Resource-Conscious Modeling for Next- Day Discharge Prediction Using Clinical Notes
arXiv:2604.03498v1 Announce Type: new 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.
arXiv:2604.03498v1 Announce Type: new 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.
Executive Summary
This article explores the feasibility of using large language models (LLMs) and traditional text-based models for predicting next-day discharge in elective spine surgery units. The study evaluates 13 models, including TF-IDF with XGBoost and LGBM, and compact LLMs, with TF-IDF with LGBM achieving the best balance in terms of precision, recall, and AUC-ROC. The findings suggest that interpretable, resource-efficient models may outperform compact LLMs in real-world clinical prediction tasks. The study contributes to the development of more efficient and effective models for predicting discharge, which can optimize bed turnover and resource allocation. The results have practical implications for healthcare providers and policy implications for the allocation of resources in healthcare facilities.
Key Points
- ▸ TF-IDF with LGBM achieved the best balance in terms of precision, recall, and AUC-ROC
- ▸ Compact LLMs underperformed in real-world clinical prediction tasks
- ▸ Interpretable and resource-efficient models may outperform compact LLMs in clinical prediction tasks
Merits
Strength in Methodology
The study employed a comprehensive evaluation of 13 models, including traditional text-based models and compact LLMs, to determine their effectiveness in predicting next-day discharge.
Insight into Model Performance
The study provided valuable insights into the performance of different models, including the limitations of compact LLMs in real-world clinical prediction tasks.
Demerits
Limited Generalizability
The study was conducted on a specific dataset and may not be generalizable to other healthcare settings or populations.
Overreliance on AUC-ROC
The study relied heavily on AUC-ROC as a metric for evaluating model performance, which may not capture other important aspects of model performance, such as interpretability and resource efficiency.
Expert Commentary
While the study's findings are promising, it is essential to consider the limitations of the study, including the limited generalizability of the results and the overreliance on AUC-ROC as a metric for evaluating model performance. Future studies should aim to address these limitations and explore the potential of compact LLMs in real-world clinical prediction tasks. Additionally, the study's results highlight the importance of developing and implementing interpretable and resource-efficient models for predicting discharge, which can optimize bed turnover and resource allocation in healthcare facilities.
Recommendations
- ✓ Future studies should explore the potential of compact LLMs in real-world clinical prediction tasks, addressing the limitations of the current study.
- ✓ Developers should prioritize the development of more interpretable and resource-efficient models for predicting discharge, which can optimize bed turnover and resource allocation in healthcare facilities.
Sources
Original: arXiv - cs.AI