New diffusion model treats missing lab data as informative signals for clinical AI

Laboratory tests in electronic health records are collected irregularly, and the absence of a test order can be as informative as the measurement itself. This phenomenon, known as informative missingness, reflects both clinicians' decisions and patient physiology. Researchers from multiple European institutions have now developed a diffusion-based approach that models these patterns directly, rather than treating missing data as a preprocessing artifact. Their method, built on the TimeDiff framework, aligns chart times into 4-hour intervals and segments admissions into 7-day windows, producing trajectories that pair each lab value with a corresponding observation indicator. Using the public DACMI benchmark derived from MIMIC-III, the model learns continuous lab values and discrete missingness patterns through complementary diffusion objectives.

Preliminary experiments show that the generated data closely matches real patient trajectories across both individual lab distributions and joint value-missingness embeddings. The model successfully captures clinically meaningful dependencies under MNAR-like missingness, demonstrating that diffusion models can preserve the crucial link between patient physiology and clinicians' testing behavior. These results position the work as an initial component toward developing clinical foundation models. By producing synthetic priors that retain key physiology-missingness relationships, the authors plan to next train Prior-Data Fitted Networks capable of leveraging informative missingness for more robust healthcare predictions.