DeepTSE: A Time-Sensitive Deep Embedding of ICU Data for Patient Modeling and Missing Data Imputation

Area of Science:

• Biomedical Informatics
• Data Science
• Critical Care Medicine

Background:

• Missing data is prevalent in intensive care units (ICUs) due to various collection challenges.
• Incomplete data compromises the accuracy of statistical analyses and prognostic models.
• Existing imputation methods like mean/median do not consider data timeliness or heterogeneous time spans.

Purpose of the Study:

• To introduce DeepTSE, a deep learning model designed to address missing data and heterogeneous time spans in ICU datasets.
• To evaluate the performance of DeepTSE against established imputation techniques.

Main Methods:

• Development of a deep learning model named DeepTSE.
• Application and evaluation of DeepTSE on the MIMIC-IV dataset.
• Comparison of DeepTSE with traditional imputation methods.

Main Results:

• DeepTSE demonstrates proficiency in handling both missing data and heterogeneous time spans.
• The model achieves competitive and in some cases superior results compared to existing imputation methods.
• Promising performance was observed on the complex MIMIC-IV dataset.

Conclusions:

• DeepTSE offers a robust solution for imputation in high-frequency ICU data with missing values and varied record lengths.
• The model shows potential to improve the reliability of analyses and prognostic models in critical care settings.
• DeepTSE represents an advancement in handling complex data challenges in biomedical research.