Scientify

Medical Research Basics / Topic Selection

How medical students can use AI to find a research topic: from clinical question to feasible study

Many first research projects stall in one of two ways: the topic is far too broad, or weeks of reading reveal that someone already did the study. A good topic usually starts with a specific clinical question and becomes feasible only after a literature search, PICO narrowing and a practical feasibility check. This guide walks through that path, where AI saves time, and where it cannot make the call for you.

Short answer

Medical students should start from a specific clinical question raised on rotations or in case discussions, not ask AI to invent a topic. Search existing work to confirm the question is still open, define population, intervention or exposure, comparison and outcome with PICO, then check sample, data access, ethics approval and statistical feasibility, and have your mentor confirm the direction.

Challenges

Four places medical students get stuck

The hard part is rarely coming up with an idea. It is judging whether an idea is worth pursuing and whether you can actually finish it.

Not knowing which questions are still open

Textbooks teach settled consensus. The real research space lives in recent papers, guideline updates and discussion sections, which are hard for a beginner to survey.

Manual searching is slow and misses things

The same concept can appear under different terms, abbreviations or subject headings. Narrow keywords miss key studies and create the illusion of a gap.

Too much literature, hard to judge quality

Case reports, observational studies, randomized trials and systematic reviews carry different weight. Beginners often treat one small study as settled, or drown in low-quality papers.

A missing step between topic and question

"Diabetes complications" is a topic, not a research question. Only after you define who you study, what you compare and which outcome you measure can you tell what data and methods you need.

PICO

Turn a clinical question into a research question with PICO

PICO is a standard question framework in evidence-based medicine. Once all four elements are explicit, the question is ready to search and shows what data the study will need.

ElementWhat it asksVague versionNarrowed version (example)
P PopulationWho are the subjects? What are the inclusion and exclusion criteria?Patients with diabetesInpatients aged 65 and older with type 2 diabetes at your hospital
I Intervention or exposureCan the treatment, exposure or factor be clearly defined and measured?On a new drugRegular use of a specific class of glucose-lowering drug for at least 3 months before admission
C ComparisonCompared with whom? How is the comparison group defined?People not on the drugPatients treated with a different drug class over the same period
O OutcomeWhich result, measured how and at what time point?Better resultsAll-cause readmission within 90 days of discharge
The table illustrates how to phrase each element and is not a study finding; real definitions should follow guidelines and your mentor's advice.

Workflow

Five steps from clinical curiosity to research question

  1. 01

    Write down one specific clinical question

    Start with something you noticed on rotations or in case discussions, such as why a lab value in a certain patient group keeps shifting on a particular postoperative day. The more specific it is, the easier it is to search and narrow.

  2. 02

    Search the literature and check for duplicate studies

    Search PubMed and similar databases, and check trial registries and systematic review registries for published, ongoing or registered studies asking the same question. AI can expand synonyms and subject headings here, but open the key papers yourself.

  3. 03

    Narrow the question with PICO

    Break the question into population, intervention or exposure, comparison and outcome. An element you cannot pin down usually marks where the topic is too broad or the data is out of reach.

  4. 04

    Check feasibility and draft a one-page brief

    Go through sample, data access, ethics, statistics and timeline, then put the background, research question, three to five key papers and a preliminary method on one page for discussion.

  5. 05

    Have your mentor review and confirm the direction

    A mentor can judge clinical relevance, department resources and whether the data source is realistic. The final direction, study design and ethics submission follow your mentor's and your institution's requirements.

Feasibility

Research topic feasibility checklist

A well-formed PICO only means the question is searchable, not that you can finish it. Before committing to a topic, work through every item below.

  • Existing evidence: Is there a published, ongoing or registered study that is the same or very similar? If so, how does your question differ in population, design or outcome?
  • Sample and data: Can you get enough cases or samples in the time you have? Will the data come from medical records, an existing cohort, a public database or prospective collection?
  • Data access: Do you need permission from the hospital, department or data provider? Some public clinical databases require registration, training and a signed data use agreement before access.
  • Ethics approval: If the study uses patient data or human samples, does it need review by an ethics committee or IRB, or a waiver of informed consent?
  • Statistical feasibility: Can the expected event count or effect size for the primary outcome support the sample size and analysis you need? Is there a statistician or faculty member you can consult?
  • Completion conditions: Are the assays, equipment, software and time available? Can you finish before your graduation or program deadline?

Scientify

Use Scientify to test a topic before you commit

Scientify is a science agent that runs in an isolated cloud computer. Give it your clinical question and draft PICO, and it keeps organizing search strategies around that question, looks for identical or similar studies, tries to retrieve full-text PDFs, and summarizes the populations, designs and outcomes of existing work so you can judge whether the gap is real.

If the question can be tested first on a public dataset, the agent can download the data, write analysis code, run descriptive statistics or a preliminary model, and refine the question based on the results. Literature, code, dependencies, data, logs, charts and results stay in one workspace. The task keeps running after you close your laptop, you can check progress from your phone, and you end up with reproducible code, data and a report to bring to your mentor.

  • Conversation history and workspace files are stored only in the isolated cloud computer; Scientify servers do not store this research data
  • The isolated cloud computers are provided by a SOC 2 Type II certified provider, and no copies are retained after deletion
  • The core agent is open source with 2k+ GitHub stars; the related paper was accepted to ICML 2026
  • New users get a free cloud computer and $5 in model credit

Sources

  • PubMed — Biomedical literature search
  • ClinicalTrials.gov — Registered and ongoing clinical trials
  • PROSPERO — Systematic review registry, useful for checking whether the same review is underway

FAQ

Can I just ask AI to give me a few research topics?

It is not a good idea. Topics generated from scratch tend to be too broad or already well studied. A more reliable approach is to start from a clinical question you actually encountered, let AI expand search terms and summarize existing studies, and then judge for yourself whether the data and your skills can support it.

Why do I need to verify a topic against real literature?

Only real published and registered studies tell you whether the direction is already saturated and how your question differs from existing work. AI-generated references may not exist or may contain errors, so open the original of every paper that goes into your proposal.

What research can I do without access to hospital data?

Options include secondary analysis of public databases, systematic reviews and meta-analyses. Public databases may still require registration and a data use agreement, and the specific design should be confirmed with your mentor.

Can Scientify handle ethics approval or decide my study design for me?

No. Scientify can search the literature, try to retrieve full texts, and write code and run preliminary analyses on data you are authorized to use. Ethics approval, data authorization and the final study design are decisions for you, your mentor and your institution.

Keep reading

Generating research hypotheses with AITurn the literature into testable hypotheses and plan how to test them.Research data privacyConversation history and workspace files stay only in the isolated cloud computer.From literature to reproducible experimentsHow Scientify keeps search, hypotheses, code and experiments in one workspace.

Start testing your topic from one clinical question

Write down your question and draft PICO. The agent searches for similar studies in the cloud, tries to retrieve full texts and runs preliminary analyses on public data. New users get a free cloud computer and $5 in model credit.

Start researching