A Generalized Tool to Assess Algorithmic Fairness in Disease Phenotype Definitions

Area of Science:

• Health Informatics
• Epidemiology
• Biostatistics

Background:

• Reliable observational studies depend on accurate patient population definitions.
• Standardizing and implementing disease definitions across datasets presents significant challenges.
• Ensuring fair representation of diverse demographic dynamics within patient populations is essential to avoid overgeneralization.

Purpose of the Study:

• To present a generalized tool for assessing the fairness of disease definitions.
• To evaluate the implementation of disease definitions using common fairness metrics.
• To provide a robust method for examining population fairness across various characteristics.

Main Methods:

• Developed a generalized tool to evaluate disease definition fairness.
• Calculated fairness metrics to assess population representation.
• Examined coarse and strongly intersecting populations across multiple demographic characteristics.
• Utilized an OMOP Common Data Model (CDM) patient database for an example analysis.

Main Results:

• Demonstrated a method for assessing fairness in disease definition implementation.
• Highlighted practical workflows for applying the fairness assessment tool.
• Provided an example analysis showcasing the tool's utility on real-world data.

Conclusions:

• The developed tool offers a robust approach to evaluating the fairness of disease definitions.
• This work facilitates more equitable and accurate patient population representation in observational research.
• Future research can further refine the tool and expand its application to broader health data contexts.