Information Adapted Machine Learning Models for Prediction in Clinical Workflow.

Area of Science:

• * Clinical informatics
• * Machine learning in healthcare
• * Predictive analytics

Background:

• * Electronic health records (EHRs) contain variable patient information.
• * Real-time prediction models often encounter patients with limited data.
• * Missing data presents a challenge for predictive model accuracy.

Purpose of the Study:

• * To assess the impact of missing data on real-time delirium prediction.
• * To evaluate prediction performance using models trained on incomplete datasets.
• * To compare models specifically designed for patients with missing data.

Main Methods:

• * Compared a model trained on missing data to the current delirium prediction model.
• * Simulated five test datasets with varying degrees of missing data.
• * Evaluated prediction accuracy against a complete data set using the same model.

Main Results:

• * A model trained for missing laboratory and nursing data outperformed the implemented model.
• * Models specifically adapted to available information showed improved prediction performance.
• * Combining procedure and demographic data yielded results closest to predictions from complete data.

Conclusions:

• * Continuous evaluation of real-time prediction models is essential.
• * Developing tailored models for specific data availability can enhance prediction accuracy.
• * Machine learning models need adaptation to handle real-world data variability.