Large language model responses to patient-oriented neurointerventional queries: A multirater assessment of accuracy, completeness, safety, and actionability
Bibliographic record
Abstract
BackgroundAs large language models (LLMs) become increasingly accessible to the public, patients are turning to these tools for medical guidance - including in highly specialized fields like interventional neuroradiology. Despite their growing use, the safety, completeness, and reliability of LLM-generated information in subspecialty medicine remain unclear.MethodsFive publicly available LLMs - ChatGPT, Gemini, Claude, Perplexity, and DeepSeek - were prompted with four neurointerventional patient-facing clinical questions spanning ischemic stroke, hemorrhagic stroke, venous disorders, and procedural device use. Each model was queried three times per question to generate unique responses. Eight blinded raters scored each response on accuracy, completeness, safety, and actionability using Likert scales. Plagiarism analyses were also performed.ResultsDeepSeek consistently outperformed other LLMs in accuracy, completeness, and actionability across four prompts, while Gemini frequently ranked worse, including in plagiarism levels. ChatGPT performed well in accuracy. Physicians were more critical than non-physicians across accuracy, completeness, and safety, whereas non-physicians rated actionability significantly lower. Overall, LLMs were rated relatively high (median of >4 on a 5-point scale) in medical safety, suggesting low risk of overtly harmful advice.ConclusionRecent-generation LLMs offer medically safe, though often incomplete or imprecise, information in response to patient-oriented neurointerventional queries. Including non-physician raters revealed valuable differences in perception that are relevant to how patients may interpret LLM outputs. As benchmark frameworks like HealthBench improve LLM evaluation, inclusion of lay perspectives and subspecialty contexts remains essential. Responsible use by clinicians and ongoing patient education will be critical as LLM use in healthcare expands.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.001 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".