Mitigating Bias in Opportunistic Screening for MACE with Causal Reasoning

Area of Science:

• Artificial Intelligence in Medicine
• Clinical Decision Support Systems
• Causal Inference

Background:

• Population drift significantly impacts AI model robustness in clinical settings.
• Existing bias mitigation strategies often neglect the influence of chronic comorbidities.
• Accurate AI predictions across diverse demographics are crucial for improving healthcare.

Purpose of the Study:

• To propose a causal reasoning framework addressing selection bias in AI models for predicting major adverse cardiovascular events (MACE).
• To evaluate the framework's effectiveness in diverse patient populations and clinical settings.
• To enhance the fairness and reliability of AI-driven clinical decision-making.

Main Methods:

• Developed a causal reasoning framework incorporating confounder adjustments.
• Trained an AI model on high-risk patient data and evaluated it on lower-risk and external datasets.
• Benchmarked the causal framework against traditional disease classification, propensity score matching, and debiasing models.

Main Results:

• The causal+confounder framework achieved superior Area Under the Curve (AUC) scores on shifted and external datasets compared to baseline models.
• The approach effectively minimized disparities related to confounding factors.
• Outperformed traditional and state-of-the-art debiasing methods in mitigating selection bias.

Conclusions:

• Integrating causal reasoning and confounder adjustments enhances AI model effectiveness in clinical applications.
• The proposed framework shows promise for creating fair and robust clinical decision support systems.
• This approach improves the reliability and ethical integrity of AI in healthcare by accounting for population shifts.