Session 2 · Find & read: the evidence

Foundations of Clinical Research: From Clinical Question to Publication

Vương Mỹ Lượng & Bernard Osang’ir

Neudata Consulting Ltd

Session 2 · Find & read: the evidence

Part 1

Week 2 · Saturday

• Recap quiz on Session 1
• 2.1 Finding the evidence: PICO to search, Boolean, MeSH, live PubMed demo
• 2.2 Managing and citing: Zotero demo, citation styles, paraphrasing, grey literature
• Break
• 2.3 Reading a paper in 15 minutes and 2.4 from literature to introduction
• Proposal builder 2: your own search, 5 references, problem statement, working title

SESSION 2 OPENING (0:00, 19:30).

Open with what the Session 1 exit question showed (for example, how many people still used vague verbs) and spend one minute on it if needed.

Teaching point: last Saturday each participant started their own question. Today we find out what is already known about it, keep the references safely, and read papers efficiently.

Pace: slower than it looks. Every block has a demo or a task. Use the ‘Check your understanding’ slides: they show you who is lost.

Practical check before starting (chat): everyone has their PICO question, their synonym list (homework) and Zotero installed. Paste two links in the chat: the Session 2 exercise sheet (PDF) and Paper 1 as a clean PDF (the unannotated original, Course/Papers/Paper 1 - Boulet 2024 Crit Care (RCT) - original.pdf, also linked by DOI on the course website page ‘The two papers’).

Timing: 1 minute, then the recap quiz.

Recap quiz: Session 1

QUICK QUIZ · 8 min

• Which course paper is a trial and which is a cohort: Paper 1 (Boulet) or Paper 2 (Hyun)?
• ‘In adults with sepsis, is a positive fluid balance by day 3, vs not, linked to death by day 28?’ Name P, E, C, O, T
• Which FINER letter did our running example fail?
• Which is an objective: ‘to understand fluid balance’ or ‘to estimate the proportion fluid-positive on day 3’?
• Write H0 for: does a restrictive protocol reduce fluid balance at day 5?

OUTPUT Answers in the online quiz form (link in the chat), checked against the next slide

RECAP QUIZ (0:00-0:10, 19:30-19:40). Individual, in a short online form (or answers typed in the chat), 6 minutes, then 2 minutes to show the form results on screen. Answers on the next slide. The questions are also on the Session 2 exercise sheet.

This also re-activates last Saturday’s vocabulary for today’s search.

Latecomers: let them join at question 2.

Timing: 8 minutes.

Recap quiz: answers

Designs

• Paper 1 is a trial (RCT)
• Paper 2 is a cohort

PECO(T)

• P: adults with sepsis
• E: positive balance by day 3
• C: neutral or negative balance
• O and T: death by day 28

Feasible

• Mortality needs more than 1,000 patients

Estimate

• ‘Estimate’ is measurable
• ‘Understand’ is not

H0

• Fluid balance at day 5 is the same in both groups

KEY MESSAGE If you got 4 or 5 right, you are ready for today.

Go through the five answers in 2 minutes. Quick poll: how many did you get right (0-3, 4 or 5)?

If many missed question 4 or 5, spend one extra minute: objectives use measurable verbs, and H0 always says ‘no difference’.

Timing: 2 minutes. Segment 2.1 starts at 0:10.

2.1 Finding the evidence

Before you study it, find out what is already known.

40 minutes: the search funnel, PICO to search blocks, Boolean, MeSH, search tools, checking for a systematic review and a live PubMed demo

SEGMENT 2.1 (0:10-0:50, 19:40-20:20).

Teaching point: the literature search decides whether the question is novel, gives the background for the problem statement, and shows how others measured the outcome.

Ask the class (chat): ‘What do you usually do when you need evidence quickly?’ Honest answers: Google, a colleague, UpToDate, a guideline. All fine for care, but a proposal needs a documented search.

Timing: 1 minute.

The search funnel

Write down each step: the date, the search string, the filters and the number of hits.

All of PubMed

• More than 40 million records
• As of October 2026

One keyword

• septic shock
• Tens of thousands

PICO blocks

• MeSH and free text, AND and OR
• A few thousand

Filters

• RCT, review, adults, years
• Hundreds

Screen

• Read titles and abstracts
• Dozens

Keep

• 5-10 key papers

KEY MESSAGE A good search starts wide, narrows step by step, and is written down.

Walk along the funnel from left to right: all of PubMed, one keyword, structured PICO blocks, filters, screening by people, a handful of key papers.

Source for the PubMed size: the PubMed ‘About’ page (https://pubmed.ncbi.nlm.nih.gov/about/) says ‘more than 40 million citations and abstracts’. Checked on 19 October 2026 (page last updated 11 March 2025). Check it again before teaching and update the ‘as of’ date if the number has changed.

The other numbers are orders of magnitude only. They change every day.

Teaching point: the last two steps are done by reading, not by the computer.

Ask the class (chat): ‘Why write down the date and the search string?’ Expected: so it can be repeated. Journals and reviewers ask for it.

This slide is built from editable shapes: change a number or add a reference directly in PowerPoint.

Timing: 2 minutes.

From PICO to search blocks

OR inside a block, AND between blocks: (P synonyms) AND (I synonyms) AND (O synonyms). C (usual care) is usually not searched.

P: septic shock

• “Shock, Septic”[MeSH]
• OR “septic shock”[tiab]
• OR sepsis[tiab]

I: restrictive fluids

• “Fluid Therapy”[MeSH]
• OR “fluid restrict*”[tiab]
• OR “restrictive fluid*”[tiab]
• OR “conservative fluid*”[tiab]

O: mortality

• Mortality[MeSH]
• OR mortality[tiab]
• OR death[tiab]

KEY MESSAGE One block per PICO concept: OR inside a block, AND between blocks.

Turn each PICO element into a concept block: the MeSH term plus free-text words.

[tiab] = title and abstract. The asterisk (*) is truncation: restrictive fluid* also finds ‘restrictive fluids’.

The comparison (usual care) is normally left out because the words vary too much.

Misconception: ‘just type the whole question into PubMed’. You lose control of synonyms and logic.

Ask the class (chat): ‘Which other words would authors use for restrictive?’ Expected: conservative, limited, de-resuscitation, fluid removal, negative fluid balance.

Timing: 4 minutes.

Another example: a ward question

Question: in adults at risk on the ward, does 2-hourly repositioning reduce new pressure injuries?

P + O: pressure injury

• “Pressure Ulcer”[Mesh]
• OR “pressure injur*”[tiab]
• OR “pressure ulcer*”[tiab]
• OR bedsore*[tiab]

I: repositioning

• “Patient Positioning”[Mesh]
• OR reposition*[tiab]
• OR turning[tiab]

Join the blocks with AND

• Tip: when P and O are the same idea (pressure injury), one block can carry both.
• Authors write ulcer, injury, bedsore, decubitus. MeSH collects them under “Pressure Ulcer”.

KEY MESSAGE Same method for any topic: concepts in blocks, synonyms with OR, blocks with AND.

Another worked example, from the ward, so colleagues outside the ICU see that the method works for any topic.

Question: in adults at risk on the ward, does 2-hourly repositioning reduce new pressure injuries?

Here P and O share the same idea (pressure injury), so one block carries both. The second block is the intervention.

MeSH: authors write pressure ulcer, pressure injury, bedsore, decubitus. The MeSH heading ‘Pressure Ulcer’ collects them.

Ask the class (chat): ‘Which other words would authors use for repositioning?’ Expected: turning, position change, mobilisation, offloading.

Timing: 3 minutes.

Boolean logic in pictures

KEY MESSAGE AND narrows, OR widens, NOT removes. Use NOT sparingly.

AND: only papers with both concepts, so fewer hits. OR: either word, so more hits. Use OR for synonyms.

NOT removes every paper that mentions the second term, even good papers that mention neonates once. That is why NOT is risky.

Tip: type the operators in capitals, and use brackets to control the order.

Ask the class (chat): ‘septic shock AND fluids vs septic shock OR fluids: which gives more hits?’ Expected: OR.

Timing: 2 minutes.

Check your understanding: AND, OR or NOT?

QUICK QUIZ · 2 min

• You want papers that mention both septic shock and fluid restriction: which operator?
• You want papers using either ‘restrictive’ or ‘conservative’ fluids: which operator?
• Your search found 40,000 hits. Will adding a third AND block give more or fewer?
• Why is ‘sepsis NOT children’ risky?

OUTPUT Answers in the chat (or the poll)

Answers: 1 AND. 2 OR. 3 Fewer (AND narrows). 4 It removes every paper that mentions children anywhere, including good adult studies that mention children once.

Timing: 2 minutes.

MeSH: the shared label behind many words

KEY MESSAGE Search the MeSH term AND the free-text words, joined with OR.

MeSH (Medical Subject Headings) is the controlled vocabulary of the US National Library of Medicine. Each MEDLINE record is tagged with MeSH terms, whatever words the authors used.

Searching a MeSH term includes the narrower terms below it (‘explode’). For example, Sepsis[MeSH] includes Shock, Septic.

The newest records may not be indexed yet, so always add free-text [tiab] words.

Ask the class (chat): ‘Why does MeSH help a nurse searching for pressure injury?’ Expected: authors write pressure ulcer, bedsore, decubitus ulcer, and MeSH collects them under one heading.

Next slide: how to find the right MeSH term.

Timing: 2 minutes.

How to find the right MeSH term

MeSH Database

• PubMed home page, under Explore
• Type a word, e.g. bedsore
• It shows the heading: Pressure Ulcer

A paper you trust

• Open a relevant paper in PubMed
• Scroll down to ‘MeSH terms’
• Reuse the ones that fit

Search details

• PubMed maps your words to MeSH
• Check which heading it used

KEY MESSAGE Look up MeSH terms, do not guess them.

Participants cannot know which words are MeSH terms. Show them three ways to find out.

Route 1, the MeSH Database: on the PubMed home page, under Explore, click ‘MeSH Database’ (or go to https://www.ncbi.nlm.nih.gov/mesh). Type an everyday word. The record shows the official heading, its definition, the entry terms it collects (for Pressure Ulcer: bedsore, decubitus ulcer, pressure injury and others) and its place in the tree. ‘Add to search builder’ sends it to PubMed.

Route 2, a known paper: open a paper you already know is relevant (for us, CLASSIC or Paper 1), scroll down to ‘MeSH terms’ and reuse the headings that match your concepts. The demo shows this.

Route 3, Search details: run your free-text words, then open ‘Search details’ on the Advanced page. PubMed’s automatic term mapping shows which MeSH heading it matched.

Check the record before you use a heading: sometimes the heading is broader or narrower than you expect.

Timing: 2 minutes.

Four tools that sharpen a search

“Quotes”

• “septic shock” = the exact phrase
• Not two loose words

Truncation *

• child* finds child, children, childhood
• Use it when one idea has many word endings

Field tags

• [tiab] title or abstract
• [mh] MeSH term
• [pt] article type

Filters

• Article type (RCT, systematic review)
• Date, age, humans

KEY MESSAGE Check ‘Search details’ to see exactly what PubMed searched.

Quotes keep a phrase together.

Truncation (the asterisk) helps when authors write the same idea with different endings. Clinical example: for a paediatric question, child* finds child, children, childhood and childhood-onset in one go, instead of typing each word with OR. Other useful stems: nurs* (nurse, nurses, nursing), restrict* (restricted, restriction, restrictive).

Cautions: child* also finds childbirth, and too short a stem (flu*) finds influenza and fluoride. In PubMed the stem needs at least four letters, and a truncated word is not mapped to MeSH, so keep the MeSH term in the block as well.

Field tags restrict where PubMed looks. Filters (left of the results) narrow by article type, date and age. Each filter needs a reason you can write down.

Misconception: ‘filter to free full text’. That throws away key papers.

Timing: 3 minutes.

PubMed vs Google Scholar

PubMed	Google Scholar
What it covers	Biomedical journals, indexed by experts
MeSH (controlled vocabulary)	Yes
Filters (design, age, date)	Yes, many
Transparent, repeatable	Yes
“Cited by” a key paper	Partial
Best for	Your main, repeatable search

KEY MESSAGE PubMed for the main search. Google Scholar to check and to follow citations.

PubMed: free, run by the US National Library of Medicine, biomedical journals, MeSH, many filters, a transparent search you can repeat.

Google Scholar: very broad (theses, books, preprints), good for grey literature and ‘Cited by’, but hidden ranking, few filters, hard to repeat.

PubMed is not the only database. Embase, the Cochrane Library, Web of Science, Scopus and CINAHL (for nursing) are the other standard sources. A systematic review searches several of them (extra slide ‘Beyond PubMed’). For a Level 1 proposal, PubMed plus a trial registry is enough.

Ask the class (chat): ‘A colleague found a perfect paper in Google Scholar. What should you check?’ Expected: peer-reviewed or preprint? Legitimate journal? (Predatory journals: Session 8.)

Timing: 2 minutes.

First check: is there already a systematic review?

Search for reviews

• Cochrane Library
• PubMed filter: Systematic Review

Reviews in progress

• PROSPERO lists registered systematic reviews

Then decide

• A recent, good review: build on it
• None, or out of date: that is a gap

Learn to do one

• Level 3, Module A: systematic review and meta-analysis

KEY MESSAGE Before you plan a study, find out what has been done, starting with systematic reviews.

Teaching point: before starting any research, get an overview of what is already known about your question. The best overview is a systematic review: one you find, or one you do yourself.

Where to look: the Cochrane Library (Cochrane systematic reviews, mostly on treatments), PubMed with the article-type filter ‘Systematic Review’, and PROSPERO (https://www.crd.york.ac.uk/prospero/), the international register of systematic reviews that are planned or under way.

If a recent, good review answers your question, do not repeat it: build on it and move your question to the gap that remains. If there is none, or it is out of date, that gap is part of your justification.

Doing a systematic review and meta-analysis yourself is a course of its own: Level 3 has a dedicated module (Module A: systematic review and meta-analysis).

Ask the class (chat): ‘Has anyone started a project and then found it had already been done?’

Timing: 2 minutes.

Live demo: searching PubMed for our question

LIVE DEMO · 10 min

• Search one keyword, septic shock, and note the number of hits
• Paste the full PICO string: P, I and O blocks joined with AND
• Look at ‘Search details’: what did PubMed really search?
• Add filters: RCT and systematic review, adults, last 10 years
• Open one result: abstract, MeSH terms, ‘Similar articles’, ‘Cited by’
• Click ‘Send to’, then ‘Citation manager’ (we import it in 2.2)

Watch for

• How the hits fall at each step
• Which papers keep appearing
• Can you find Paper 1 and the CLASSIC trial (Meyhoff 2022)?

OUTPUT A search string we can all copy (Exercise 1 on the sheet)

Follow Course/Demos/s2_demo.md, Part A (PubMed): exact strings, where to click, and the offline fallback.

Do not promise hit counts. Expect tens of thousands for ‘septic shock’ and a few thousand for the full PICO string. The numbers change every day.

CLASSIC (first time we name it): a large randomised trial, Meyhoff et al., N Engl J Med 2022, ‘Restriction of Intravenous Fluid in ICU Patients with Septic Shock’. The name stands for Conservative versus Liberal Approach to Fluid Therapy of Septic Shock in Intensive Care. In 1,554 adults with septic shock in European ICUs, restricting IV fluid did not change 90-day mortality. It is the main trial behind our running example, so say this when it appears in the results.

Share your screen. Narrate slowly and pause after each step so participants can note the number of hits on their sheet. Invite synonyms for the blocks in the chat and add one or two live.

If your internet fails, switch to the screenshots (the second trainer can share them). Do not troubleshoot live.

Timing: 10 minutes.

Your turn: build a Boolean search

INDIVIDUAL TASK · 5 min

• Take your own PICO question and your synonym list (Session 1 homework)
• Exercise 1: write 2-4 synonyms for P, for I (or E) and for O
• Join synonyms with OR in brackets, and the blocks with AND
• Check which words are MeSH terms: MeSH Database or a known paper
• We call on 2-3 people to read their string aloud, microphone on

OUTPUT One search string for your own question, to run in Proposal builder 2

Individual: participants write their string on the exercise sheet or in a text file. They do not search yet.

After 4 minutes, call on 2 or 3 participants by name (tell the class at the start that you will do this). Each unmutes and reads one block or the whole string aloud. Type it on your shared screen as they read, and ask the class for one missing synonym.

Listen for: missing brackets, AND and OR mixed inside one block, the comparison included as a block, NOT used without a reason.

Fast finishers: add one filter and write why.

Timing: 5 minutes. Segment 2.1 ends at 0:50.

2.2 Managing and citing

Keep every reference you find, and use other people’s work honestly.

20 minutes: reference managers (Zotero demo), citation styles, paraphrasing safely, grey literature

SEGMENT 2.2 (0:50-1:10, 20:20-20:40).

Teaching point: two habits protect every proposal and paper: save references in a reference manager from day one, and never copy wording from a source.

Timing: 30 seconds.

Reference managers: save as you go

Collect

• Tick papers in PubMed
• Clipboard or My NCBI Collections

Export

• ‘Send to’, then ‘Citation manager’
• Gives an .nbib file

Import

• Zotero or Mendeley (free)
• EndNote (paid)

Cite

• Insert citations in Word with the plug-in

List

• Reference list built for you
• Any style

KEY MESSAGE A reference manager prevents the commonest error in proposals: wrong references.

Teaching point: never type references by hand. A reference manager stores each paper once and formats it in any journal style.

Zotero (free, open source) and Mendeley (free) work with Word. EndNote is common in universities but paid. In this course everyone uses Zotero (installed as Session 1 homework).

Browser connectors (e.g. Zotero Connector) save a paper from the journal page in one click, often with the PDF.

Misconception: ‘I will sort the references at the end’. Saving as you go takes seconds, but rebuilding a list takes hours and causes errors.

Timing: 2 minutes.

Live demo: from PubMed to Zotero to Word

LIVE DEMO · 5 min

• Import the .nbib file from the PubMed demo into Zotero
• Look: authors, title, journal, year, DOI, with nothing typed by hand
• In Word, on the Zotero tab, click ‘Add/Edit Citation’ and type Meyhoff
• Click ‘Add/Edit Bibliography’: the reference list appears
• Open ‘Document Preferences’: switch Vancouver to APA and back

Watch for

• [1] becomes (Meyhoff et al., 2022)
• The list reformats itself

OUTPUT Everyone has seen the full cycle: collect, cite, list

Instructor script: Course/Demos/s2_demo.md, Part B (Step 6), with what to click, what to say and what to do if Zotero fails. The same script is Course/Demos/s2_demo.pdf and the ‘Session 2 demo’ page of the course website.

The words in quotation marks on the slide are Zotero’s own button names in Word.

Preparation: Zotero desktop and its Word plug-in installed, an empty test library, a blank Word document.

Everyone installed Zotero as Session 1 homework, so participants can repeat each step on their own computer. Anyone whose installation fails watches now and fixes it after the session with the second trainer.

If Zotero fails, show the screenshots and say ‘the same file works in Mendeley and EndNote’.

Timing: 5 minutes.

Citation styles: same paper, two looks

Use the style your target journal asks for. The reference manager switches in one click.

Vancouver

• Restricting IV fluid did not change 90-day mortality [1].
• Reference list:
• Author AA, Author BB, et al. Title. J Abbrev. Year;Vol(Issue):pages.
• Numbered in the order cited. Used by most medical journals.

APA

• Restricting IV fluid did not change 90-day mortality (Meyhoff et al., 2022).
• Reference list:
• Author, A. A., & Author, B. B. (Year). Title. Journal, Vol(Issue), pages.
• Author and year in the text. Common in nursing and social science.

KEY MESSAGE Pick the style your target journal uses, and let the software do the formatting.

Vancouver (numbered, in the order cited) is used by most medical journals and follows the ICMJE recommendations. APA (author and year) is common in nursing, psychology and social science.

Every journal states its style in the ‘Guide for authors’. A reference manager switches between thousands of styles in one click.

Teaching point: a citation lets the reader check your claim. Cite the original study for facts and numbers, not a review that quotes it, wherever you can.

Ask the class (chat): ‘Which style uses [1]?’ Expected: Vancouver.

Timing: 2 minutes.

Citing vs copying: paraphrase safely

Method: read, close the paper, write from memory, check facts, cite.

Source (Boulet et al. 2024, abstract)

• “In ICU patients with septic shock, an optimized restrictive fluid strategy targeting hidden fluid intakes did not reduce the overall fluid balance at day 5.”

Word swaps (“patchwriting”)

• “In septic shock ICU patients, an optimised restrictive approach targeting hidden fluid intakes did not lower the total fluid balance at day 5.”
• Same structure, same words = plagiarism, even with a reference

Own words + citation

• “A pilot trial in 48 patients found that limiting maintenance and drug-dilution fluids did not significantly change fluid balance after five days [1].”
• New structure, adds context, cited

KEY MESSAGE Change the structure as well as the words, and always cite the source.

Left: one sentence from the conclusion of Paper 1’s abstract (quoted with attribution).

Middle: ‘patchwriting’, the same sentence with a few words swapped. Even with a reference this is plagiarism, because the structure and most of the wording are copied. Similarity checkers used by journals detect it easily.

Right: a real paraphrase, with a new structure, extra context from the paper (pilot, 48 patients, which fluids), and a citation.

Method: read, close the paper, write from memory, check the facts, cite. If you need the exact words, use quotation marks and a citation. Quotes are rare in medical writing.

Misconception: ‘if I cite it, I can copy it’. Citation credits the idea, but it does not license copying the wording. AI rewording tools do not solve this either. You remain responsible (Session 7).

Ask the class (chat): ‘Is copying the Methods wording from your own earlier paper allowed?’ Expected: it is text recycling. Check the journal policy, and usually rewrite or cite.

Timing: 3 minutes. Exercise 2 on the sheet practises this at home.

Four rules for citing safely

Cite facts

• Numbers, findings and others’ ideas need a citation

Cite sources

• Read and cite the original paper
• Not a summary of it

Own words

• Use a new structure
• Exact words only in quotation marks

Check

• Every reference says what you claim it says

KEY MESSAGE Never cite a paper you have not read.

Common knowledge (‘sepsis is an infection-related emergency’) does not need a citation. Specific facts and numbers do.

Citing a paper you have only seen quoted elsewhere spreads errors. Check the original.

AI chat tools may invent references that look real. Every reference must be found in PubMed or on the journal site.

Timing: 1 minute.

Check your understanding: OK or not OK?

QUICK QUIZ · 2 min

• You copy a sentence from an abstract, change two words and add [3]
• You quote one sentence word for word, in quotation marks, with [3]
• You write a finding in your own words but forget the citation
• You cite a paper you found quoted in a review, without reading it

OUTPUT For each: a thumbs-up (OK) or thumbs-down (not OK) reaction, or the poll

Answers: 1 Not OK: patchwriting is plagiarism even with a citation. 2 OK (but rare in medical writing). 3 Not OK: someone else’s finding needs a citation. 4 Not OK: you may repeat an error, so read and cite the original.

Timing: 2 minutes.

Grey literature, including trial registries

KEY MESSAGE Look below the waterline: searching grey literature, including trial registries, reduces publication bias.

Journal articles in PubMed are only the tip of the iceberg. Below the waterline is the grey literature: everything not published in a journal. In this course it includes trial registries (as in the course glossary), together with conference abstracts, theses, reports, guidelines and preprints.

Why it matters: studies with negative or disappointing results are less likely to be published (publication bias), so the published literature can look more positive than the truth. Searching below the waterline reduces this bias in what you find.

What a trial registry is: a public database where a clinical trial is recorded before the first patient is enrolled. The record gives the question, the design, the planned primary outcome, the planned sample size and the status (recruiting, completed, results posted). Medical journals that follow the ICMJE recommendations only publish trials that were registered.

The main registries: ClinicalTrials.gov (run by the US National Library of Medicine) and the WHO International Clinical Trials Registry Platform (ICTRP), which searches many national registries at once. Systematic reviews are registered in the same way, in PROSPERO.

Use: compare the planned primary outcome with the published paper. A switched outcome is a warning sign. Registries also show trials that finished but never published their results.

Paper 1 is registered as NCT04947904, registered before the first patient was enrolled. A good homework habit: open the registry record and compare its planned primary outcome with the paper.

Preprints (e.g. medRxiv) are not peer reviewed. Read them with extra care.

Ask the class (chat): ‘Why check a registry before starting your own trial?’ Expected: someone may already be running the same study.

Timing: 3 minutes. Segment 2.2 ends at 1:10, then the break.

Break

15 min

BREAK (1:10-1:25, 20:40-20:55).

Say: stand up, stretch and rest your eyes. Before we restart, open Paper 1 (link in the chat).

During the break: post the link to Paper 1 as a clean PDF in the chat for the reading task (Course/Papers/Paper 1 - Boulet 2024 Crit Care (RCT) - original.pdf, or the DOI link on the course website page ‘The two papers’).

Say clearly in the chat what time you will begin again, and restart on time.

2.3 Reading a paper in 15 minutes

You do not have to read every word. Read the right words in the right order.

40 minutes: anatomy of a paper, reading order, the abstract, spin, and a reading task on Paper 1

SEGMENT 2.3 (1:25-2:05, 20:55-21:35).

Teaching point: busy clinicians need a reading routine. In 15 minutes you can decide whether a paper is relevant, what it found and whether to trust it.

Check in the chat that everyone has opened the clean abstract of Paper 1 before the task.

Timing: 30 seconds.

Anatomy of a paper: IMRaD

KEY MESSAGE Every original paper has the same shape: why, how, what, so what.

IMRaD = Introduction, Methods, Results and Discussion: the standard structure of an original research paper.

The title and the abstract sit on top of IMRaD. They are not one of its letters, which is why they come first in the picture. The small ‘a’ in IMRaD stands for ‘and’, not for ‘abstract’.

The hourglass: the Introduction narrows from the broad problem to the specific question. Methods and Results are narrow and factual. The Discussion widens again to what it means for practice.

Teaching point: knowing the shape tells you where to look. The question is at the end of the Introduction, the primary outcome is in the Methods, and the main number is in the Results (and the abstract).

Misconception: ‘the Discussion tells me what was found’. The Discussion is the authors’ interpretation. The Results are the evidence.

Ask the class (chat): ‘Where do you find who was included?’ Expected: Methods (eligibility) and Results (flow chart, Table 1).

Timing: 4 minutes. Sessions 7 and 8 teach how to write each section.

The 15-minute reading order

Stop after step 1 or 2 if it does not answer your question. Total: about 15 minutes.

1 Title

• 30 s
• Relevant to my question?

2 Abstract

• 3 min
• PICO, design, main result

3 Figures and tables

• 4 min
• Flow chart, Table 1, main result table

4 Methods

• 5 min
• Who, how chosen, primary outcome, bias protection

5 Discussion

• 2-3 min
• Limitations. Do conclusions match the data?

KEY MESSAGE Title, abstract, figures and tables, methods. Stop early if it is not relevant.

Do not read front to back. Step 1, title (30 seconds): is it about my question? Step 2, abstract (3 minutes): PICO, design, main result. Most papers are rejected here.

Step 3, figures and tables (4 minutes): the flow chart (who was lost), Table 1 (were groups similar), the main results table.

Step 4, methods (5 minutes): who, how chosen, the primary outcome, how bias was reduced. Step 5, discussion (2-3 minutes): limitations, and do the conclusions match the data?

Teaching point: you read the Introduction and full Discussion only for papers you will cite or appraise in depth.

Misconception: ‘reading only the abstract is enough’. Abstracts omit limitations and can contain spin (next slides).

Timing: 3 minutes.

What an abstract must contain

Section	What it must contain
Background	The problem and the aim (1-2 sentences)
Methods	Design, setting, participants, intervention/exposure, comparison, primary outcome
Results	Numbers screened and analysed, primary result as effect size + 95% CI (+ p), key harms
Conclusions	Answer to the primary question, no more
Registration	Trial registry number (for trials)

Usually 250-300 words

KEY MESSAGE A good abstract lets you write the PICO and the main result without opening the paper.

Most clinical journals require a structured abstract: Background, Methods, Results, Conclusions (plus the trial registration number for trials).

Methods must name the design, setting, participants, intervention or exposure, comparison and the primary outcome.

Results must give the numbers screened and analysed and the primary result as an effect size with its 95% confidence interval, not only a p-value or ‘significant’.

Conclusions answer the primary question, no more and no less. Reporting guidelines have abstract checklists (CONSORT for Abstracts, STROBE), covered in Session 8.

Ask the class (chat): ‘An abstract says only: mortality was significantly lower (p < 0.05). What is missing?’ Expected: the actual rates, the effect size and the confidence interval.

Timing: 3 minutes.

A results sentence: poor vs good

Illustrative numbers. Ask of every abstract: how big, how sure, by when?

Poor results sentence

• “Mortality was significantly lower in the intervention group (p < 0.05).”
• No numbers per group
• No effect size
• No 95% CI
• No time point

Good results sentence

• “28-day mortality was 30% with restriction vs 38% with usual care (risk difference -8 percentage points, 95% CI -15 to -1).”
• Numbers per group
• Effect size
• 95% CI
• Time point

KEY MESSAGE Every main result needs: numbers per group, an effect size, a 95% CI and a time point.

Worked example (illustrative numbers, not from a real trial).

Left: ‘significantly lower (p < 0.05)’ tells us nothing about how big the difference is, how sure we are, or when it was measured.

Right: rates in each group, the size of the difference (8 percentage points), the 95% confidence interval and the time point (28 days). A reader can judge whether the difference matters clinically.

Teaching point: ask ‘how big, how sure, by when?’ of every result you read. Confidence intervals are explained properly in Session 6.

Ask the class (poll, left or right): ‘Which sentence would you trust to change practice?’ Expected: the right one, because you can see the size and the uncertainty.

Timing: 3 minutes.

Spin: when words go beyond the data

What the data say	What the words say (spin)	Seen in
p-value	p = 0.42	“a non-significant trend”
Outcomes	Primary outcome: no difference	Headline = a positive secondary outcome
Design	Observational association	“Impact of…”, “reduces”
Comparison	No between-group test	“Patients improved” (before vs after only)

KEY MESSAGE Read the numbers first, then check whether the words match them.

Spin = reporting that makes results look more favourable than the data justify. It is common in abstracts of trials with non-significant primary outcomes.

Row 1: Paper 1’s Discussion calls p = 0.42 for the primary outcome a ‘non-significant trend’. A p of 0.42 is no evidence of a difference. The correct phrase is ‘no significant difference, and the confidence interval was wide’. To its credit, the conclusion of Paper 1’s abstract has no spin.

Row 2: highlighting a positive secondary outcome when the primary outcome showed nothing.

Row 3: causal words for observational data. Paper 2’s title says ‘Impact of…’ although a cohort can only show association (its abstract conclusion correctly says ‘associated’).

Row 4: claiming improvement from before-after change within one group, with no comparison between groups.

Misconception: ‘peer review removes spin’. It reduces it, but spin is frequent in published papers.

Timing: 4 minutes.

Check your understanding: spot the spin

QUICK QUIZ · 3 min

• A cohort study: ‘A positive fluid balance increases deaths in sepsis’
• Primary outcome p = 0.31: ‘a promising trend towards benefit’
• The primary outcome shows no difference, but the abstract headline says ‘improved patient satisfaction’
• ‘Mortality fell from 30% to 22% after our new fluid protocol’ (no control group)

OUTPUT For each, we call on 2-3 people, microphone on: what is the spin, and how would you rewrite it?

Take one item at a time. Give the class 30 seconds to think, then call on one or two participants by name to unmute and say where the spin is and how they would rewrite it. Ask the class whether anyone disagrees before you give the answer.

Answers: 1 Causal word for an observational design: write ‘was associated with higher mortality’. 2 p = 0.31 is no evidence of a difference: write ‘no significant difference, and the CI was wide’. 3 The headline is a secondary outcome: lead with the primary result. 4 Before-after change without a comparison, and other things may have changed: write ‘mortality was lower after implementation, but without a control group we cannot attribute this to the protocol’.

Timing: 3 minutes.

Your turn: read Paper 1’s abstract

INDIVIDUAL TASK · 12 min

• Open Paper 1 (Boulet 2024, Crit Care): the abstract is on page 1
• Fill the reading grid (Exercise 3): design, P, I vs C
• Primary outcome, numbers screened and analysed
• Main result with its 95% CI and p-value
• Does the conclusion match the result? Any spin?
• Type your primary outcome and main result in the chat

Clock

• 8 minutes reading alone
• 4 minutes checking in the chat

OUTPUT A completed abstract reading grid

Individual work, quietly for 8 minutes (cameras may go off), then 4 minutes in which everyone types the primary outcome and main result in the chat and you comment on the answers. Beginners need the extra time. Do not rush it.

Where the paper is: Course/Papers/Paper 1 - Boulet 2024 Crit Care (RCT) - original.pdf (the clean original, without our notes). It is open access (CC BY-NC-ND), so the unchanged original may be shared. The course website page ‘The two papers’ links to it by DOI (10.1186/s13054-024-05155-z), and you post the link in the chat at the break. The annotated versions (English and Vietnamese, also in Course/Papers) are for after the task.

Watch the chat. The hardest cells: the primary outcome (many write ‘mortality’, but it is cumulative fluid balance at day 5) and the main result (they should copy the difference with its 95% CI, -15.2 to 41.6, which crosses zero).

Fast finishers: read Paper 2’s abstract and spot the word in its title that overstates the design (‘Impact’).

Timing: 12 minutes. The answers are on the next slide.

Paper 1’s abstract: the answers

Abstract item	What Paper 1 reports
Design	Pilot RCT, single-blind (patients and statisticians), 3 ICUs
P	Adults 18-85 y, septic shock, first 24 h of vasopressors
I vs C	Restrict maintenance and drug-dilution fluids (7 days) vs standard
Primary outcome	Cumulative fluid balance at day 5 (ml/kg)
Numbers	1,201 screened, 50 randomised, 48 analysed (24 per group)
Main result	Median 6.9 vs 35.0 ml/kg. Difference 13.2 (95% CI -15.2 to 41.6), p = 0.42
Conclusion	Did not reduce fluid balance at day 5 (matches the result)
Registration	ClinicalTrials.gov NCT04947904

KEY MESSAGE Primary outcome: fluid balance, not mortality. Read what the study actually measured.

Go through the grid quickly. Ask for each row in the chat before you reveal it.

Key points. (1) A pilot trial with 48 analysed patients. (2) The primary outcome is cumulative fluid balance at day 5, not mortality. (3) The confidence interval for the difference (-15.2 to 41.6 ml/kg) includes zero, so there is no evidence of a difference. (4) The abstract conclusion matches the result (good practice). (5) The trial was registered.

If a sharp participant asks: the medians (6.9 vs 35.0 ml/kg) differ by more than the reported difference of 13.2 because the 13.2 is a difference in means (the Methods report the mean difference with its 95% CI), not a difference in medians. The course annotations flag this mismatch. Park it for Sessions 6-8.

Link to our question: this trial does not answer the 28-day mortality question. It answers a feasibility-type question about fluid balance. That is exactly the ‘feasible rewrite’ from the FINER slide.

Timing: 5 minutes, row by row. Segment 2.3 ends at 2:05.

2.4 From literature to introduction

The papers you read become the first page of your proposal.

15 minutes: the literature matrix, the problem-statement funnel, a model problem statement, working title and keywords

SEGMENT 2.4 (2:05-2:20, 21:35-21:50).

Teaching point: reading is not the end. A literature matrix turns papers into a gap, the gap into a problem statement, and the PICO into a working title.

Timing: 30 seconds.

The literature review worksheet

Study	Design, size	Compared	Primary outcome: result	Key limitation
Boulet 2024 Crit Care	Pilot RCT, 3 ICUs, n = 48	Restrict hidden fluids vs standard	Fluid balance at day 5: no significant difference	Small, not powered for mortality
Hyun 2023 Ann Intensive Care	Matched cohort, registry, 20 hospitals	Positive vs negative fluid balance	28-day death higher if positive by day 2-3	Observational: confounding possible
Meyhoff 2022 N Engl J Med	Large RCT (CLASSIC), n = 1,554	Restrictive vs standard IV fluid	90-day death 42.3% vs 42.1%: no difference	Clinicians not blinded
Your paper 4	…	…	…	…

KEY MESSAGE One row per paper: 5 rows today, the same columns for every paper.

The worksheet is in Course/Templates (literature review worksheet). One row per paper, the same columns each time.

Rows 1-3 show the course papers: Paper 1 (Boulet 2024, pilot RCT: fluid balance, no significant difference), Paper 2 (Hyun 2023, matched cohort: positive balance on days 2-3 associated with higher 28-day mortality), and CLASSIC (Meyhoff 2022: no difference in 90-day mortality).

Notice: the three papers measure different outcomes at different times. A matrix makes that visible at once.

Teaching point: the ‘key limitation’ column is where the gap, and your study’s justification, comes from.

Timing: 3 minutes.

From literature to your Introduction

KEY MESSAGE Known, not known, why here: the question then follows naturally.

The papers you read become the Introduction of your proposal. It is the same funnel as the top of the IMRaD hourglass.

Paragraph 1, what is known: the size of the problem and current practice, with references.

Paragraph 2, what is not known: the gap (conflicting studies, no studies in patients like ours, outcomes not measured). Your literature matrix’s ‘limitation’ column feeds this paragraph.

Paragraph 3, why it matters here: local data (an audit, an incident review, a number from your unit). Then state the question and aim.

Misconception: ‘the background must review everything ever written’. Paper 1’s introduction does the whole funnel in under 500 words.

Timing: 3 minutes.

A model problem statement

Paragraph	Model text
1 What is known	Septic shock is a leading cause of ICU death. Early fluids restore perfusion, but continued fluids often produce a positive fluid balance, which cohort studies associate with higher mortality [2].
2 What is not known	Large trials of restrictive fluids found no mortality difference [3, 4], and a pilot targeting hidden fluids did not significantly reduce fluid balance [1]. Whether restriction is achievable in our setting is unknown.
3 Why it matters here	Our 3-month audit found that most septic shock patients were fluid-positive by day 3, and practice varies between consultants. We therefore ask: [PICO question].

Model text: replace the audit numbers with your own

KEY MESSAGE Three short paragraphs, each with its own job, and references in brackets.

This is what the funnel looks like as real text, for the feasible rewrite of the running example.

Paragraph 1 (known) cites Paper 2. Paragraph 2 (not known) cites CLASSIC, CLOVERS and Paper 1 and ends with the gap. Paragraph 3 (why here) uses local data (replace the audit statement with your own numbers) and ends with the question.

Teaching point: every factual sentence carries a citation number, and the last sentence leads straight into the PICO question.

Misconception: ‘the problem statement should be long and impressive’. About 300 words is enough for a proposal.

Model text in English. You may write your proposal in Vietnamese, but search and cite in English.

Timing: 4 minutes.

Working title and keywords from PICO

From the PICO

• P: adults with septic shock
• I: restrictive fluids
• C: usual fluids
• O: fluid balance
• D (design): pilot RCT

Working title (13 words)

• “Restrictive versus usual fluids in adults with septic shock: a pilot randomised trial”
• 10-16 words, with population, comparison and design visible
• No abbreviations, and no ‘impact’ or ‘effect’ for observational data

Keywords (3-6)

• MeSH terms that are not already in the title
• Fluid Therapy
• Water-Electrolyte Balance
• Critical Care
• Vasoconstrictor Agents

KEY MESSAGE Your PICO already contains your title: add the design and keep it short.

A good title names the population, the intervention and comparison (or exposure), and the design, in about 10 to 16 words.

Example from the feasible rewrite: ‘Restrictive versus usual fluids in adults with septic shock: a pilot randomised trial’ (13 words).

Avoid abbreviations and promotional or causal words (‘impact’, ‘novel’, ‘effect of’), especially for observational designs (compare Paper 2’s title).

Keywords: 3 to 6, ideally MeSH terms not already in the title, so the paper is found by more searches (the Paper 1 annotations make the same point).

Ask the class (chat): ‘Is Fluids in the ICU a good title?’ Expected: no. It has no population, no comparison and no design.

Timing: 4 minutes. Segment 2.4 ends at 2:20.

Proposal builder 2

Now find the evidence for your own question: one search, five papers, one problem statement, one title.

35 minutes: we search live for the running example, then you search for your own question

PROPOSAL BUILDER 2 (2:20-2:55, 21:50-22:25). The 35 minutes start with this slide.

Teaching point: everything taught today is applied to each participant’s own question from last Saturday. Everyone works on their own proposal. Nobody works in a group.

By the end, each participant has a documented search, some of the 5 references in Zotero and the worksheet, and a first draft of the problem statement and working title. The rest is homework.

Move straight to the next slide.

Proposal builder 2: the evidence for your own question

PROPOSAL BUILDER · 35 min

• Watch us screen the running-example search live, and suggest papers and terms in the chat
• Run your own search string in PubMed. Note the date and the number of hits
• Screen titles and abstracts, pick 5 key papers and save them to Zotero
• Fill one row per paper in the literature review worksheet
• Draft your problem statement (known, not known, why here) and a working title
• Two volunteers read their title and gap sentence aloud

Timing

• 0-10 min: we model it live
• 10-30 min: you search and draft your own
• 30-35 min: two volunteers share

OUTPUT Your Proposal Part 1 continued: search, 5 references, problem statement, title

Minutes 0-10, whole class: re-run the running-example string from the demo, screen the first titles together (participants say ‘keep’ or ‘drop’ in the chat and suggest extra terms), save two papers to Zotero, fill one worksheet row and point to the model problem statement and title.

Minutes 10-30, everyone works alone on their own question. Cameras may go off. The trainers answer questions in the chat. Common problems: too many hits (add a block or a filter), too few (add synonyms), a comparison block, AND inside a block.

Anyone who cannot finish the 5 references records the string, database and date. The remaining references become homework.

Minutes 30-35: two volunteers read their title and gap sentence aloud, microphone on, and the class suggests one improvement each in the chat.

What good looks like: Course/Solutions/s2_solutions.md (Exercise 4).

Timing: 35 minutes. Wrap-up starts at 2:55 (22:25).

Key words of the day

Key word	In one line
MeSH	The shared label PubMed gives to many ways of writing one idea
AND / OR / NOT	Narrows / widens / removes. OR inside a block, AND between blocks
Reference manager	Stores each paper once and formats citations in any style
Patchwriting	Copying with a few words swapped: plagiarism even if cited
IMRaD	Introduction, Methods, Results and Discussion
Spin	Words that claim more than the numbers show

WRAP-UP (2:55-3:00, 22:25-22:30).

Read the six key words quickly. They are also in the course glossary (EN-VN).

Timing: 1 minute.

Session 2 recap

Search

• PICO blocks, MeSH and free text
• OR inside, AND between
• Write the search down

Cite honestly

• Zotero from day one
• Your own words
• Read what you cite

Read smart

• Title, abstract, figures, methods
• Numbers before words

KEY MESSAGE Weeks 1-2 are done: you have a question, five papers and a title.

Read the three messages, then ask one participant: ‘What is the first thing you look for in an abstract?’ Expected: the primary outcome and its result with a CI.

Link forward: over the next two Saturdays (Sessions 3 and 4) we choose the right design to answer each question, and meet bias, confounding and measures of effect using the two fluid papers.

Timing: 1.5 minutes.

Exit question (in the chat)

QUICK QUIZ · 2 min

• Answer now in the chat, or in the 1-question form linked there
• Q1: which gives more hits: septic shock AND fluids, or septic shock OR fluids?
• Q2: what was the primary outcome of Paper 1?

Why we ask

• It is not homework and not marked
• It shows us what is still unclear
• We read every answer before Session 3

OUTPUT One short answer per person, before you leave

Purpose: a 2-minute check of understanding. It is not homework.

Answers: Q1 OR (either term). Q2 Cumulative fluid balance at day 5 (ml/kg), not mortality.

Save the chat (or the form responses) and read them after the session. If more than a quarter get either wrong, open Session 3 with it and revisit it in the recap quiz.

Timing: 1.5 minutes.

Homework before Session 3 (next Saturday)

INDIVIDUAL TASK

• Finish your 5 key references, worksheet rows and problem statement
• Read one of your papers with the 15-minute routine
• Read the abstract of Paper 2 (Hyun 2023) with the reading grid (Exercise 3)
• Exercise 2 (paraphrasing), at home
• Optional: look up Paper 1 (NCT04947904) on ClinicalTrials.gov

Ready for next Saturday

• Your Proposal Part 1
• Your literature review worksheet

OUTPUT Proposal Part 1 complete before Session 3

Be specific: about 60-90 minutes over the week, on your own.

The two papers are in Course/Papers (annotated versions in English and Vietnamese). Session 3 uses both abstracts to compare an RCT with a cohort.

Remind participants to email their Part 1 to the trainers if you offer written feedback between sessions.

Close on time. Thank the class.

Timing: 1 minute.

Extra slides: Session 2

Use if time allows, or for self-study

EXTRA SLIDES (Session 2).

Everything after this slide is optional. Use it if the class is fast or asks, or point participants to it for self-study.

The core session is complete without these slides.

Beyond PubMed: other places to search

Database	What it covers	Access
Cochrane Library	Cochrane systematic reviews (CDSR) and a register of controlled trials (CENTRAL)	Search and abstracts free
Embase	Large biomedical database, strong on drugs and conference abstracts	Subscription
Web of Science, Scopus	Many disciplines, with links to the papers that cite each paper	Subscription
CINAHL	Nursing and allied health	Subscription
Epistemonikos	Systematic reviews in health, in many languages	Free
Registries	Trials: ClinicalTrials.gov, WHO ICTRP. Reviews: PROSPERO	Free

EXTRA: other databases.

Key message: for a Level 1 proposal, PubMed and a registry are enough. A systematic review searches several databases.

Before any new study, look for an existing systematic review first (core slide ‘First check’). The Cochrane Library is the place to start for treatment questions: the Cochrane Database of Systematic Reviews (CDSR) holds Cochrane’s own reviews, and CENTRAL is its register of controlled trials.

Embase overlaps with MEDLINE but adds many drug and European journals and conference abstracts. Web of Science and Scopus cover many disciplines and are strong for following citations. CINAHL is the main nursing database. APA PsycInfo covers mental health and behaviour.

Epistemonikos (free) collects systematic reviews from many sources. WHO Global Index Medicus brings together regional databases, including the Western Pacific Region Index Medicus, which can matter for local-language journals.

Embase, Web of Science, Scopus and CINAHL usually need an institutional subscription. Ask your library.

A systematic review searches at least two or three of these, plus registries and grey literature. Level 3 has a dedicated module on systematic review and meta-analysis (Module A): protocol and PROSPERO registration, a multi-database search, risk of bias, pooling and PRISMA 2020 reporting.

Timing: 2 minutes if used.

PubMed Clinical Queries: a ready-made filter

Open

• PubMed home page
• Clinical Queries, under Find

Type

• Your topic words
• e.g. septic shock fluid restriction

Choose

• Category: therapy, diagnosis, etiology or prognosis

Scope

• Broad: finds more
• Narrow: more precise

KEY MESSAGE A quick shortcut for ‘is there good evidence?’, not a replacement for your own search.

EXTRA: PubMed Clinical Queries.

Clinical Queries applies validated search filters for study types that answer clinical questions.

Good for a quick look before a ward round or to check your own search did not miss an obvious trial.

The interface changes from time to time. Check it before showing it live.

Timing: 2 minutes if used.

Optional demo: Google Scholar and ‘Cited by’

LIVE DEMO · 3 min

• Search: restrictive fluid septic shock mortality
• Look: journals, theses, preprints, ranked by a hidden algorithm
• Under a key trial, click ‘Cited by’, then search within citing articles
• Click the quote icon to see export formats

OUTPUT Participants see what Scholar adds and what it lacks

EXTRA: Google Scholar demo.

Course/Demos/s2_demo.md, Part C.

Say: ‘Scholar is excellent for following citations forward and for grey literature, but there is no MeSH, few filters, and the same search may give different results next month.’

If Scholar shows a CAPTCHA, do not try to solve it while sharing your screen. Skip to the screenshots.

Timing: 3 minutes if used.

Text recycling, AI tools and similarity checks

Text recycling

• Reusing your own published wording
• Check the journal policy

AI rewording

• Does not remove plagiarism
• You stay responsible

AI references

• May be invented
• Find every one in PubMed

Similarity checks

• Journals screen submissions for copied text

KEY MESSAGE Write from understanding, cite what you use, and check every reference yourself.

EXTRA: text recycling, AI and similarity checks.

Text recycling (‘self-plagiarism’): some reuse of standard Methods wording may be accepted if cited and allowed by the journal. Never reuse Results or Discussion text.

AI tools: journals require disclosure of AI use (Session 7 covers integrity and AI in detail).

Timing: 2 minutes if used.

Copyright © Neudata, 2026

All property rights and copyright are reserved.

This presentation contains proprietary data, information, and materials developed by Neudata for dedicated use only and may not be communicated, copied, reproduced, distributed, published, or cited, in whole or in part, without the prior written consent of Neudata. Where explicit written permission has been granted, appropriate attribution must be included, followed by “by courtesy of Neudata”.

Any unauthorized use or infringement of these materials may give rise to legal action and claims for damages, without prejudice to any other rights of Neudata, including rights relating to patents, trademarks, or other forms of intellectual property protection.