How to do a systematic review as a medical student (even without a mentor)

Can a medical student do a systematic review alone, without a mentor?

You can lead a systematic review without a mentor, but you can't do every step on your own. The Cochrane Handbook, the standard methods reference for systematic reviews, says at least two people must work independently to decide which studies are included and to extract the outcome data. The real job, then, is finding a second reviewer and, ideally, someone experienced to check a few key decisions.

Early on, a mentor told one of us that you could do 99% of a project yourself and it still wouldn't be a whole project. A systematic review makes that literal. It answers one focused question by finding every relevant study with a pre-planned, repeatable method, judging how trustworthy each one is and combining their results. One person can carry most of that; the missing 1% is a second pair of eyes on two steps.

Here's how the work splits:

• What you can do on your own
• What needs two people
• What skipping the second person costs

What you can do on your own: the question, the protocol, the search, the analysis and the writing. Cochrane doesn't even want two people running the same searches, though it strongly recommends a librarian peer reviews your search strategy first (Cochrane Handbook, section 4.2.2). One person screening titles and abstracts is acceptable too, although two is better (section 4.6.4).

What needs two people: the final decision on each full-text article, and pulling out the outcome data. Both are mandatory in Cochrane's conduct standards (MECIR items C39 and C46), because doing them twice, independently, cuts both honest mistakes and one person's bias (sections 4.6.4 and 5.5.2). PRISMA 2020, the reporting guideline journals expect for systematic reviews, then asks you to state how many reviewers did each step and whether they worked independently (Page et al., BMJ 2021, items 8 and 9). If you did them alone, the editor can see it.

What skipping the second person costs: In a randomised trial run on Cochrane's crowdsourcing platform, a single reviewer screening abstracts missed 13% of relevant studies, while two reviewers missed 3% (Gartlehner et al., 2020). Extraction goes wrong often too: one study cited in the Handbook found data extraction errors in 20 of 34 reviews. And AMSTAR 2, a standard tool for appraising systematic reviews, has two items asking whether study selection and data extraction were done in duplicate (Shea et al., BMJ 2017).

Honestly, the second reviewer is easier to find than most students think. These four routes work from anywhere:

• A classmate or batchmate
• A junior who wants to learn
• A friend at another college or in another country
• A swap with someone running their own review

A classmate or batchmate: They don't need to know your topic. Cochrane even notes that a second reviewer who isn't a content expert can be an advantage, because experts tend to bring pre-formed opinions (section 4.6.4). They need time, care and an hour's training on your criteria.

A junior who wants to learn: Screening and extraction show them how a review is built from the inside, and if they also help revise and approve the paper, they can earn authorship (see who should be an author).

A friend at another college or in another country: Everything happens online, and Rayyan's free plan lets you invite two reviewers (as of October 2026). If you're an IMG preparing for exams or working abroad, your second reviewer can be a batchmate back home.

A swap with someone running their own review: Be their second reviewer and they'll be yours. Our Accelerator cohort is built this way: students pair up as each other's second reviewer (ours).

A second reviewer catches slips; a checker catches wrong decisions, the kind a mentor would spot. Our publishing guide covers the general routes to a mentor. For a systematic review, three small, specific checks do most of that job:

• A librarian for your search strategy: Cochrane strongly recommends this check, ideally with PRESS, a short evidence-based checklist for peer reviewing searches (section 4.4.8). Ask your college librarian first.
• A clinician for your eligibility criteria: "Could you look at my inclusion criteria for 15 minutes?" gets a yes far more often than "Will you be my mentor?" Bring one page.
• Someone who has published a meta-analysis: A senior resident or a biostatistics teacher can tell you in one sitting whether your plan to pool makes sense.

Does a systematic review count as original research?

By method, no: a systematic review is secondary research, because it analyses studies that other people ran instead of collecting new data. Under the rules that score your applications, often yes: official rules in the UK and India count a systematic review or meta-analysis as original research or as a full research publication, and some journals publish it as an original article.

Primary research collects new data from patients, samples or records; secondary research analyses data that already exist, usually published studies. A meta-analysis is the statistical part many systematic reviews contain: it combines the results of several studies into one estimate. Here's how three rule-books treat them, as of October 2026:

• UK specialty training: Internal Medicine Training's application scoring defines original research to include systematic reviews and meta-analyses, so a first-author review cited in PubMed can earn the top 6 publication points; other reviews, editorials and case reports score less (IMT application scoring).
• Faculty posts in India: the National Medical Commission's 2025 faculty regulations define a research publication as original papers, meta-analyses, systematic reviews and case series in journals indexed in databases such as Medline, PubMed Central, Scopus or DOAJ (NMC, June 2025).
• Journals: JAMA publishes systematic reviews with a meta-analysis as Original Investigations, and those without one as Reviews (JAMA instructions for authors).

For the US Match, research goes in the Scholarly Work section of the ERAS application; our publishing guide covers what counts there. And one thing to get right everywhere: a systematic review isn't a "review article". A narrative review summarises the literature without a registered protocol, a reproducible search or a formal judgement of each study's risk of bias, meaning how likely its design or conduct pushed its result away from the truth. Call yours a systematic review only if you did all of those.

What are the steps of a systematic review, and how long does each take?

We break a systematic review into nine steps, from checking your topic to submitting the paper. In our guided method, the hands-on work from question to submission-ready manuscript comes to about 100 hours, at 4 to 5 hours a week (ours). Published figures are far bigger, because they count total project hours on reviews of every size, and calendar time includes a lot of waiting.

Two published benchmarks are worth knowing. Meta-analyses done by a specialist company took a median of 1,110 hours each (216 to 2,518), with the biggest share spent searching, retrieving and organising papers (Allen and Olkin, 1999, as reported by Borah and colleagues). And 195 completed PROSPERO-registered reviews took 67.3 weeks on average from registration to publication, with searches finding anywhere from 27 to 92,020 records (Borah et al., BMJ Open 2017).

Here's every step: what you do, how long it takes and the mistake we see most often. Published times come from a 2021 scoping review of 34 studies on review workload (Nussbaumer-Streit et al., 2021) unless the cell says otherwise; the fourth column shows the hours our guided method budgets for each step (ours).

Scroll the table sideways to see every column →

Scope and experience explain much of the gap. Screening 1,000 records at one a minute takes about 17 hours per reviewer, and 10,000 take about 170, so a tightly framed question is the biggest time-saver there is. The same scoping review names inexperience and unfamiliar software among the reasons steps take longer, and blocking out time as one thing that speeds them up.

The table shows the classic mistake at each step. These five cut across all of them:

How do you find a systematic review topic that hasn't already been done?

Search before you commit: check PROSPERO for registered reviews still in progress, PubMed, the Cochrane Library and Epistemonikos for published ones, then count the new studies that have appeared since the newest good review. It takes a few hours and can save you months.

The reason is volume. Between 1991 and 2014, yearly publication of systematic reviews rose 2,728% while PubMed as a whole grew 153%, and some topics carry more than 20 overlapping meta-analyses (Ioannidis, Milbank Quarterly 2016). PROSPERO, the international register of planned systematic reviews, now holds more than 516,000 records (PROSPERO, October 2026).

Here's the check we use, in this order:

1. Write the question as a PICO
2. Search PROSPERO
3. Search PubMed with the systematic review filter
4. Search the Cochrane Library and Epistemonikos
5. Count the new primary studies

Write the question as a PICO: population, intervention, comparison and outcome. Our publishing guide shows how to narrow it, and the PICO words become your search terms. If you're an IMG aiming at one specialty, keep the question inside it, so you can discuss it at interview.

Search PROSPERO: Enter two or three core terms on its search page and filter for ongoing reviews. One registered in the last two years is the most likely to beat you to it. PROSPERO also flags similar reviews when you register, and asks why yours is needed (PROSPERO).

Search PubMed with the systematic review filter: Add AND systematic[sb] to your search or tick the Systematic Review filter, which catches more than the publication-type tag alone (PubMed User Guide). Sort by most recent and read every title from the last three years.

Search the Cochrane Library and Epistemonikos: A Cochrane review on your question sets a high bar, and its abstract is free to read (Cochrane). Epistemonikos describes itself as the largest source of systematic reviews for health decision-making.

Count the new primary studies: Note the search date of the newest good review, then look for trials published or completed since, in PubMed and on ClinicalTrials.gov.

Then decide. These thresholds are our rule of thumb, not a published standard:

• Go: no good review in the last three years, nothing matching registered as ongoing, and enough primary studies to say something.
• Narrow or shift: a recent review leaves out something you can name, such as a population, comparator or outcome; your review fills that gap, and your PROSPERO record says so.
• Update: the newest good review is several years old, and several new trials have appeared since its search date.
• Drop: several recent reviews plus ongoing registrations, or too few primary studies to answer anything.

If you don't have a question yet, these are the places good ideas usually come from:

• The wards: a "why do we do this?" question that nobody could answer with a citation.
• The end of recent reviews and guidelines: the research gaps their authors name for you.
• Older reviews that new trials have overtaken: candidates for an update.

Do you need ethics committee approval or PROSPERO registration first?

Register first, before you screen or extract anything. PROSPERO won't accept a review that has already started data extraction, and in India, ICMR's February 2024 addendum to its ethics guidelines says a systematic review or meta-analysis that follows standard guidelines and is prospectively registered "does not require submission to the ethics committee for clearance".

PROSPERO is a free international register run by the University of York's Centre for Reviews and Dissemination. Registering publishes the key details of your plan (question, eligibility criteria, outcomes and analysis) before you start, so nobody can quietly change it later, and you can attach your full protocol, best written to the PRISMA-P 2015 checklist. Registration also counts when your review is judged: AMSTAR 2 treats a protocol registered in advance as a critical domain (Shea et al., 2017). Before you register, know these three things:

• What PROSPERO accepts
• How long registration takes
• What ethics approval you need

What PROSPERO accepts: systematic reviews with outcomes directly relevant to human health. It turns away reviews that have started data extraction, scoping reviews (for now), literature reviews that use a systematic search but no other systematic review methods, and a few other types listed on its eligibility page. A scoping review can go on a free general registry such as OSF Registries instead. Course exercises belong in PROSPERO's unpublished student project option; a real review you intend to publish goes on the live register.

How long registration takes: Since PROSPERO's new system went live on 24 February 2025, there's no editorial queue. Once the guarantor (usually the most senior member of the team) approves the record and every named team member confirms their participation, PROSPERO assigns the number and publishes the record straight away (PROSPERO; FAQs). So the wait is your own team: name your second reviewer before you submit, and chase the confirmations the same day.

What ethics approval you need: In India, ICMR's 2017 National Ethical Guidelines expected every study, systematic reviews included, to go to an ethics committee for review or a formal exemption. The February 2024 addendum quoted above changed that for reviews that follow standard guidelines and are registered in advance, and its list of registries includes PROSPERO, OSF Registries, INPLASY and Research Registry. If your college or a journal still asks for an ethics statement, show them the addendum and your registration number. If your project uses patient records from your own hospital, it's an original study, not a review, and it must go to your ethics committee. Outside India, your institution sets the rule; our publishing guide covers the general case under registration and ethics.

Is 7 papers enough for a systematic review or a meta-analysis?

Yes, seven studies can be enough for both. There's no minimum number of studies for a systematic review, and the Cochrane Handbook (version 6.5, chapter 10, updated November 2024) defines a meta-analysis as the statistical combination of results from two or more separate studies; what a small number limits is how much you can say about why the results differ.

Seven is also normal: in a random sample of 300 systematic reviews indexed in 2020, only 28% contained a meta-analysis of 10 or more studies (Handbook, section 13.3.4.4). Whether to pool depends on whether the studies ask the same question closely enough for an average to mean something, not on the count; Cochrane's meta-analysis chapter opens by warning against rushing to the statistics (section 10.1).

With few studies, four things change:

• Heterogeneity is hard to measure
• Random-effects confidence intervals get unreliable
• Funnel-plot tests stop working
• Subgroup analyses run out of power

Heterogeneity is hard to measure: Heterogeneity is how much the true effects differ from study to study, and I² is the share of the variation in results that comes from those differences rather than chance. With few studies, I² and tau² (the estimated between-study variance) are very uncertain, so Cochrane advises against diagnosing heterogeneity with simple cut-offs such as "above 50% is high" (section 10.10.2). Report a confidence interval for tau² instead; Cochrane notes it starts to be informative from about five studies (section 10.10.4.4).

Random-effects confidence intervals get unreliable: A random-effects meta-analysis assumes each study estimates its own, related effect and averages them; a fixed-effect analysis assumes one true effect shared by all. With few studies, the standard (Wald-type) random-effects interval can be too narrow, and the Hartung-Knapp-Sidik-Jonkman (HKSJ) adjustment that widens it can overshoot (section 10.10.4.5). Part of the reason is plain arithmetic: HKSJ multiplies by a t-value with one fewer degree of freedom than the number of studies, where the standard method uses 1.96. With seven studies that's 2.45, about 25% bigger; with three, 4.30, more than double; with two, 12.71, which is why the Handbook's own two-study example ends with an excessively wide interval. Our advice: run both, and report both if they disagree. A fixed-effect method gives a steadier average with few studies, at the cost of ignoring heterogeneity (section 10.10.4.2).

Funnel-plot tests stop working: A funnel plot checks whether small studies show bigger effects than large ones, a possible sign of missing studies. Cochrane's rule of thumb is to test for funnel-plot asymmetry only with at least 10 studies, because below that the tests have little power (section 13.3.4.4). With seven, say so instead of running the test.

Subgroup analyses run out of power: Cochrane says meta-regression, which tests whether a study characteristic explains differences in results, should generally not be considered with fewer than ten studies (section 10.11.4). Treat any subgroup difference across seven studies as a hypothesis, not a finding.

Knowing when to pool is a judgement you build with examples, and our course has a lesson on exactly that. You can start free with Phase 0.

Can you do a systematic review without a meta-analysis?

Yes. A systematic review doesn't need a meta-analysis, and Cochrane says so directly (section 10.10.3). When the studies can't sensibly be combined, you synthesise them with a different, stated method and report it with SWiM, the Synthesis Without Meta-analysis reporting guideline.

Cochrane's legitimate reasons for not pooling include limited evidence, results that are incompletely reported or use different effect measures, and bias in the evidence. Too much clinical or methodological diversity is another commonly cited reason, though Cochrane suggests a meta-analysis can still be worth considering then (section 12.1). If you don't pool, these are the methods Cochrane accepts:

• Summarising effect estimates: the median, interquartile range and range of the effects across studies, for when you have the effects but not their variances.
• Combining P values: this tells you whether there's evidence of an effect in at least one study, and nothing about its size.
• Vote counting by direction of effect: counting how many studies point towards benefit and how many towards harm, ignoring statistical significance.

Vote counting by statistical significance, where you tally the "significant" studies, is the one to avoid: Cochrane calls it unacceptable (section 12.2.2). Cochrane also wants the actual method named, not just the label "narrative synthesis", with the results always set out in tables or plots (section 12.2). The SWiM guideline turns that into nine reporting items, used alongside PRISMA 2020 (Campbell et al., BMJ 2020). It's still a complete systematic review; it just tells readers less about the size of the effect.

Which free tools do you need for a systematic review?

You can run a whole systematic review on free tools: PubMed and Europe PMC to search, Rayyan to screen, a spreadsheet to extract, PRISMA's templates for the flow diagram and R to analyse. Paid tools such as Covidence, Embase or RevMan Web are useful, not essential.

Here's the free kit, with prices checked on 25 October 2026:

• Searching: PubMed plus at least one more source, because Cochrane doesn't consider a search of MEDLINE alone adequate (section 4.2.2). Use Embase or Scopus if your college library subscribes; if not, add Europe PMC. Also search the two trial registers Cochrane rates most important, ClinicalTrials.gov and the WHO ICTRP portal (section 4.3.3).
• Screening: Rayyan's free plan gives you three active reviews and two free reviewers. Covidence costs $339 a year for one review, with a free trial limited to 500 records, unless your university has a licence.
• Extraction: a spreadsheet form, piloted on two or three studies before you start.
• Flow diagram: the PRISMA 2020 flow diagram shows how many records you found, screened, excluded and included, and PRISMA offers free Word templates and an online app for it. Rayyan keeps its own PRISMA diagrams for paid plans.
• Analysis: R, free, with the meta or metafor packages. RevMan Web is free for Cochrane reviews and in Research4Life countries; otherwise it's a yearly subscription, £80 on the student rate, plus VAT in the UK and EU (Cochrane). We compare the two in Is RevMan free? RevMan vs R.
• Reporting: the 27-item PRISMA 2020 checklist (Page et al., 2021) and, if you don't pool, SWiM.

If you're in India, also download ICMR's free Beginner's Guide for Systematic Reviews (2022). It walks through the whole method and the software options; it predates the 2024 ethics addendum and today's tool prices, which is what this page adds.

Can you use ChatGPT or other AI tools for a systematic review?

Yes, for parts of the work, as long as a person checks everything the tool produces and you disclose how you used it. An AI tool can't be one of the two independent reviewers, and it can't be an author.

Three sources set the rules, and they agree:

• ICMJE: the International Committee of Medical Journal Editors (recommendations updated January 2026) says journals should ask at submission whether and how you used AI; chatbots can't be authors, because they can't take responsibility for the work; and undisclosed AI use can be treated as misconduct (ICMJE).
• Cochrane, Campbell, JBI and CEE: in a joint 2025 statement, these four evidence-synthesis organisations agreed that authors stay responsible for any AI they use, that it needs human oversight, and that AI use that makes or suggests judgements must be fully reported (Flemyng et al., 2025).
• PRISMA 2020: items 8 and 9 ask for details of any automation tools you used in screening and extraction.

In practice, reasonable uses include suggesting search synonyms (check each against PubMed's subject headings), explaining an R error message and polishing your English, all disclosed. Letting a tool make inclusion decisions on its own, trusting numbers it extracted without checking the paper, or having it draft the paper are not.

Who should be an author on a systematic review, and in what order?

Everyone who meets all four ICMJE authorship criteria, and nobody who doesn't. The order is for the authors to agree together, and the safest time to agree it is in writing, before screening starts.

For a systematic review, two cases come up again and again:

• Your second reviewer: Screening and extraction count as acquiring data, which is part of the first ICMJE criterion. To be an author, they also need to help draft or critically revise the paper, approve the final version and agree to be accountable for it. Someone who only screened for a weekend belongs in the acknowledgements.
• Your checkers: A librarian who reviewed your search or a faculty member who gave general advice goes in the acknowledgements unless they meet all four criteria; ICMJE says general supervision alone doesn't qualify (ICMJE).

The full criteria, and the paid-authorship offers to walk away from, are in our publishing guide.

Frequently asked questions