Disaggregating Latino nativity in equity research using electronic health records

Area of Science:

• Health Services Research
• Health Equity
• Machine Learning Applications

Background:

• Health equity research requires data disaggregation to identify and address disparities.
• Determining Latino country of birth (nativity) is crucial for understanding health inequities.
• Existing methods lack the ability to infer nativity from electronic health records (EHRs).

Purpose of the Study:

• To develop and validate prediction models for inferring Latino nativity.
• To improve health equity research by enabling data disaggregation.
• To utilize EHR data, surname, and community composition for nativity inference.

Main Methods:

• Utilized EHR data from 19,985 Latino children across 456 community health centers (CHCs).
• Linked EHR data with census-tract geocoded neighborhood composition and surname data.
• Employed a machine learning framework (Super Learner) to construct and evaluate prediction models.

Main Results:

• Prediction models demonstrated outstanding external validation performance for nativity inference.
• Area Under the Curve (AUC) values were high: 0.90 (US-born vs. foreign), 0.89 (Mexican vs. non-Mexican), 0.95 (Guatemalan vs. non-Guatemalan), and 0.99 (Cuban vs. non-Cuban).
• 10.7% of Latino patients in the study reported non-US country of birth.

Conclusions:

• Developed and validated novel prediction models to infer Latino nativity from EHR data.
• These models represent a significant methodologic advance for health disparities research.
• The approach facilitates data disaggregation, crucial for informing health equity research in primary care.