MedEvi-NS at ArchEHR-QA 2026: Using Clinical Reasoning Principles to Improve Zero-shot Capabilities of Large Language Models in Evidence Alignment

Abstract

The ArchEHR-QA shared task focuses on grounded question answering using patient EHR data. For the given clinical interpretation of the patient question, note excerpt (E) and answer text (A), subtask 4 (evidence alignment) aims to cite supporting sentences from E for each sentence in A. In this paper, we propose a prompt-engineering methodology that features clinical-reasoning principles in related alignment. We adopt this methodology for GPT-5.2 in zero-shot learning mode. According to our experiments on ArchEHR-QA, incorporating clinical reasoning principles into the prompt improves F 1overall by +2.0%. Our final submission resulted in 77.4% by F 1overall, which positions us at 10th out of 16 teams. Our code is publicly available: https://github.com/nicolay-r/ArchEHR-QA-2026-Task-4-MedEvi-NS

Resources

Details