Towards Multidisciplinary Summarization of Hospital Stays: Efficient Sentence-Level Clinical Provenance Categorization
Baris Karacan, Vaibhav Bhargava, Barbara Di Eugenio, Natalie Parde, Catherine Craven, Karen Dunn Lopez, Andrew Boyd, Clinical Data Team · 2026
Read Towards Multidisciplinary Summarization of Hospital Stays: Efficient Sentence-Level Clinical Provenance Categorization by Baris Karacan, Vaibhav Bhargava, Barbara Di Eugenio, Natalie Parde, Catherine Craven, Karen Dunn Lopez, Andrew Boyd, Clinical Data Team (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-02487.txt.
Description
Karacan and co-authors work toward team-wide summaries of hospital stays for babies in intensive care. A neonatal stay spans doctors, nurses, and therapists. Their notes run into the hundreds. Pooling such mixed text yields incoherent output. A structured path needs each sentence tagged by its source discipline first.
The pilot study builds a categorization pipeline. Supervised fine-tuning adapts two Llama-3 models, at 8B and 70B sizes. Training uses MedSecId, a corpus of 2,002 adult ICU notes with provenance headers. Both models pass 92 percent Macro F1 in domain. Transfer is then tested on neonatal summaries. Fine-tuning lifts the 70B model by 7 points. The 8B model barely moves. A quantized fine-tuned 70B model beats its full-precision baseline. It also cuts compute needs sharply. Scale, the authors argue, preserves flexibility in cross-domain clinical transfer.
Catalog record
| CREATOR | Baris Karacan, Vaibhav Bhargava, Barbara Di Eugenio, Natalie Parde, Catherine Craven, Karen Dunn Lopez, Andrew Boyd, Clinical Data Team |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 12 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2606.02487