Risk assessment of ICU patients through deep learning technique: A big data approach.

Area of Science:

• Artificial Intelligence in Medicine
• Clinical Data Analysis
• Pharmacovigilance

Background:

• Intensive Care Unit (ICU) patients receive numerous medications, posing risks from infusion rates and dosages.
• Current methods for analyzing patient treatment histories often struggle with missing data and temporal complexities.
• Deep learning offers potential for continuous patient monitoring and optimizing treatment plans.

Purpose of the Study:

• To develop a novel deep learning model for predicting patient outcomes in critical care settings.
• To address challenges in analyzing heterogeneous, incomplete, and uneven patient prescription data.
• To improve the monitoring and adjustment of medication therapies for ICU patients.

Main Methods:

• Utilized Long Short-Term Memory Recurrent Neural Networks (LSTM-RNN) driven by heterogeneous medication events.
• Incorporated Regular Language Handling and Gaussian Cycle for processing noisy and incomplete prescription records.
• Employed a piece-based Gaussian cycle approach to manage missing data values.

Main Results:

• The proposed LSTM-RNN model effectively predicts patient outcomes using complex medication event data.
• The model demonstrates robustness in handling various data quality issues, including missing and heterogeneous information.
• Semantic relevance of medication events and drug event grouping were emphasized for accurate analysis.

Conclusions:

• LSTM-RNN and variants like Phased LSTM-RNN are suitable for analyzing temporal medication data and attributing treatment effects.
• The bit-based Gaussian cycle effectively addresses data deficiencies in patient prescription records.
• This approach enhances the ability to monitor and adapt treatment plans for ICU patients.