. When a supervisor asks why you included a variable, the answer should be traceable to that sentence.

Then choose the journal early

Most students pick a journal after the paper is written, which is backwards. Identify two or three realistic target journals while the question is still forming, read their scope statements and three recent papers from each, and note their word limits and reporting requirements. You will learn what that field considers a contribution, which is the most direct available answer to whether your question is publishable.

Check indexing and legitimacy at the same time. A student's first paper should be in a journal indexed where it will be found and where the peer review is real. Learn to recognise a predatory invitation now, before one arrives in your inbox promising review in seventy-two hours for a fee.

Get the question right and the rest of the project is work — hard, ordinary, finishable work. Get it wrong and no amount of effort downstream will rescue it. This is why in our programme nobody touches a dataset until their question has survived all four tests, in writing, with a mentor.