A patient brings in a study they read online

The situation: a headline in the consultation

A patient arrives with a printout, a screenshot or a phone held up across the desk: a headline saying that a common medicine causes dementia, that a diet doubles heart risk, or that a supplement works as well as a prescription drug. Patients bring these because they care about their health, and how the question is handled affects their trust in the advice that follows. The difficulty is that the clinician has rarely seen the study, the article often does not link to it, and the consultation has a fixed length. The honest answer depends on details that take time to find: whether the study was a randomised trial or an observational cohort, how many people were in it, whether it was published in a peer-reviewed journal or only presented at a meeting, and how the result compares with the rest of the evidence. Questions that need this look like:

• Does long-term use of a common acid-suppressing drug class cause dementia, and how strong is that evidence?
• A news story says time-restricted eating raises the risk of cardiovascular death. What kind of study was this?
• Is a popular supplement as effective as standard treatment, and what do systematic reviews say?

What goes wrong without a cited source

Without a cited source, the conversation becomes one opinion against a headline, and the clinician often loses. Dismissing the study without having read it can sound defensive; agreeing with it can lead to stopping a useful treatment. A general web search tends to surface more news stories about the same headline rather than the paper, and general AI chatbots have produced references that do not exist when asked for studies, so a confident summary can describe the wrong paper. Common misreadings repeat: an association in an observational study reported as a cause, a relative risk increase presented without the absolute risk, a result from a small or non-randomised group presented as settled, and a conference abstract treated as peer-reviewed evidence. Each makes the eventual explanation harder and can send the patient away less sure of their treatment than before.

• Association reported as cause. Observational data can show a link, not that one thing causes the other.
• Relative risk without absolute risk. "Doubles the risk" can mean a change from very rare to rare.
• Abstract treated as a paper. Meeting abstracts are often not yet peer reviewed.

How AskMed helps with a study a patient read

AskMed helps by placing the headline study inside the wider evidence, with every claim linked so you can show the patient where it comes from. Ask AskMed about the claim and it searches a fixed list of medical sources: PubMed, the Cochrane Library and journals such as NEJM, The Lancet, JAMA and The BMJ, guideline bodies such as NICE, WHO and the US Preventive Services Task Force, and drug labels. It prefers guidelines first, then systematic reviews and meta-analyses, then reviews and large trials. For an evidence question, AskMed is instructed to cover the key trials, meta-analyses and guideline positions with their populations and results, and to say where the evidence is weak or conflicting. Research mode reads the full text of the most relevant result before writing. If the patient brings a text-based PDF, attach it: AskMed can cite the file for what it says.

• Study type filters. Choose systematic reviews and meta-analyses, review articles, clinical trials or other studies.
• Sources that disagree. The answer says so and says which evidence is stronger.
• The patient's language. Answers are written in the language of the question: English, French, Spanish, Portuguese, Italian, Japanese or Arabic.
• AskMed works on iPhone and iPad, Android and the web, with one account on all of them.

Checklist: answering a patient's study

The checklist below gives a structure for answering a patient's study in a consultation, or at a follow-up if it needs more time. It mirrors the questions any critical appraisal asks: what the study is, who was in it, what it measured, how large the effect was in absolute terms, and how it fits with the rest of the evidence. Doing those steps in order stops the conversation from turning into a debate about the headline. The steps use AskMed on iPhone, iPad, Android or the web for the evidence search and the plain-language explanation, and the cited sources mean you can show the patient the guideline or review rather than just your opinion. It also helps to thank the patient for bringing the study, since the aim is a shared decision, and to agree what would change the plan if stronger evidence appears later.

1. Ask where they read it. A news site, a social media post or the paper itself, and whether there is a link.
2. Find the study type and population. Randomised trial, cohort, case-control, abstract or animal study, and who was included.
3. Look for the absolute effect. How many people per hundred or thousand were affected, not only the relative change.
4. Compare with reviews and guidelines. Ask AskMed what systematic reviews and the current guideline say, and open the citations.
5. Explain in plain words. Ask for a plain-language version, then check it still matches the sources.
6. Agree a plan. Decide together whether anything changes, and when to review it.

Limits when answering a patient's study

AskMed has limits that matter when a patient's study is the question. AskMed searches a fixed list of journals, databases and guideline bodies, so a paper published in a journal outside that list may not appear unless PubMed indexes it, and a preprint or a conference abstract may not be found at all. Answers are written by an AI language model, with no human review before you see them, so an answer can misread a study or miss one; open the citation before you quote a number to a patient. AskMed is built for healthcare professionals and students, not for patients' self-diagnosis, and it is not a symptom checker. It does not decide treatment for a person, and it never gives drug doses. Attach a patient's document only if you are permitted to share it, because attached text is sent with the question to generate the answer.

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.
• ScenarioChecking a NICE guideline as a GPFind what NICE recommends for a UK primary care question, with the year and the link.