A Machine Learning-Based Clinical Decision Support System to Improve End-of-Life Care.

Area of Science:

• Healthcare Informatics
• Clinical Decision Support Systems
• Machine Learning in Medicine

Background:

• End-of-life care (EoLC) improves patient outcomes and reduces costs but is underutilized.
• Machine learning (ML) models can identify high-risk patients for timely EoLC interventions.

Purpose of the Study:

• To evaluate the impact of an ML-based mortality prediction model on EoLC provision within an academic health system.

Main Methods:

• A random forest ML model was integrated into a clinical decision support system.
• Interrupted time series analysis compared EoLC documentation rates before and after implementation.
• Outcomes included advance directives, palliative care consults, and DNAR orders for high-risk patients.

Main Results:

• Advance directive documentation improved post-implementation, but other EoLC measures did not.
• The ML model's predictive performance was significantly worse in production than in development.
• The system was withdrawn after 16 months due to lack of clinical benefit.

Conclusions:

• ML-based clinical decision support systems may not effectively improve EoLC provision.
• System-level factors, CDS design, and poor model performance can hinder successful implementation.