Informative Missingness to Generate Irregular Clinical Time Series
Researchers have developed a novel diffusion-based method to generate synthetic clinical time series data, which effectively models both laboratory values and their irregular observation patterns. This approach, detailed in a new arXiv paper, treats missing data not as an artifact but as an informative signal reflecting clinical decisions and patient physiology. By extending the TimeDiff framework, the model captures realistic sampling and clinically meaningful dependencies, demonstrating potential for use in developing clinical foundation models. AI