arXiv:2605.26560v1 Announce Type: new 
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.

تم تطوير نظام هجين عصبي-رمزي جديد لاستخراج التعليمات السريرية للمتابعة بشكل موثوق من ملاحظات المرضى الخارجيين، محققًا دقة عالية في تحديد أزواج العمل-التاريخ. تتكامل هذه الطريقة مع BioBERT للتوسيم ورابط ثنائي الاتجاه لعلاقات الكيانات، مما يظهر أداءً متفوقًا مقارنة بالنماذج التوليدية مثل GPT-4o-mini وLLaMA-3.

Se ha desarrollado un nuevo pipeline híbrido neuro-simbólico para la extracción fiable de instrucciones de seguimiento clínico a partir de notas de pacientes ambulatorios, logrando una alta precisión en la identificación de pares acción-fecha. Este método integra BioBERT para el etiquetado y un enlace biafino para las relaciones entre entidades, demostrando un rendimiento superior en comparación con modelos generativos como GPT-4o-mini y LLaMA-3.

Un nouveau pipeline hybride neuro-symbolique a été développé pour l'extraction fiable des instructions de suivi clinique à partir de notes de patients externes, atteignant une grande précision dans l'identification des paires action-date. Cette méthode intègre BioBERT pour le balisage et un lien biaffine pour les relations entre entités, montrant des performances supérieures par rapport aux modèles génératifs comme GPT-4o-mini et LLaMA-3.

A new hybrid neural-symbolic pipeline has been developed for the reliable extraction of clinical follow-up instructions from outpatient notes, achieving high accuracy in identifying action-date pairs. This method integrates BioBERT for tagging and a biaffine linker for entity relationships, demonstrating superior performance compared to generative models like GPT-4o-mini and LLaMA-3.