Detecting clinician implicit biases in diagnoses using proximal causal inference.

Area of Science:

• Health equity research
• Causal inference in medicine
• Big data analytics in healthcare

Background:

• Implicit biases in healthcare professionals, stemming from stereotypes like racism and sexism, contribute to systemic discrimination.
• Existing bias measurement methods are limited to individual attitudes and controlled settings, not real-world patient outcomes.
• Large-scale electronic health records (EHRs) and biobanks offer new opportunities to study health outcome disparities.

Purpose of the Study:

• To develop and apply a causal inference approach for detecting the impact of clinician implicit biases on patient outcomes.
• To leverage large observational medical datasets for a more comprehensive understanding of bias effects.
• To provide a tool for raising awareness about unequal health outcomes due to implicit biases.

Main Methods:

• Utilizing a causal inference framework with proximal mediation analysis.
• Analyzing large-scale, real-world observational medical data, specifically from the UK Biobank.
• Disentangling pathway-specific effects of patient sociodemographic attributes on clinician diagnostic decisions.

Main Results:

• The proposed method successfully detected the influence of clinician implicit biases on patient health outcomes within the UK Biobank data.
• Demonstrated the capability of causal inference to uncover subtle biases in large datasets.
• Provided evidence of how implicit biases can lead to disparities in patient care and diagnosis.

Conclusions:

• Causal inference methods applied to big data are effective tools for identifying the real-world impact of implicit biases in healthcare.
• This approach can highlight systemic discrimination and its contribution to unequal health outcomes.
• Findings underscore the need for interventions to mitigate implicit bias and promote health equity.