Identifying treatment heterogeneity in atrial fibrillation using a novel causal machine learning method

Area of Science:

• Cardiology
• Pharmacology
• Data Science

Background:

• Atrial fibrillation (AF) necessitates lifelong oral anticoagulation to prevent stroke.
• Several novel oral anticoagulants (OACs) are available, but subgroup-specific effectiveness is not well-understood.

Purpose of the Study:

• To investigate patient subgroups with differential treatment effects among OACs.
• To identify which OACs provide the best risk reduction for specific patient profiles.

Main Methods:

• Analysis of 34,569 patients with nonvalvular AF using claims and medical data.
• Application of causal machine learning to identify subgroups and compare OACs (apixaban, dabigatran, rivaroxaban, warfarin).
• Primary outcome: composite of ischemic stroke, intracranial hemorrhage, and all-cause mortality.

Main Results:

• Machine learning identified 9 subgroups with varying OAC benefits.
• Apixaban showed advantages in 7 subgroups compared to dabigatran or rivaroxaban.
• No subgroup favored warfarin; most dabigatran vs. warfarin users showed no preference.

Conclusions:

• OAC effects are heterogeneous across AF patient subgroups.
• Findings support personalized OAC selection based on patient characteristics.
• Further prospective studies are needed to confirm clinical impact.