A Pipeline for Generating Longitudinal Synthetic Clinical Notes Using Large Language Models

A new pipeline uses large language models to generate longitudinal synthetic clinical notes, offering a privacy-safe resource for developing clinical AI tools without real patient data [1][2]. The modular pipeline combines structured patient generation, semi-structured patient journey simulation, and unstructured clinical note generation using large language models [1][2]. It is designed to maintain internal consistency across longitudinal patient records while capturing variation in writing style, note structure, and clinical detail [1][2]. The resulting dataset comprises 70 synthetic patients, each associated with 20-50 clinical notes spanning a full hospital journey [1][2]. Large language models, or LLMs, are machine learning models with many parameters trained on vast amounts of text for natural language processing tasks such as language generation [8]. The pipeline incorporates LLM-based validation and augmentation steps to improve the faithfulness, realism, and diversity of the generated notes [1][2]. The dataset is released at multiple levels of validation, enabling users to balance realism and scalability depending on their use case [1][2]. Synthetic data is increasingly used to enable AI development and evaluation in domains where access to real-world data is restricted [2]. In healthcare, clinical documentation presents particular challenges due to its sensitivity [2]. This dataset supports the development, testing, and evaluation of clinical AI systems, including summarisation tools, coding models, and decision support systems, without reliance on real patient data [1][2]. The work was submitted on 25 June 2026 [1]. The paper is hosted on arXiv, a preprint server that allows researchers to share findings before formal peer review. Associated code and data are linked through platforms including Hugging Face and DagsHub [1]. Hugging Face is a company known for hosting machine learning models and datasets, and its former Head of Research, Douwe Kiela, co-authored the foundational 2020 paper introducing Retrieval-Augmented Generation [7].