The Problem Across medicine, public health and environmental science, experts and instruments often provide evaluations or measurements of the same thing (e.g., disease diagnoses, medical image ratings or air quality readings). Decisions based on these ratings depend on how much the raters agree. Traditional agreement measures can give misleading results, which can lead to flawed conclusions, wasted resources and poor policy decisions. There is a critical need for agreement measures that are accurate, interpretable and work for modern datasets. The Approach To address this need, Dr. Hughes developed Neural Bayes Agreement (NBA), a new statistical method that: Uses a flexible model to represent agreement between multiple raters. Employs advanced Bayesian and neural network techniques to make the method usable with complex data. Works in diverse settings and is designed for use by researchers, clinicians, policy analysts and data scientists who need trustworthy agreement statistics. Improving the Science of Measuring Agreement "I create statistical tools that help scientists answer the question: ‘How much can we trust these measurements?’ Better agreement measures mean better science, better policy and better outcomes." - John P. Hughes, PhD Short Term Impact Provided a reliable, reproducible way to measure agreement in challenging real-world data. Demonstrated improved accuracy over widely used alternatives. Released as open-source R and Julia software for wide use. Longer Term Impact Improve the quality and trustworthiness of research across fields. Support more cost-effective decision-making Improve public trust by ensuring reported agreement reflects reality. Societal Impact Funding for this research will advance access to this new statistical method, allowing for the improved interpretation of complex health data in diverse settings. This work offers societal benefits in the following areas: For more information visit https://health.lehigh.edu/research-partners or email INRSRCH@lehigh.edu 14 Data Driven Innovation: Show how advanced statistics and machine learning can solve longstanding problems. Policy: Guide evidence-based policies in healthcare, public health and environmental protection. Education: Offer a compelling way of thinking about agreement that can improve statistical practice.
RkJQdWJsaXNoZXIy MTA0OTQ5OA==