Enhancing Clinical Decision Support in Nephrology: Addressing Algorithmic Bias Through Artificial Intelligence...

Area of Science:

• Health Informatics
• Medical Ethics
• Artificial Intelligence in Medicine

Background:

• Clinical decision support (CDS) tools are increasingly used in nephrology and general clinical practice.
• Concerns about algorithmic bias in CDS tools have risen, prompting investigation into the use of sensitive variables like race.

Purpose of the Study:

• To analyze sources of algorithmic bias in CDS tools developed using electronic health record data.
• To explore the complex question of whether sensitive variables, such as race, should be included in CDS tools.
• To highlight the role of health system governance in navigating these ethical considerations.

Main Methods:

• Identification and breakdown of three primary sources of bias: proxy variables, observability concerns, and underlying heterogeneity.
• Qualitative and quantitative analysis of the function of sensitive variables within CDS tools.
• Examination of institutional experience with CDS governance committees.

Main Results:

• Algorithmic bias in CDS tools can arise from proxy variables, observability issues, and data heterogeneity.
• Decisions regarding the inclusion of sensitive variables often depend more on qualitative factors than quantitative data.
• Health system governance committees are crucial for guiding decisions on sensitive variables in CDS.

Conclusions:

• A deeper understanding of algorithmic bias sources is necessary for developing equitable CDS tools.
• The inclusion of sensitive variables in CDS tools requires careful consideration of their specific function and potential impact.
• Fostering a community practice focused on sensitive variables and equity in model development and governance is essential.