WT-LSTM for Intensive Care Unit Length-of-Stay Prediction with Real-Time Signal (Preprint)

• DOI: 10.2196/preprints.71247
• Publication Date: 2025-01-15T16:51:16Z
• Authors: Yiqun Jiang; Qing Li; Wenli Zhang
• Keywords: Preprint; SIGNAL (programming language)

<sec> <title>BACKGROUND</title> Efficient allocation of healthcare resources is essential for sustainable hospital operation. Effective intensive care unit (ICU) management alleviating the financial strain on systems. Accurate prediction length-of-stay in ICUs vital optimizing capacity planning and resource allocation, with challenge achieving early, real-time predictions. </sec> <title>OBJECTIVE</title> This study aims to develop a predictive model, namely WT-LSTM, ICU using only sign data. The model designed urgent settings where demographic historical patient data or lab results may be unavailable; leverages inputs deliver early accurate <title>METHODS</title> proposed integrates discrete wavelet transformation Long Short-Term Memory (LSTM) neural networks filter noise from patients’ series improve accuracy. Model performance was evaluated eICU database, focusing ten common admission diagnoses database. <title>RESULTS</title> demonstrate that WT-LSTM consistently outperforms baseline models, including linear regression, LSTM, BiLSTM, predicting data, significant improvements Mean Squared Error (MSE). Specifically, component enhances overall WT-LSTM. Removing this an average decrease 3.3% MSE; such phenomenon particularly pronounced specific cohorts. model's adaptability highlighted through predictions 3-hour, 6-hour, 12-hour, 24-hour input Using three hours delivers competitive across most diagnoses, often outperforming APACHE IV, leading outcome system currently implemented clinical practice. effectively captures patterns signs recorded during initial patient’s stay, making it promising tool optimization ICU. <title>CONCLUSIONS</title> Our based offers solution prediction. Its high accuracy capabilities hold potential enhancing practice, supporting critical administrative decisions management.