First Published on 4 July 2023.

Background:

In the realm of healthcare outcomes, it is widely recognized that considering race and ethnicity (R/E) is crucial for quantitative studies. However, it’s important to continually strengthen our fundamental understanding of why R/E is important. This blog post aims to provide a comprehensive overview of the significance of R/E in model-building, summarizing conclusions from existing literature. Whether you’re a statistician, epidemiologist, data scientist, or simply interested in the topic, this article will help deepen your understanding of R/E’s role in healthcare outcomes.

Encoding R/E in Models:

Race and ethnicity are typically treated as nominal categorical variables in model-building. We discuss different methods for encoding these variables, including simple encoding and dummy encoding. Exploring these approaches and understanding their implications allows us to make meaningful comparisons and choose the most appropriate coding system for our goals. It’s crucial to avoid using R/E as a continuous variable, as it may lead to

Predictive Models:

This section focuses on R/E within predictive modeling.

We discuss the types of potential bias that can arise in “non-R/E-representative” datasets, such as representation bias and measurement bias. Drawing from examples in healthcare, we highlight the importance of training algorithms on representative datasets and the need for explicit caveats when dealing with non-representative data. Furthermore, we explore the controversy surrounding the inclusion of R/E as a predictor variable, considering arguments for and against its incorporation in predictive models.

Takeaway:

Remember, by continuously building our knowledge and engaging in discussions around this topic, we can contribute to the advancement of data-driven research that is inclusive and fosters equitable healthcare for all.