Assessing Bias in EHR Data: Agent-Based Model of Geographic Disparities in CA-MRSA

Overview

This study used an agent-based model (ABM) to simulate health care–seeking behaviors for community-acquired methicillin-resistant Staphylococcus aureus (CA-MRSA) infections across California subregions. The model assumed uniform infection prevalence but found that geographic disparities observed in electronic health record (EHR) data were only partially reproduced, suggesting factors beyond bias and care-seeking behavior contribute to observed disparities.

Background

Health disparities, including geographic differences in disease prevalence, often arise from complex social, economic, and environmental factors. Differential access to health care and variations in health care–seeking behaviors can influence observed disparities, especially when using EHR data that only capture individuals who seek treatment. In California, EHR data from emergency departments (EDs) have revealed geographic disparities in CA-MRSA infections, but it is unclear if these reflect true prevalence differences or biases due to care-seeking patterns. Agent-based modeling offers a method to simulate these behaviors and assess their impact on observed disparities.

Data Highlights

The ABM simulated health care–seeking behavior assuming no true geographic variation in CA-MRSA prevalence. It reproduced observed prevalence in 9 of 21 geographic areas. However, the magnitude of simulated geographic differences was smaller than in empirical data, and spatial pattern agreement was low to moderate.

Key Findings

• The ABM assumed uniform CA-MRSA prevalence across regions but incorporated variable health care–seeking behaviors.
• Simulated prevalence matched observed EHR data in 9 out of 21 geographic subregions.
• Simulated geographic disparities were less pronounced than those observed in real EHR data.
• Spatial patterns of infection prevalence showed low to moderate concordance between simulated and observed data.
• Results suggest that geographic disparities in CA-MRSA prevalence are not solely due to bias from health care–seeking behavior or differential access.
• The ABM framework can be adapted to other health outcomes by modifying care-seeking parameters and disease progression.

Clinical Implications

Clinicians and public health professionals should recognize that observed geographic disparities in CA-MRSA from EHR data may reflect true differences in disease burden beyond biases related to health care access or utilization. Interventions targeting disparities should consider broader social and environmental determinants. Additionally, modeling approaches like ABM can help disentangle complex factors influencing observed health disparities.

Conclusion

The agent-based model indicates that while health care–seeking behavior contributes to observed geographic disparities in CA-MRSA, it does not fully explain them. Further research is needed to identify additional determinants driving these disparities.

References

1. Original Article -- Assessing Bias in Electronic Health Record Data: An Agent-Based Model Approach