Governed clinical variable mapping with browser-local reasoning

2025 – present · AstraZeneca

Mappings proposed, not applied — 426 nodes, 332 edges, 0 of 84 approved. A mapping waits for a reviewer.

Runs against CDISC's public pilot, ranking 415 candidate variables against 110 known-good pairs with embeddings frozen between runs so the comparison holds. Inference happens in the browser on WebGPU. On derivations a frontier model reaches 0.38 top-1 against 0.06 lexical, on a leak-free benchmark.

Keyframes recorded live in the browser build, on the public CDISC pilot; the notes either side are read off the page. AE.AEACN is selected, Qwen2.5-0.5B-Instruct (q4) loads into the tab (786 MB, first use only) and runs on the GPU through WebGPU, and proposes ADAE.AEACN at 80%. The panel says so outright — proposal only, not applied — and 0 of 84 mappings are approved. Time is compressed; the timings on the cards are the measured ones.

Mapping study variables to analysis variables is slow, manual and repeated per study. A model can propose mappings, but a proposal accepted because the model sounded confident is not a mapping — it is a guess with provenance attached.

Built on the public CDISC pilot: embedding retrieval over a 415-variable candidate pool against 110 gold pairs, with identical cached embeddings across every run so comparisons mean something. Candidate scoring runs in-page over WebGPU, so the data does not leave the browser. A mapping is always a proposal a reviewer accepts, never something applied automatically.

On the derivation split, the hard half, lexical and embedding retrieval reach top-1 accuracy of 0.06 and 0.08; a frontier model reranking the same candidates reaches 0.38, and a 7B on-premises model 0.10. The benchmark was rebuilt after the gold answers were found leaking into the baselines' candidate text. The gain depends on CDISC variable names, and recall is now the constraint: the right answer is in the top 25 for 31 of 50 derivations.

Accepted at the 2nd Workshop on Large Language Models and Big Data Analytics for Health and Medicine, part of the 2026 IEEE International Conference on Bioinformatics and Biomedicine (IEEE BIBM 2026).

References

CDISC. Analysis Data Model (ADaM). CDISC foundational standard. cdisc.org
CDISC. SDTM/ADaM pilot project. GitHub, cdisc-org. github.com/cdisc-org
Reimers and Gurevych (2019). Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks. EMNLP-IJCNLP 2019, 3980–3990. doi:10.18653/v1/D19-1410

No public URL yet — paper accepted; the proceedings record follows the conference.