Can public health experts make a forecast for the flu or another infectious disease as accurately as a computational model?
Thomas McAndrew, assistant professor, Department of Biostatistics and Health Data Science in the College of Health, has received an award from the National Science Foundation (NSF) to research and develop a novel approach to forecasting. His project, “IHBEM: Enhancing Influenza Forecasting Through an Integrated Platform for User-Generated Temporal Forecasts,” will help improve evidence-based public health decision making for infectious diseases.
The first goal of the project is to understand cultural norms and thought processes of public health practitioners, epidemiologists, and other infectious disease clinicians associated with infectious disease decision making. Which data and models are these experts using to make their decisions, and what challenges are they facing in these processes?
Second, McAndrew will be developing new software to be a central source of information to aid in decision-making. This technology is a human judgment platform constructed to systematically collect people’s opinions and forecasts about infectious disease in an unbiased way.
According to McAndrew, there will be both a public-facing, citizen science component, where members of the public can pose and answer questions about public health issues, as well as an internal side for public health experts.
When a user on the platform asks a question to the crowd, “they’ll know how many people offered forecasts, and they’re going to see each of the individualized rationales for them, as well as a summary about those rationales and a summary of the forecasts in the form of an ensemble,” McAndrew said.
McAndrew is extending the work of ensemble forecasting by adding a temporal component, looking at how accurately individuals make predictions continuously over time and weighing them accordingly.
The third part of the project will compare influenza forecasts generated by citizen scientists to state-of-the art statistical models, as part of the CDC’s FluSight Ensemble—an epidemic predictive initiative which incorporates forecasts from multiple models.
“Can people beat the machine?” McAndrew asked. “We have past work that shows that people are just as good at it. I’m fascinated to see what will happen.”
This project is an extension of McAndrew’s expertise in computational epidemiology and infectious disease forecasting. Specifically, McAndrew uses human judgment forecasts to augment epidemic models. The most typical way is asking people to make predictions about attributes of an epidemic, and then incorporating it into a model to make it perform better, he said. Improving health outcomes through the use of new technologies is at the heart of the College of Health’s work.
Partners in the four year-long project include College of Health professor Rochelle Frounfelker, who will lead the work to identify the characteristics of public health decision processes; and Shaun Truelove, an epidemiologist with Johns Hopkins Bloomberg School of Public Health.
“I hope that by improving evidence-based decision making,” he said, “we will reduce mortality and morbidity of infectious diseases. A second, important, goal for me is empowering public health officials to produce forecasts of similar performance to computational models, to show them that what is in their head and their decision processes are important and have a ton of value.”
