Transparency-First Medical Language Models: Datasheets, Model Cards, and End-to-End Data Provenance for Clinical NLP

TL;DR

This paper presents TeMLM, a comprehensive transparency framework for clinical language models, including artifacts and checklists for provenance, data, and model transparency, demonstrated on a synthetic dataset and real clinical tasks.

Contribution

The authors introduce a unified, machine-checkable transparency artifact suite for clinical NLP models, enhancing accountability and reproducibility in medical AI development.

Findings

TeMLM artifacts enable systematic auditing of clinical language models.

Models evaluated on synthetic data show promising results for PHI de-identification and ICD-9 code extraction.

Synthetic benchmarks are useful but real data validation remains essential.

Abstract

We introduce TeMLM, a set of transparency-first release artifacts for clinical language models. TeMLM unifies provenance, data transparency, modeling transparency, and governance into a single, machine-checkable release bundle. We define an artifact suite (TeMLM-Card, TeMLM-Datasheet, TeMLM-Provenance) and a lightweight conformance checklist for repeatable auditing. We instantiate the artifacts on Technetium-I, a large-scale synthetic clinical NLP dataset with 498,000 notes, 7.74M PHI entity annotations across 10 types, and ICD-9-CM diagnosis labels, and report reference results for ProtactiniumBERT (about 100 million parameters) on PHI de-identification (token classification) and top-50 ICD-9 code extraction (multi-label classification). We emphasize that synthetic benchmarks are valuable for tooling and process validation, but models should be validated on real clinical data prior to deployment.