A proof-of-concept study for patient use of open notes with large language models
Bibliographic record
Abstract
Abstract Objectives The use of large language models (LLMs) is growing for both clinicians and patients. While researchers and clinicians have explored LLMs to manage patient portal messages and reduce burnout, there is less documentation about how patients use these tools to understand clinical notes and inform decision-making. This proof-of-concept study examined the reliability and accuracy of LLMs in responding to patient queries based on an open visit note. Materials and Methods In a cross-sectional proof-of-concept study, 3 commercially available LLMs (ChatGPT 4o, Claude 3 Opus, Gemini 1.5) were evaluated using 4 distinct prompt series—Standard, Randomized, Persona, and Randomized Persona—with multiple questions, designed by patients, in response to a single neuro-oncology progress note. LLM responses were scored by the note author (neuro-oncologist) and a patient who receives care from the note author, using an 8-criterion rubric that assessed Accuracy, Relevance, Clarity, Actionability, Empathy/Tone, Completeness, Evidence, and Consistency. Descriptive statistics were used to summarize the performance of each LLM across all prompts. Results Overall, the Standard and Persona-based prompt series yielded the best results across all criterion regardless of LLM. Chat-GPT 4o using Persona-based prompts scored highest in all categories. All LLMs scored low in the use of Evidence. Discussion This proof-of-concept study highlighted the potential for LLMs to assist patients in interpreting open notes. The most effective LLM responses were achieved by applying Persona-style prompts to a patient’s question. Conclusion Optimizing LLMs for patient-driven queries, and patient education and counseling around the use of LLMs, have potential to enhance patient use and understanding of their health information.
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 machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.098 | 0.133 |
| Meta-epidemiology (narrow) | 0.002 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.002 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.001 | 0.002 |
| Scholarly communication | 0.003 | 0.003 |
| Open science | 0.003 | 0.002 |
| Research integrity | 0.003 | 0.003 |
| Insufficient payload (model declined to judge) | 0.009 | 0.001 |
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 source (direct Gemma or distilled Codex), 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".