This continuation project builds on work conducted within the ERC project ActDisease (ERC-2021-STG-101040999) at Uppsala University, which investigates how European patient organizations shaped modern medical practices through their communicative strategies. The project is based on a multilingual dataset of historical medical periodicals (1875–1990) in English, German, French, and Swedish. In the previous NAISS allocation, our efforts focused primarily on historical medical magazines, where we developed and validated multimodal pipelines combining Vision-Language Models with layout embeddings derived from LayoutXLM and OCR data. These experiments demonstrated the potential of multimodal approaches while also revealing important limitations related to model robustness and computational cost. In parallel, we developed a training-free approach to biomedical entity linking for historical medical texts, which resulted in a peer-reviewed publication accepted to be published in EACL 2026 proceedings and the release of supporting code and benchmark dataset. The primary goal of this continuation is to extend these approaches to historical medical journals, which represent a distinct and methodologically important subset of the corpus. Compared to magazines, journals exhibit different structural conventions, denser informational content, and more complex relationships between textual and visual elements. This shift enables us to test the generalisability of our previously developed methods and to analyse communicative strategies across a broader spectrum of medical publishing practices. A central focus of the project is a systematic evaluation of multiple Vision-Language Model architectures and sizes. We will compare different model families, including Phi-based multimodal models, LLaVA, Pixtral, and Qwen-VL variants, examining how architectural differences and parameter scale influence performance in genre inference and communicative purpose classification. Particular attention will be paid to zero-shot and few-shot settings, as well as to model robustness in the presence of OCR noise, multilingual variation, and complex page layouts typical of historical journals. This evaluation requires substantial computational resources. Running repeated inference across multiple models and configurations on high-resolution journal page images is computationally intensive and cannot be performed efficiently with local infrastructure. Additionally, parts of the dataset are under copyright, which necessitates secure local processing and excludes the use of external commercial APIs. Access to NAISS resources is therefore essential to enable systematic experimentation and ensure compliance with data handling constraints. The expected outcomes include a comparative assessment of multimodal model performance on historical medical journals, methodological advances in multimodal document analysis, and publications in venues such as Digital Humanities Quarterly and the Association for Computational Linguistics. By extending our prior work from magazines to journals and incorporating insights from our biomedical entity linking research, this project will establish a more general and empirically grounded framework for analysing historical medical communication.