Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

StayLTC: A Cost-Effective Multimodal Framework for Hospital Length of Stay Forecasting

Sudeshna Jana, Manjira Sinha, Tirthankar Dasgupta

TL;DR

StayLTC addresses real-time hospital Length of Stay forecasting by fusing structured EHR data with unstructured clinical notes through continuous-time Liquid Time-Constant Networks. The framework encodes notes into dense embeddings, augments them with four Severity of Illness scores, and leverages a LTC-based time-series predictor for daily LOS forecasts. On the MIMIC-III dataset, StayLTC demonstrates strong accuracy, often surpassing traditional time-series models and transformer-based baselines, while maintaining a dramatically smaller computational footprint than time-series LLMs. The work highlights StayLTC’s potential for scalable, resource-efficient LOS forecasting and points to explainability and data quality improvements as promising directions for future research.

Abstract

Accurate prediction of Length of Stay (LOS) in hospitals is crucial for improving healthcare services, resource management, and cost efficiency. This paper presents StayLTC, a multimodal deep learning framework developed to forecast real-time hospital LOS using Liquid Time-Constant Networks (LTCs). LTCs, with their continuous-time recurrent dynamics, are evaluated against traditional models using structured data from Electronic Health Records (EHRs) and clinical notes. Our evaluation, conducted on the MIMIC-III dataset, demonstrated that LTCs significantly outperform most of the other time series models, offering enhanced accuracy, robustness, and efficiency in resource utilization. Additionally, LTCs demonstrate a comparable performance in LOS prediction compared to time series large language models, while requiring significantly less computational power and memory, underscoring their potential to advance Natural Language Processing (NLP) tasks in healthcare.

StayLTC: A Cost-Effective Multimodal Framework for Hospital Length of Stay Forecasting

TL;DR

StayLTC addresses real-time hospital Length of Stay forecasting by fusing structured EHR data with unstructured clinical notes through continuous-time Liquid Time-Constant Networks. The framework encodes notes into dense embeddings, augments them with four Severity of Illness scores, and leverages a LTC-based time-series predictor for daily LOS forecasts. On the MIMIC-III dataset, StayLTC demonstrates strong accuracy, often surpassing traditional time-series models and transformer-based baselines, while maintaining a dramatically smaller computational footprint than time-series LLMs. The work highlights StayLTC’s potential for scalable, resource-efficient LOS forecasting and points to explainability and data quality improvements as promising directions for future research.

Abstract

Accurate prediction of Length of Stay (LOS) in hospitals is crucial for improving healthcare services, resource management, and cost efficiency. This paper presents StayLTC, a multimodal deep learning framework developed to forecast real-time hospital LOS using Liquid Time-Constant Networks (LTCs). LTCs, with their continuous-time recurrent dynamics, are evaluated against traditional models using structured data from Electronic Health Records (EHRs) and clinical notes. Our evaluation, conducted on the MIMIC-III dataset, demonstrated that LTCs significantly outperform most of the other time series models, offering enhanced accuracy, robustness, and efficiency in resource utilization. Additionally, LTCs demonstrate a comparable performance in LOS prediction compared to time series large language models, while requiring significantly less computational power and memory, underscoring their potential to advance Natural Language Processing (NLP) tasks in healthcare.

Paper Structure

This paper contains 10 sections, 2 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Proposed StayLTC Framework
  3. Processing of Unstructured Clinical Notes
  4. Compute Severity of Illness (SOI) Scores
  5. LTC network for real time LOS prediction
  6. Results and Discussions
  7. Dataset
  8. Baseline Models
  9. Results
  10. Conclusion

Figures (3)

  • Figure 1: Schematic overview of the StayLTC framework for predicting LOS.
  • Figure 2: Predictions for a patient's real-time LOS using autoencoded health vectors and SOI scores: LTC vs. TIME-LLM
  • Figure 3: Predictions of real-time LOS for a Deceased patient using autoencoded health vectors and SOI scores: LTC vs. TIME-LLM