Unbiased inference for echocardiogram urgency prediction using double machine learning.

Area of Science:

• Cardiology
• Health Informatics
• Machine Learning

Background:

• Echocardiography is crucial for diagnosing cardiovascular conditions but faces challenges in patient prioritization due to limited test availability.
• Existing methods for assessing appointment urgency struggle with the complex interplay of clinical and administrative variables.

Purpose of the Study:

• To develop and evaluate a novel model for predicting patient urgency for echocardiography appointments.
• To leverage double machine learning techniques to accurately stratify patient urgency by disentangling variable relationships.

Main Methods:

• Utilized Electronic Health Record data, extracting both clinical and administrative variables.
• Applied double machine learning (DML) to model the urgency of echocardiography appointments.
• Compared the DML model's performance against traditional machine learning approaches.

Main Results:

• The developed double machine learning model significantly outperformed traditional methods in predicting appointment urgency.
• Identified administrative variables and cancer-related comorbidities as critical factors in patient prioritization.
• Provided robust estimations of variable effects, revealing complex interdependencies.

Conclusions:

• The double machine learning model enhances the efficiency and effectiveness of echocardiography utilization by improving patient prioritization.
• Actionable insights are provided for clinicians to identify urgent cases and optimize resource allocation.
• The methodology can be extended to prioritize other advanced, limited diagnostic tests.