Prompt design for medical question answering with Large Language Models
Bibliographic record
Abstract
The combination of prompting technique and the choice of a foundational model determines end-to-end workflow performance on a given task. We aim to provide comprehensive guidance for the best-performing prompting techniques for various LLMs for medical question-answering. We aim to provide comprehensive guidance for the best-performing prompting techniques for a variety of LLM for medical question-answering. We evaluated 15 large LLMs (incl. Claude 3.5 Sonnet, Gemini pro, Llama, Mistral, OpenAI GPT-4o and 4.1) and 6 smaller models (incl. Gemma, Mistral Nemo, Llama 3.1, Gemini flash) across five prompting techniques on neuro-oncology exam questions. Using the established MedQA dataset and a novel neuro-oncology question set, we compared basic prompting, chain-of-thought reasoning, and more complex agent-based methods incorporating external search capabilities. Results showed that the Reasoning and Acting (ReAct) approach combined with giving LLM access to Google Search performed best on large models like Claude 3.5 Sonnet (81.7% accuracy and 85.5% for v2). We also showed that large models significantly outperformed smaller ones on the MedQA dataset (79.3% vs. 51.2% accuracy) and that complex agentic patterns like Language Agent Tree Search provided minimal benefits despite 5x higher latency. We recommend practitioners to experiment with various techniques given their specific use case and foundational model, and favor simple prompting patterns with large models, as they offer the best balance of accuracy and efficiency.
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 | 0.005 | 0.021 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.002 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.002 | 0.004 |
| Open science | 0.002 | 0.002 |
| Research integrity | 0.002 | 0.003 |
| Insufficient payload (model declined to judge) | 0.008 | 0.004 |
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".