Work

Resource-Conscious Modeling for Next-Day Discharge Prediction Using Clinical Notes

Year: 2026

Type: book-chapter

Abstract: Timely discharge prediction is important for surgical unit operations. Using 3,928 postoperative patient notes (32.2% positive), we compared TF-IDF models, sentence embeddings, and LoRA-fine-tuned small language models (SLMs). TF-IDF with LightGBM performed best (F11 = 0.47, AUC = 0.80), while embeddings and SLMs showed limited sensitivity, with only modest recall gains from LoRA. Overall, lightweight and interpretable text models outperformed transformer-based approaches in this imbalanced next-day discharge prediction task.

Source: Studies in health technology and informatics

DOI: 10.3233/shti260335

Authors: Ha Na Cho, Sairam Sutari, Alexander Lopez, Hansen Bow, Kai Zheng

Institution: University of California, Irvine

Language: English

Topic: Machine Learning in Healthcare

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences

SDG: Good health and well-being

Open Access status: hybrid

OpenAlex ID: W7161968795

FWCI:

Cites:

Cited by:

Related to: