Informative Missingness to Generate Irregular Clinical Time Series

Short summary

Researchers developed a diffusion-model approach to generate synthetic clinical time series that preserve realistic patterns of missing lab tests, addressing the challenge that test absence reflects clinician decisions as much as patient values. Using MIMIC-III data, they demonstrated their method captures meaningful dependencies between patient physiology and testing behavior. Initial results suggest this could help develop clinical foundation models that treat missingness as informative signal.

• •Diffusion models generate synthetic clinical time series preserving realistic missingness patterns
• •Missing data in medical records reflects clinician decisions and patient physiology, not random absence
• •Foundation for clinical AI models that treat missingness as meaningful signal rather than noise

Generated with AI, which can make mistakes.