Problem
Research questions and friction points this paper is trying to address.
follow-up instructions
clinical text extraction
action-date pairing
outpatient notes
temporal information extraction
Reliable Extraction of Clinical Follow-Up Instructions: A Hybrid Neural-Symbolic Pipeline
🤖 AI Summary
This work addresses the challenge of accurately extracting follow-up instructions—such as “schedule a brain MRI in two weeks”—from outpatient clinical notes, which requires precise identification and reliable association of action verbs with future time expressions to support scheduling and auditing. Existing generative approaches struggle to consistently link these elements. To overcome this limitation, the authors propose a neuro-symbolic hybrid method: first, BioBERT combined with BIO tagging and a biaffine coreference linker extracts action and time entities; then, deterministic rules normalize temporal expressions into day offsets and map actions to a 28-class ontology. This decoupled design substantially improves the accuracy of action–time pair extraction, achieving F1 scores of 0.997 and 0.986 on seen and unseen action test sets, respectively, with a mean absolute error of 0 days—significantly outperforming generative baselines, which attain only 0.51–0.57 Pair F1.
📝 Abstract
Objective. Outpatient notes carry follow-up instructions pairing actions with future times ("MRI brain in two weeks"). Extracting (action, date) pairs supports scheduling and audit, but generative extractors miss the date because linking and arithmetic are implicit in decoding. We test a hybrid neural-symbolic pipeline against direct generation. Methods. We define TestSpecification and TimeSpecification entities and a ScheduledFor relation. BioBERT feeds BIO tagging and a biaffine linker; entities are canonicalized via a 28-action ontology and times normalized to day offsets deterministically. We evaluate on a 2,000-note synthetic outpatient corpus with action-disjoint splits (18 train, 6 OOV-test) against zero-shot GPT-4o-mini and LoRA-fine-tuned LLaMA-3 8B with note-level bootstrap 95% CIs. Results. On 259-note seen and OOV splits the hybrid pipeline achieves Test-Time Pair F1 of 0.997 and 0.986 with 0.00-day MAE. Baselines reach high action F1 (LLaMA-3 0.992; GPT-4o-mini 0.963 seen) but Pair F1 stays at 0.51-0.57 (LLaMA-3) and 0.53 (GPT-4o-mini), CIs non-overlapping with the hybrid. Conclusion. Separating learned entity extraction from deterministic date arithmetic outperforms generation on this benchmark, generalizes to held-out actions, and exposes failure modes. Transfer to real EHR notes is the next validation; a first-pass realism check is in Limitations.
Innovation
Methods, ideas, or system contributions that make the work stand out.
neural-symbolic pipeline
clinical follow-up extraction
temporal normalization
entity-relation extraction
out-of-vocabulary generalization
🔎 Similar Papers
No similar papers found.
