Analysis Code for Paper "Human-AI Collectives Most Accurately Diagnose Clinical Vignettes"

This repository contains the code and example data used for the analyses presented in our publication.

Repository Contents

• analyses.ipynb — Main analysis notebook used to generate results in the paper.
• snomed_matching.ipynb — Notebook for matching diagnoses to SNOMED terms.
• ext_data_figs9.R — R script to reproduce Extended Data Figure 9.
• get_llm_responses.py — Script to query large language models (LLMs) for diagnostic responses.
• llm_prompts.py — Contains functions for interacting with LLMs.
• api_keys.py — Placeholder for API credentials (not included for security reasons).
• requirements.txt — List of Python dependencies.
• data/ — Folder containing example data for one case vignette.