This repository is intended to support reproducibility for the manuscript:
Patient Representation and Social Context in Digital Medical Education Question Banks: Cross-Sectional Content Analysis
The code supports a cross-sectional content analysis and dataset quality audit of public medical examination question-bank records used as digital medical education resources and, secondarily, AI medical question-answering benchmarks.
MedQA-US, MedQA-Mainland China, and MedQA-Taiwan are the primary source partitions. MedMCQA is a conditional extension only.
The names MedQA-US, MedQA-Mainland China, and MedQA-Taiwan reproduce the labels used in the source dataset files and directories and are retained solely to identify corpus variants and examination contexts. They are provenance labels, not geopolitical classifications, nationality designations, or patient-origin variables. No inference about patient origin is made from these fields.
This public release candidate does not include raw question-bank files, full question stems, answer options, source-text excerpts, reviewer notes, R1/R2 coding files, adjudication files, consensus files, restricted UID mapping, or other internal-only materials.
Public data files use study-specific pseudonymized identifiers and contain derived categorical coding fields only. Internal reliability/adjudication/free-text note tables are not included in this GitHub candidate bundle.
The workflow documents:
- Source registry and immutable raw-file manifest.
- Standardized source-record structure.
- Transparent high-sensitivity candidate rules.
- Calibrated false-negative audits and residual safety inclusion.
- Independent human eligibility review.
- Independent R1/R2 main coding with field-level reliability checks.
- Human adjudication followed by dataset lock before analysis.
The v2.2 candidate algorithm is a high-sensitivity sampling aid, not a final eligibility classifier.
Full replication from raw sources requires access to the original public question-bank files under their source licenses and terms. Public verification of manuscript values can be performed from the public no-stem data, figure source data, public figures, and scripts.
Analysis code in this repository is licensed under the MIT License.
The public no-stem coded datasets, aggregate tables, figure source data, data dictionaries, and documentation are licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0), unless otherwise noted.
Complete question stems, answer options, source-text excerpts, reviewer notes, internal coding text, and any third-party source item content are not redistributed in this repository and are not licensed through this repository. Raw question text may be subject to source licensing or copyright restrictions.