Block diffusion offers faster and efficient models for clinical event sequences
Block Diffusion for Longitudinal EHR: Accuracy, Efficiency, and Representation Trade-offs
Machine LearningArtificial Intelligence
Summary
Models that predict what happens in a patient’s medical history usually have to look at clinical events one by one, which can be slow. The authors studied a new way called block diffusion that looks at groups of events together, making the process faster while still keeping predictions accurate. They tested this using hospital data and found that block diffusion learned medical patterns similarly well as traditional methods. Though it isn’t perfect yet for generating full event sequences, it shows promise for quicker clinical predictions.
What this means in practice
- •For hospital data teams: Speed up clinical outcome predictions from patient histories without sacrificing accuracy using block diffusion models.
- •For healthcare software developers: Implement faster inference methods for electronic health record analysis tools by adopting block-based masked diffusion techniques.
Tested on one dataset.
Authors
Ashvin Gupta, Joshua Placidi, Brendan Delaney, Alessandra Russo
Abstract
Autoregressive (AR) models have proved effective for zero-shot prediction from longitudinal electronic health records, but impose strictly sequential generation on trajectories containing temporally clustered and interdependent clinical events. We investigate block-based masked diffusion (BD) as an alternative on MIMIC-IV under matched tokenisation, context length, model capacity, and evaluation conditions. Across block sizes, BD exhibits a clear quality--efficiency trade-off, with small-block configurations retaining much of the predictive performance of AR while enabling substantially faster inference. A representative $B{=}8$ configuration achieves approximately $1.43\times$ faster evaluation. Although generative rollout remains below AR in most settings, frozen linear probes on BD and AR representations achieve near-identical performance across clinical prediction tasks, indicating that BD learns comparably informative clinical representations. Increasing the denoising budget yields diminishing returns, while alternative decoding strategies provide limited gains. Overall, our results show that block diffusion is a viable and computationally attractive approach to longitudinal EHR modelling, with further improvements in iterative decoding offering a promising route toward closing the remaining generative performance gap.