April 06, 2020 – ONC is collaborating with NIH’s National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) to apply machine learning and artificial intelligence to patient-centered outcomes research (PCOR) on chronic kidney disease.
Through a project called Training Data for Machine Learning to Enhance PCOR Data Infrastructure (the PCOR Machine Learning Project), ONC and NIDDK are seeking to produce new scientific evidence to support the healthcare decisions of patients, their family members, and their providers.
“Through the PCOR Machine Learning Project, ONC and NIDDK aim to advance the application of AI and machine learning algorithms in PCOR by defining the requirements for high-quality training data sets,” Stephanie Garcia, MPH, wrote in a blog post.
“These data sets are essential to train prediction models that use machine learning algorithms, to extract features most relevant to specified research goals, and to reveal meaningful associations in the data.”
Current AI workflows make it possible to conduct complex studies and uncover deeper insights than traditional analytical methods, ONC noted. As the amount of digital data increases, PCOR researchers need better tools to analyze and interpret this data. Machine learning tools are a viable solution, and training data sets are essential for developing these algorithms.
The project is using chronic kidney disease as a sample use case, a condition that impacts 37 million Americans. Chronic kidney disease (CKD) involves the gradual loss of function in the kidneys over time, and can lead to high blood pressure, low blood count, weak bones, and an increased risk for heart disease.
Patients with early-stage CKD often don’t experience symptoms, but the disease can progress to end-stage kidney failure, which is deadly without routine dialysis or a kidney transplant. Millions of Americans are at higher risk of CKD, including people with diabetes, high blood pressure, and a family history of kidney failure.
The PCOR Machine Learning Project will identify which unanswered questions would benefit most from AI and machine learning applications using data from multiple sources. Current and planned project activities include capturing lessons learned from the process of developing high-quality training data sets focusing on data annotation, data curation, and data quality and quantity requirements.
The project will also focus on developing machine learning models and identifying approaches to evaluate model performance, as well as disseminating resources and materials to encourage future applications of these methods by PCOR researchers.
The PCOR Machine Learning Project has established a working group to bring federal partners together on a quarterly basis. The project is also leveraging expertise from leaders in biomedical, technology, and patient-advocacy organizations, as well as academic researchers and participating clinicians.
Both groups met for the first time in January 2020 and discussed use cases to consider, data sources, techniques for addressing health disparities and bias, and other relevant topics.
“Engagement and expertise from a variety of stakeholders are critical to understanding which kidney disease use cases would benefit most from AI, where relevant data are stored and how to retrieve them, and what the best practices are for training data and model development,” Garcia concluded.
“Throughout the next two years, the PCOR Machine Learning Project will help identify factors that can contribute to the development of robust training data sets and build a foundation for AI and machine learning approaches in PCOR.”