Temporal Structure in Clinical Narratives in Portuguese: Insights from Cross-Document Annotation
Proceedings of the 8th Workshop on Clinical Natural Language Processing (Clinical NLP) @ LREC 2026
Abstract
Medical reports, by documenting disease progression and patient responses to treatment, form continuous narratives in which each new document adds a chapter to the patient’s clinical story. Constructing coherent patient timelines requires identifying temporal relations across multiple medical reports that compose a patient’s clinical journey. However, cross-document temporal annotation remains an underexplored area, largely due to the methodological and conceptual challenges it entails. This study addresses these challenges by investigating the identification and characterization of cross-document temporal relations in Portuguese medical records. For this purpose, cross-document annotation was performed on different types of reports (Group Consultation Reports, Discharge Reports, and General Reports) from patients diagnosed with Acute Myeloid Leukemia and followed at IPO-Porto, Portugal. Annotation was carried out using the Med2Story scheme, specifically designed to capture both temporal and medical information. Our results indicate that, although cross-document annotation of temporal information is more demanding in terms of both the annotation scheme and the annotation process, it enables the construction of coherent chronological representations of patients’ clinical journeys. Furthermore, the analysis reveals key characteristics of these clinical narratives, including the predominance of nominal events and the prevalence of simultaneity as the most frequent temporal relation type.