Tabular Population Priors as Structured Retrieval for EHR Medication Set Prediction
Paper · https://arxiv.org/abs/2611.17341
arXiv:2611.17341v1 Announce Type: new Abstract: Medication prediction from electronic health records is a fixed vocabulary structured prediction problem, yet recent clinical RAG systems often treat it like open ended case based reasoning: retrieve similar patient admissions and ask an LLM to infer a medication set. This retrieval object is indirect the model receives retrieved patient records while the target is a discrete set of medication class labels. We propose a more task aligned retrieval object: a tabular population prior ranked medication class candidates from a sparse condition medication co occurrence table estimated from training admissions. Using PACERAG as a fixed downstream pipeline and replacing only its patient retriever, this drop in swap improves F1 from 0.303 to 0.338 (+11.5% relative) and Jaccard from 0.218 to 0.237 on 1,004 heldout MIMIC IV admissions, with no additional LLM calls and no inference time patient index. A non LLM top k baseline reaches F1=0.248, confirming the prior is predictive before LLM integration. Aggregate population statistics can provide a more task aligned, efficient, and auditable retrieval source than patient level nearest neighbors for fixed vocabulary EHR set prediction.
First seen by Oriel on .
AI summary
This paper proposes retrieving ranked medication candidates from population-level condition–medication statistics, improving EHR medication set prediction over patient-level retrieval.
- The method uses a sparse condition–medication co-occurrence table estimated from training admissions to rank medication class candidates.
- Replacing only PACERAG’s patient retriever improved F1 from 0.303 to 0.338 and Jaccard from 0.218 to 0.237 on 1,004 heldout MIMIC IV admissions.
- The improvement required no additional LLM calls or inference-time patient index.
- A non-LLM top-k baseline achieved F1=0.248, showing the population prior is predictive before LLM integration.
- The approach targets fixed-vocabulary EHR set prediction and offers a more efficient, auditable retrieval source than patient nearest neighbors.
#ehr #medication prediction #retrieval augmented generation #structured prediction
Generated by gpt-5.6-sol. Check the original for details.
Aliases
| Source | External ID | URL |
|---|---|---|
| arXiv AI Papers | oai:arXiv.org:2611.17341v1 | https://arxiv.org/abs/2611.17341 Read discussion |