A primer on quantitative bias analysis with positive predictive values in research using electronic health data

Area of Science:

• Health Informatics
• Epidemiology
• Biostatistics

Background:

• Concerns exist regarding bias in electronic health data reuse for research due to outcome misclassification.
• Positive predictive values (PPVs) are commonly used to validate outcome definitions but may not fully capture bias.
• Even minor misclassification can significantly impact research findings.

Purpose of the Study:

• To review predictive value-based quantitative bias analysis (QBA) methods.
• To explain how to quantify bias caused by outcome misclassification in electronic health data.
• To guide health informaticians and researchers in assessing data quality impacts.

Main Methods:

• Review of epidemiologic principles for misclassification bias.
• Introduction to two predictive value-based QBA methods.
• Demonstration using simulations in hypothetical electronic health record studies.
• Emphasis on stratifying PPVs by exposure for accurate bias assessment.

Main Results:

• Predictive value-based QBA quantifies bias from outcome misclassification.
• Factors influencing bias include misclassification levels, prevalence, and exposure-specific differences.
• Stratifying PPVs by exposure is crucial for robust bias assessment.

Conclusions:

• Predictive value-based QBA provides a quantitative approach to address data quality issues in health research.
• This tutorial bridges health informatics and epidemiology to improve the use of electronic health data.
• Accurate quantification of bias enhances the reliability of research using electronic health records.