arXiv:2602.00541v2 Announce Type: replace Abstract: Clinical events captured in Electronic Health Records (EHR) are irregularly sampled and may consist of a mixture of discrete events and numerical measurements, such as laboratory values or treatment dosages. The sequential nature of EHR, analogous to natural language, has motivated the use of next-token prediction to train prior EHR Foundation Models (FMs) over events. However, this training fails to capture the full structure of EHR. When a given event occurs must be captured, but the event value (abnormal lab) also modulates the likelihood
The proliferation of Electronic Health Records and advancements in AI foundation models are converging, making this a critical area for innovation in clinical data processing.
Improving how AI models interpret complex, irregular EHR data can lead to more accurate clinical predictions, personalized medicine, and more efficient healthcare systems.
The proposed 'marked time-to-event' loss function offers a more nuanced way to train EHR foundation models, potentially overcoming limitations of simpler next-token prediction methods by integrating temporal context and event values.
- · AI healthcare startups
- · Pharmaceutical companies (drug discovery)
- · Hospitals and healthcare providers
- · Patients
- · AI models reliant on simplified EHR representations
- · Traditional clinical predictive analytics firms
More robust and accurate AI models for predicting patient outcomes and disease progression.
Accelerated development of personalized treatment plans and proactive healthcare interventions.
Potential for AI-driven healthcare to significantly reduce diagnostic errors and improve overall public health metrics globally.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG