Published May 20, 2020
Jacobs School of Medicine and Biomedical Sciences researchers Gabriel Anaya, MD, and Sarah G. Mullin are putting their statistical skills to use in real time in the fight against COVID-19.
Anaya, a trainee in the clinical informatics fellowship program, and Mullin, a student in the doctoral program in biomedical informatics — both mentored by Peter L. Elkin, MD, professor and chair of biomedical informatics — are part of a team that has created an application that epidemiologically models COVID-19 hospitalizations, intensive care cases, ventilator use and other factors to help Erie County health care providers prepare for the surge in COVID-19 cases.
The application has been put to use by another part of the team to inform local health officials about local trends related to COVID-19, as the researchers showed the degree to which Western New York has been protected from the novel coronavirus due to the social distancing that has been promoted by public health authorities.
In addition to the Erie County Department of Health, the UB team has been working in close collaboration with Kaleida Health, Erie County Medical Center, Catholic Health, UBMD Physicians’ Group, Independent Health Association and HEALTHeLink.
Even before the novel coronavirus reached Western New York, the team started doing research on various models that could prove beneficial.
“The University of Pennsylvania (UPenn) had a model that was more acceptable for us to modify the code and make it more geared toward Erie County,” Anaya says.
“Once we started working on this we got more specialized people involved, including Sarah and other people from the epidemiology department who assisted in developing these models,” Anaya adds.
“A lot of my work is using statistics in medicine,” says Mullin, who has a master’s degree in statistics from Ohio State University. “After I entered in mid-March, we came up with an initial iteration of the model in four or five days.”
The fluid nature of the pandemic meant that researchers had to adapt almost daily.
“We quickly transitioned from UPenn’s SIR model (susceptible, infected and recovered) to an SEIR model (susceptible, exposed, infected and recovered), which we thought was more precise based on the reality of people being exposed, a latent period of the virus and a proportion of the population getting infected,” Anaya says.
The first iteration comparing their model to that of UPenn’s was on March 23.
“Since then, we’ve had eight different modifications to the model to keep making it more precise, adding different compartments such as asymtomatic and pre-symptomatic,” Anaya says.
“Initially when we started in early March, the only data available to us was from China and Italy. As we got more information from the literature, the news and our own data, we were able to evolve the model to better match the reality of disease transmission and progression,” Mullin says.
Their evolving model was the result of gaining local data.
“Once we started getting cases in Erie County, we began extracting real data from our community to be able to better understand the initial parameters that would predict the model,” Anaya says. “We are now very far from the initial model we took from UPenn to a modified and fitted model with step-wise social distancing using real data from Erie County. We’re sort of fitting our model and adjusting parameters using the real data from COVID-19 patients at the hospitals.”
“The fact that Sarah and I had already worked on projects before really helped. We know our strengths and our weaknesses. That’s how we knew who we needed to call and who we needed to get involved to make this project successful,” adds Anaya, who has been working on the front lines with COVID-19 patients in hospitals to collect some of that data. “We have been lucky that we have a good relationship with the Erie County Department of Health. They’ve been able to share daily hospital census data or COVID data.”
The research has meant highly evolved team collaboration and a drive to make sure their predictions are accurate and up to date.
“I’ve worked pretty closely with the clinical informatics fellows throughout my career. I think that has enabled us to work fluidly together,” Mullin says. “Gabriel and I have pretty consistently worked together on the statistical models. I’ll derive and code them up, then he’ll check them and update the app. Then we’ll check with the analytical group and get feedback from the statistical fellows and the professors and fix them, and then they’ll go out.”
In addition to Elkin, they are working under the direction of Jacobs School faculty member Peter Winkelstein, MD, executive director of the Institute for Healthcare Informatics, clinical professor of pediatrics and chief medical officer at UBMD Physicians’ Group and Kaleida Health; along with School of Public Health and Health Professions faculty members Matthew R. Bonner, PhD, associate professor of epidemiology and environmental health, and Gregory E. Wilding, PhD, professor and chair of biostatistics.
Other trainees in the clinical informatics fellowship program who are helping with the model are Arlen B. Brickman, MD; Jinwei Hu, MD; and Brianne E. Mackenzie, MD. Jonathan C. Blaisure, a doctoral student in biomedical informatics, is also part of the team.