Abstract
<jats:p>Background: Unstructured clinical text in electronic health records contains essential patient information but is difficult to reuse systematically due to limited semantic interpretability beyond keyword search. Objectives: This paper presents the theoretical foundations of transformer-based text vectorization and its benefit potentials for semantic analysis and secondary use of clinical free text. Methods: A conceptual framework is described in which transformer models generate context-sensitive semantic vector representations of clinical narratives, enabling advanced analyses such as semantic retrieval and similarity-based comparison. Results: Transformer-based embeddings support meaning-oriented access to clinical text, automated document structuring, case similarity analysis, and semantic linking across heterogeneous sources, extending classical retrieval and rule-based approaches. Conclusion: Transformer-based text vectorization provides a scalable semantic layer for unstructured clinical documentation and supports systematic secondary use when integrated with appropriate validation, bias control, and governance mechanisms.</jats:p>