Close
Enter your
text here

Uncovering Racial and Ethnic Disparities in Pediatric Sleep Research

Uncovering Racial and Ethnic Disparities in Pediatric Sleep Research 1024 683 Erin Gregory

In a recent study published in the Journal of Biomedical Informatics, Mattina Davenport, PhD, principal investigator in the Center for Child Health Equity and Outcomes Research at the Abigail Wexner Research Institute at Nationwide Children’s Hospital, explores the impact of biases in clinical documentation on pediatric sleep research.

The Motivation Behind the Study

Pediatric sleep disorders affect children across all racial and ethnic groups, yet disparities in clinical documentation can lead to misdiagnosis or underdiagnosis, perpetuating health inequities. This study highlights the necessity of using unbiased data in training machine learning models to ensure accurate diagnoses and effective treatments for all children.

“The goal of the study is to create an algorithm that identifies patients at risk for sleep deficiency. To achieve this, assessing clinical documentation for racial and ethnic variations in sleep-related concerns was crucial,” says Dr. Davenport. As Electronic Health Records (EHRs) and Natural Language Processing (NLP) become more integral to health care, it is essential to address any biases in the data they rely on. This research aims to reveal how racial/ethnic differences in clinical notes might affect pediatric sleep studies and the models built from this data.

Analyzing Clinical Notes for Bias

Dr. Davenport and her team analyzed clinical notes from pediatric patients aged 5-18, collected between 2018 and 2021, using 178 keywords to examine the documentation of sleep issues. By applying LASSO regression and mixed-effects logistic regression models, they assessed racial and ethnic differences in keyword usage and their potential impact on model training.

The team identified significant racial and ethnic disparities in the documentation of sleep issues. Their study revealed that clinical notes for non-Hispanic white patients contained more specific keywords likely to prompt follow-up care, while notes for minoritized patients were characterized by broader, less specific descriptions. This pattern of documentation could result in the under-detection of sleep problems in minoritized groups, highlighting a bias in the data sources used for machine learning models. These disparities could skew the data used to train these models, potentially leading to inaccurate predictions and treatment recommendations.

“The most important step is not just changing documentation templates but introducing a standardized and multidimensional way of assessing every patient’s sleep health status,” says Dr. Davenport. The study advocates for more inclusive and representative data collection practices to improve the accuracy and fairness of healthcare outcomes.

Ensuring Fairness in Pediatric Sleep Research

“Sleep is a multidimensional construct,” says Dr. Davenport. “While a patient may report no concerns, many lack health literacy about what sleep truly entails. Pediatricians and other providers must understand the wide range of sleep-related issues, as this shapes how patients may express subclinical symptoms and compensatory behaviors, such as napping or drinking caffeinated beverages.”

This research highlights the importance of addressing biases in clinical documentation. By doing so, clinicians can ensure that machine learning models used in pediatric sleep research are accurate and equitable. This is crucial for improving the detection and treatment of sleep disorders across all racial and ethnic groups, ensuring that every child has the opportunity for better health outcomes.

 

Reference:

Davenport MA, Sirrianni JW, Chisolm DJ. Machine learning data sources in pediatric sleep research: assessing racial/ethnic differences in electronic health record-based clinical notes prior to model training. Front Sleep. 2024;3:1271167.

Image credit: Adobe Stock

About the author

You might also like