How to Do Medical Research: Topic Selection to Approved Protocol
A step-by-step guide for MD, MS and PhD researchers on choosing a research topic, framing it with PICO and FINER, selecting the right study design, calculating sample size and getting a protocol through ethics review.
Most medical research fails long before the first patient is recruited. It fails at the question, which was too broad to answer, or at the design, which could never have supported the claim the researcher wanted to make. This guide walks through the first eight steps of a clinical or laboratory research project, from finding a gap that matters to holding an approved and registered protocol. A companion guide covers the remaining ten steps, from data collection through to publication.
The Eighteen Steps at a Glance
| Phase | Steps | What you hold at the end |
|---|---|---|
| Question and design (this guide) | 1 to 8 | An approved, registered protocol |
| Execution and analysis | 9 to 16 | A locked dataset and an answered question |
| Reporting and dissemination | 17 to 18 | A submitted manuscript |
The sequence matters more than the speed. Every step skipped early becomes a reviewer comment later, and reviewer comments about design cannot be fixed by reanalysis.
Step 1: Start From a Gap, Not From a Topic
"Diabetes" is a topic. "Whether structured pharmacist-led review reduces hypoglycaemia admissions in older adults on sulfonylureas" is a gap. The difference is that a gap has a boundary, a population and a claim someone could disagree with.
Gaps come from four reliable places. The limitations paragraph of recent papers in your field, where authors name what they could not do. The "further research is needed" line of systematic reviews, which is often specific enough to lift directly. Clinical audit results in your own department, where a deviation from a standard is a question waiting to be asked. And guideline documents, where a recommendation graded on weak evidence marks an evidence gap the guideline committee has already conceded.
Write the gap as one sentence and show it to a clinician who is not your supervisor. If they cannot tell you why it matters within ten seconds, it is not yet a gap.
Step 2: Turn the Gap Into an Answerable Question
PICO is the standard framing for comparative clinical questions, and its discipline is the point. Each element forces a decision you would otherwise leave vague until analysis.
| Element | Question it forces | Vague version | Specified version |
|---|---|---|---|
| Population | Who exactly, and who is excluded? | Elderly diabetics | Adults aged 65 and over on sulfonylurea monotherapy, excluding eGFR under 30 |
| Intervention | Delivered by whom, how often, for how long? | Pharmacist review | Structured 30-minute review at 0 and 12 weeks by a clinical pharmacist |
| Comparator | Against what alternative? | Usual care | Scheduled GP review at 12 weeks with no medication reconciliation |
| Outcome | Measured how, and when? | Fewer hypos | Hospital admissions coded for hypoglycaemia within 12 months |
For non-comparative work, adapt the frame rather than abandon it. Diagnostic accuracy studies use population, index test, reference standard and target condition. Qualitative work uses population, phenomenon of interest and context. Prognostic work uses population, prognostic factor, outcome and timing.
Then apply the FINER test to the finished question. Is it Feasible with your time, funding and access? Is it Interesting to someone besides you? Is it Novel, meaning it adds rather than repeats? Is it Ethical? Is it Relevant to practice, policy or a subsequent study? A question that fails feasibility is the most common failure among doctoral researchers, and it usually fails on recruitment numbers rather than on ideas.
Step 3: Scope the Literature Before You Commit
A scoping search is not a literature review. Its only job is to tell you whether your question has already been answered, and whether enough prior work exists to justify and design your study.
- Search at least two databases, because coverage differs. PubMed or MEDLINE plus Embase is a common minimum, with CINAHL for nursing and PsycINFO for behavioural work.
- Check the trial registries at ClinicalTrials.gov and the WHO ICTRP portal. A registered but unpublished trial answering your question changes your plan.
- Check PROSPERO if you are planning a systematic review, since an identical registered review means you should adapt or move on.
- Record the search, including database, date, full strategy and hit counts. You will need it for the protocol and for the methods section.
Set a hard stop. Scoping is the phase most likely to expand indefinitely, because reading feels productive. Two focused weeks is usually enough to know whether the question survives.
Step 4: Choose the Design the Question Demands
Two practical constraints override preference. Rare outcomes push you towards case-control, because a cohort large enough to accumulate events may be unaffordable. Outcomes that take years to appear push you towards retrospective designs or towards a surrogate outcome, and if you use a surrogate you must justify why it predicts the outcome that matters.
Step 5: Define Every Variable Before You Collect One
A variable that is not defined in advance will be defined inconsistently during collection. Build a data dictionary at protocol stage, not at analysis stage.
| Field | What it records |
|---|---|
| Variable name | The exact name used in the dataset, no spaces |
| Definition | The operational definition, including thresholds |
| Type | Continuous, ordinal, nominal or binary |
| Units and permitted range | What a valid entry looks like, so implausible values are caught |
| Source | Case notes, questionnaire item, laboratory system |
| Role | Primary outcome, secondary outcome, exposure, covariate |
Name one primary outcome. Research with three primary outcomes has none, because the multiplicity inflates the chance of a false positive and reviewers will say so. Everything else is secondary and should be labelled as exploratory where appropriate.
Step 6: Sample Size Is a Design Decision, Not a Formality
The calculation needs four inputs, and each one is a claim you must defend. The expected effect size, which should come from prior literature or from the smallest difference that would change practice. The variability of the outcome, from published data or a pilot. The significance level, conventionally 0.05. And the power, conventionally 0.80 or 0.90.
State all four in the protocol along with the source of each, then add the attrition assumption. A trial powered at 200 that expects 15 per cent loss to follow-up must recruit approximately 236. For qualitative work, replace power with a justification of sample adequacy based on the concept of information power or on planned saturation, and say how you will judge that you have reached it.
Step 7: Write the Protocol as the Contract It Is
The protocol is what an ethics committee approves and what a journal will later compare your paper against. Deviations are permitted, but undeclared deviations look like outcome switching.
| Protocol section | The question it answers |
|---|---|
| Background and rationale | Why does this need doing now? |
| Objectives and hypotheses | What exactly will be tested? |
| Design and setting | What kind of study, where, over what period? |
| Participants and eligibility | Who is in, who is out, how recruited? |
| Interventions or exposures | What is done, by whom, with what fidelity checks? |
| Outcomes and measurement | What is measured, with what instrument, when? |
| Sample size | How many, and on what assumptions? |
| Statistical analysis plan | Which test for which outcome, decided in advance? |
| Data management | Where data live, who can see them, how long they are kept? |
| Ethics and consent | Approval route, consent process, participant materials |
| Dissemination | Where results go, including negative results |
For clinical trials, the SPIRIT statement gives the itemised checklist that funders and committees expect, and it is available through the EQUATOR Network alongside the reporting guideline for the finished paper.
Step 8: Ethics Approval and Prospective Registration
The Declaration of Helsinki sets the ethical frame for research involving human participants, and every institutional process descends from it. Two requirements catch researchers out most often.
The first is that approval must precede recruitment, including for retrospective chart reviews and for service evaluations that later become research. Data collected before approval usually cannot be rescued.
The second is prospective registration. The ICMJE requires that clinical trials be registered in a public registry before the first participant is enrolled as a condition of consideration by member journals. Registering after recruitment starts can make a trial unpublishable in the journals you were targeting, and no amount of good data will fix it.
Build in the waiting time. Committee review cycles, amendments and site permissions routinely take longer than a first-time researcher plans for, and the delay belongs in your timeline rather than in your contingency.
Suggested Images, Free to Reuse
These are real Wikimedia Commons files with the licence recorded on each file page. Check the page before use and attribute as the licence requires.
| Use it for | File on Wikimedia Commons | Licence |
|---|---|---|
| Illustrating trial flow | Flowchart of Phases of Parallel Randomized Trial, Modified from CONSORT 2010 | CC BY-SA 3.0 |
| A worked trial flow diagram | TwiNS CONSORT Flow Diagram of identical twins diet trial | CC BY 4.0 |
| Review screening flow | PRISMA flow diagram | CC BY-SA 4.0 |
| Historic informed consent | Informed Consent Agreement Between Antonio Benigno and Walter Reed | Public domain |
| Laboratory context | Lab technician looking through the microscope | CC BY 2.0 |
Where Ethical Support Fits
The design decisions in this guide are yours to make and to defend, because they are the intellectual content of your degree. Legitimate support sits around them rather than inside them. A PhD expert writer can talk through whether your question is answerable as framed, review a protocol draft for internal consistency between objectives, outcomes and analysis plan, check that your search strategy is reproducible, and proofread participant-facing documents for readability. What no one should do for you is choose your hypothesis, invent your data or write the sections your examiners will assess as your own reasoning.
Frequently Asked Questions
How long should the question and protocol phase take?
For a doctoral project, two to four months is common by the time ethics approval is in hand. Committee cycles dominate the back half of that, so submit early and use the waiting period to build your instruments.
Can I change my primary outcome later?
Only before you look at outcome data, and only with a documented amendment approved by the same committee. Changing it afterwards is outcome switching and is treated as a research integrity issue.
Do audits and service evaluations need ethics approval?
Frequently not, but the decision belongs to your institution and not to you. National bodies such as the Health Research Authority publish decision tools for exactly this question. Get the determination in writing before you start.
What if a similar study is already registered?
That is useful information rather than a dead end. Replication in a different population, a different setting or with a longer follow-up is publishable, provided you say plainly what is new.
How do I pick between a cohort and a case-control design?
Work backwards from the outcome. Rare outcomes favour case-control because you sample on the outcome. Rare exposures favour cohort because you sample on the exposure.
Is a pilot study worth the time?
Yes, when your uncertainty is about feasibility, recruitment rate or instrument performance. A pilot is not for estimating effect size, and reporting a pilot as though it tested the hypothesis is a common reviewer complaint.
Your Next Step Today
Write your question in full PICO form in a single sentence, then read it aloud and mark every word that could be interpreted two ways. Each ambiguous word is a decision you have not made yet, and making it now costs an afternoon rather than a resubmission.
Trusted Sources
- WMA Declaration of Helsinki, ethical principles for medical research involving human participants. Accessed 13 August 2026.
- ICMJE Recommendations, including trial registration requirements. Accessed 13 August 2026.
- EQUATOR Network, the index of reporting guidelines. Accessed 13 August 2026.
- CONSORT and SPIRIT, trial protocol and reporting checklists. Accessed 13 August 2026.
- ClinicalTrials.gov and the WHO ICTRP Search Portal. Accessed 13 August 2026.
- PROSPERO, the register of systematic review protocols. Accessed 13 August 2026.
- Health Research Authority, UK approval routes and decision tools. Accessed 13 August 2026.
Regulatory requirements differ by country and by institution. Where this guide and your own committee's guidance differ, follow the committee.
Related services
Related guides
- Medical Research: Data Collection, Analysis and Reporting
- Writing a Medical Research Proposal
- Systematic Review and Meta-Analysis with PRISMA
- Writing a Clinical Audit Report
All academic writing guides or browse our academic writing services.