Duke AI Health Fellow
In the United States, one in 36 children is diagnosed with autism. Early detection of autism ensures timely access to intervention, and accurate screening is a critical first step to diagnosis and linkage to early services. Sophisticated computational tools, such as machine learning models, may increase autism screening accuracy and help reduce disparities in access to diagnosis and intervention.
Angel Huang, PhD, a Duke AI Health Data Science fellow and member of the Duke Autism Center of Excellence research team, is working hard to advance such computational models by taking advantage of information about a child’s medical history that can be gleaned from electronic health records. “Autism can be hard to diagnose early because there aren't a lot of specialists available, and autism is not super common,” Huang explains. “By using computer programs that can learn from data, we can develop a better way to screen for autism early on. These programs can look at all the information in electronic health records in an objective, consistent way.”
Huang’s goal is to create a tool based on a infant’s electronic health records that doctors can use to quickly identify infants and toddlers that may be likely to have a later diagnosis of autism. “Catching autism early is so important,” she says. “It allows families to get assessments, start interventions, and access resources during crucial development periods, and this early support leads to the best outcomes for autistic kids and reduces stress for families.”
Huang says that the Duke AI Health fellowship has taught her invaluable lessons about cross-disciplinary collaboration and instilled in her a passion for bridging the gaps between medicine and data science to advance human health. A key skill is translating clinical problems into tractable analytical projects by clearly defining the study sample, goal, and hypothesis. Huang’s current project explores whether the accuracy of predicting if an infant will later be diagnosed with autism can be improved by having the model learn from data on other similar conditions affecting brain development, such as language delay. “This could potentially allow clinicians to identify children with a higher likelihood for any developmental condition early, and get them to the right specialist,” she explains.
The model she and her mentor, Ben Goldstein, PhD, associate professor of bioinformatics and statistics and Duke Autism Center of Excellence investigator, have developed currently makes predictions using data based on procedures and other standard medical codes, but Huang says that doctor's visit notes could improve accuracy, so in the near term, she is eager to enhance the model with the notes.
“With new advances in computer language technology, a physician’s notes might teach our model things that official codes miss. Descriptions of a child's behaviors or concerns could help the model better predict a child’s likelihood of autism,” she says.
An important further step will be integrating these predictive models directly into electronic health records. “This would allow doctors to quickly check a child's probability of receiving a diagnosis of autism during routine visits,” she explains.
For Huang, the most rewarding part of working with the Duke Autism Center of Excellence is knowing that her efforts can make a real difference for families. “Knowing these projects have the potential for real-world implementation keeps me motivated,” she says. “Contributing to that effort alongside the amazing team at the Duke Autism Center of Excellence showed me the impact that data science can have. I'm grateful for the opportunity to learn both broad technical skills and how to guide research that drives positive change. My dream is that this research will allow autistic individuals to get assistance as early as possible, giving them the best shot at reaching their full potential. I'm driven to do this work because I want to have a tangible, positive impact on the autism community.”