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

TL;DR

This study assesses lightweight models, including fine-tuned LLMs and traditional text models, for predicting next-day discharge from clinical notes, emphasizing resource efficiency and practical performance.

Contribution

It compares 13 models, highlighting that simple TF-IDF with LGBM can outperform compact LLMs in clinical discharge prediction tasks.

Findings

TF-IDF with LGBM achieved F1-score of 0.47 and AUC-ROC of 0.80.

LoRA improved recall in DistilGPT-2, but transformer models underperformed overall.

Resource-efficient models can outperform compact LLMs in real-world clinical tasks.

Abstract

Timely discharge prediction is essential for optimizing bed turnover and resource allocation in elective spine surgery units. This study evaluates the feasibility of lightweight, fine-tuned large language models (LLMs) and traditional text-based models for predicting next-day discharge using postoperative clinical notes. We compared 13 models, including TF-IDF with XGBoost and LGBM, and compact LLMs (DistilGPT-2, Bio_ClinicalBERT) fine-tuned via LoRA. TF-IDF with LGBM achieved the best balance, with an F1-score of 0.47 for the discharge class, a recall of 0.51, and the highest AUC-ROC (0.80). While LoRA improved recall in DistilGPT2, overall transformer-based and generative models underperformed. These findings suggest interpretable, resource-efficient models may outperform compact LLMs in real-world, imbalanced clinical prediction tasks.