Discrete Neural Interlingua for Ontology-Agnostic EHR Modeling

MEDIATOR is an ontology-agnostic discrete tokenizer that maps arbitrary clinical concept text into a shared vector-quantized codebook, without manual mappings or a fixed vocabulary. It raises mean AUPRC over the strongest prior tokenizer by 4.0 points on 20 MIMIC-III tasks, and by 8.6 points in zero-shot transfer to 14 eICU tasks while using all events. Control experiments show the text encoder drives in-domain accuracy, while discretization objectives drive transfer robustness.