Real references for a case report

The situation: a case worth writing up

A resident or clinician sees an unusual presentation, a rare adverse reaction or a new association and decides to write it up as a case report, or needs to support a clinical note or teaching document with references. Journals that publish case reports usually ask for a short introduction, the case itself, and a discussion that sets the case against what has already been published: how many similar cases exist, what mechanisms have been proposed, and what the guidelines say. Many follow the CARE guidelines, a reporting checklist for case reports. The case is the author's own; the discussion depends on finding the right literature and citing it accurately. That search is where most of the time goes, particularly for rare conditions where the evidence is scattered across small case series and reviews. Discussion sections typically need sources for statements like these:

• How many cases of this adverse reaction have been reported, and in which populations?
• What mechanisms have been proposed for this association?
• What do current guidelines recommend for screening or prevention?

What goes wrong with references from a chatbot

References written by a general AI chatbot from memory can be wrong in ways that are hard to see. Studies of early ChatGPT versions found that many of the references it produced were fabricated, with plausible authors, journals and years, and many of the real ones contained errors in the details. A reference that does not exist can pass a quick read and be caught by a reviewer or an editor, which damages the submission and the author's credibility. A subtler problem is a real paper cited for a claim it does not make: the study exists, but it was in a different population, measured a different outcome, or reached the opposite conclusion. Copying references from another paper's list without reading them carries the same risk. Each reference needs three checks: that it exists, that it says what the sentence says, and that the citation details are correct.

• Fabricated references. Plausible authors, journal and year, but no such paper.
• Real paper, wrong claim. The source exists but does not support the sentence.
• Wrong details. Incorrect year, volume, pages or DOI.

How AskMed helps you find real references

AskMed helps by searching real medical sources and linking every claim to the exact page it came from. AskMed searches a fixed list of sources before it writes: PubMed and PubMed Central, the Cochrane Library, NEJM, The Lancet, JAMA, The BMJ and specialty journals; guideline bodies such as NICE, WHO and CDC; FDA and DailyMed drug labels; and references such as StatPearls. Each sentence links to the source that supports it using that source's exact web address, and AskMed is instructed to cite a source only for what it says and never to invent web addresses, studies or statistics. An answer that comes back citing nothing is replaced by a fixed reply rather than shown as evidence. Study type filters include "other studies, such as observational studies and case reports", so you can ask for case-level evidence or for reviews only.

• Share with sources. Sharing an answer copies the question, the answer and the numbered sources with their titles and links, ready to paste into your notes.
• Labelled sources. Each source is tagged Guideline, Research, Drug label or Reference.
• History. Answers stay in History while you write, unless Private Mode is on.
• AskMed works on iPhone and iPad, Android and the web, with one account on all of them.

Checklist: references for a case report

The checklist below takes a case report discussion from a list of claims to a reference list you can defend. It separates finding a source from checking it and formatting it, because each is a different kind of error. AskMed fits the finding step: it returns linked sources for each claim in seconds, which replaces much of the first round of database searching, and it works on iPhone, iPad, Android and the web. The checking and the formatting stay with you, because no tool can confirm that a source supports your exact sentence as well as reading it can, and because AskMed links sources but does not format them in Vancouver, AMA or a journal's own style. Keep a simple table while you work: the sentence, the source, the section you read and the formatted reference. Then de-identify the case and get the patient's consent as your journal requires.

1. List the claims your discussion makes. Each claim that is not about your own case needs a source.
2. Ask one specific question per claim. Use study type filters for case reports, trials or reviews as needed.
3. Open every linked source. Confirm the paper exists, and read the part that supports your sentence.
4. Check the population and outcome. A study in a different group or on a different outcome does not support the claim.
5. Copy the citation details from the source. Take authors, title, journal, year and DOI from PubMed or the publisher's page.
6. Format in the journal's style. Vancouver, AMA or the journal's own instructions for authors.

Limits for case reports and notes

AskMed has limits as a reference-finding tool for case reports and notes. It searches a fixed list of sources, so a relevant paper published only in a journal outside that list may not appear unless PubMed or PubMed Central indexes it, and a search for a rare condition may return few sources; when the evidence does not answer the question, AskMed is instructed to say so rather than fill the gap. AskMed does not run a systematic, reproducible search, so it cannot show that you found every published case. Answers are written by an AI language model, with no human review before you see them, so a link can sit beside a sentence its source does not fully support: read every source you cite. AskMed does not format references. Never attach identifiable patient information unless you are permitted to share it, and follow your journal's consent rules.

Related

• GuideIs ChatGPT accurate for medical questions?What studies found on chatbot accuracy and made-up references, and how to check any AI answer.
• GuideHow AskMed answersWhere answers come from, how citations work, why there are no doses, and the limits.
• ScenarioPreparing for journal clubPlace one paper among the trials, reviews and guidelines around it.
• ScenarioA rash differential with sourcesDescribe the morphology and get a cited differential with the first test for each.
• ScenarioPreparing for ward roundsCheck the criteria behind each plan before your consultant asks.