TRV-2026-2904Version 1 · Published

Saved · see the current record

Why this version exists

Certified into the record

The exact text we saved

This is the version exactly as it was saved, character for character. The SHA-256 hash below is a 64-character code worked out from that text. Run the same text through SHA-256 and you get the same code; change a single letter and the code comes out completely different. Each version also stores the previous version’s hash, which chains them together so no version can be swapped out quietly.

TRUVACE RECORD VERSION
record: TRV-2026-2904
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-11-14T00:03:10.616220Z
status: published
lens: p_space
sector: health
headline: Hallucinated Evidence and Real-World Risk-Physicians' Intent to Act on AI-Generated Abstracts in Emergency Medicine and Trauma Surgery: Cross-Sectional, Vignette-Based Diagnostic Accuracy Study
dek: Background: Large language models (LLMs) are increasingly used in academic writing and can generate linguistically polished but factually incorrect scientific content. Such misinformation may distort evidence appraisal and, if translated into clinical practice, compromise patient care.
Objective: This study aimed to estimate physicians' likelihood of misperceiving false information embedded in LLM-generated abstracts as true and reporting an intent to use that information in future practice. We also explored whi…
gain_title: (none)
problem_title: In a vignette survey of 20 emergency and trauma physicians at 4 German hospitals, doctors said they would act on 57% (31/54) of the abstracts they chose to act on that were GPT-5-generated fakes, and rated true and false abstracts similarly for accuracy.
trace_subject: (none)
gain_reading: (none)
gain_evidence: (none)
problem_reading: In a vignette survey of 20 emergency and trauma physicians at 4 German hospitals, doctors said they would act on 57% (31/54) of the abstracts they chose to act on that were GPT-5-generated fakes, and rated true and false abstracts similarly for accuracy.
problem_evidence: Overall, physicians reported that they would act on 68% (54/80) of abstracts. | True and false abstracts received similar ratings for perceived accuracy | Physicians frequently reported an intent to act on LLM-generated abstracts containing false claims
quick_read: Twenty emergency medicine and trauma surgery physicians at 4 hospitals in Germany each rated 4 abstracts written with GPT-5 Thinking mode. Two abstracts held deliberately false claims backed by plausible explanations; two held accurate claims. Across 80 ratings, physicians said they would act on 68% (54/80) of the abstracts.

Intent to act barely separated truth from falsehood. Sensitivity was 78% and specificity 43%, so more than half of the false abstracts (23 of 40) drew a stated intent to act. The positive likelihood ratio of 1.35 had a confidence interval (0.98-1.85) reaching down to 1.

The result matters because LLMs are increasingly used in academic writing and can produce polished but wrong content. The sample was small and the task was hypothetical, so the study measures stated intent rather than actual clinical decisions.
limitation: The study involved only 20 physicians from 4 hospitals in Germany, used vignettes, and measured reported intent to act rather than real clinical decisions.
tag: Evidence-backed problem
key_points: Design: cross-sectional vignette survey of physicians in emergency medicine and trauma surgery at 4 hospitals in Germany. | Twenty physicians gave 80 ratings; each assessed 4 GPT-5 Thinking abstracts, 2 false and 2 accurate. | Physicians said they would act on 68% (54/80) of abstracts. | Using truth status as the reference, acting had 78% sensitivity (31/40), 43% specificity (17/40), positive predictive value 57% and negative predictive value 65%. | Perceived accuracy was similar for true and false abstracts (mean 3.15 vs 2.88; P=.21). | Abstracts physicians would act on scored higher for accuracy, precision and structure (all P<.001); grammar did not differ (P=.80).
rundown: The study tested whether LLM-generated misinformation could pass as credible to practising clinicians. Each participant read four GPT-5 Thinking abstracts, half with intentionally false claims and half accurate. The main outcome was the proportion of doctors reporting an intent to use or act on recommendations in the false abstracts.

Diagnostic-accuracy measures showed weak discrimination: positive predictive value was 57% (31/54) and negative predictive value 65% (17/26). The diagnostic odds ratio was 2.55 (95% CI 0.96-6.73), with intervals that include no discrimination.

Ratings of precision (3.15 vs 2.93; P=.29), structure (3.23 vs 3.20; P=.89) and grammar (3.40 vs 3.43; P=.86) were also similar for true and false abstracts. Yet abstracts that doctors would act on scored substantially higher on perceived accuracy, precision and structure, suggesting that polished presentation drives credibility.
sources:
- peer_reviewed | Journal of Medical Internet Research | https://doi.org/10.2196/91095 | 2026-11-13
prev: 0000000000000000000000000000000000000000000000000000000000000000
SHA-256 hash
d9ef7a7eb618f8e520482d2fe8235410c044f19d45a2cb7c847255b6077a99f9
Previous version’s hash
0000000000000000000000000000000000000000000000000000000000000000
Check this record hasn’t been altered
How to check it yourself, without trusting this page

Download any version from /api/record/TRV-2026-2904 and hash its canonical field yourself, for example with shasum -a 256. You should get its content_hash. Each version’s text also ends with prev:and the hash of the version before it (version 1 points to 64 zeros). If anyone had changed a single character of any version after it was saved, those hashes wouldn’t line up.