A team of University of Iowa faculty, led by Joseph Cavanaugh, PhD, professor and head of the Department of Biostatistics, is working with the Iowa Department of Public Health (IDPH) to analyze data and develop predictive models to help Iowa respond to the COVID-19 pandemic.
Other faculty involved in the COVID-19 modeling project are:
- Grant Brown, PhD, Assistant Professor of Biostatistics
- Aaron Miller, PhD, Assistant Professor of Epidemiology
- Jacob Oleson, PhD, Professor of Biostatistics and Director of the Center for Public Health Statistics
- Michael Pentella, PhD, Clinical Professor of Epidemiology and Director of the State Hygienic Laboratory
- Eli Perencevich, MD, MS, Professor of Internal Medicine and Epidemiology and Director of the Center for Access and Delivery Research and Evaluation
- Daniel Sewell, PhD, Assistant Professor of Biostatistics
What are the goals of the COVID-19 data modeling project?
The project is comprised of three phases. The first phase was to critique a frequently cited model developed by researchers at the Institute for Health Metrics and Evaluation (IHME) at the University of Washington. The IHME model provides both national and state-specific projections for hospitalizations, ICU and ventilator needs, and mortality.
The second phase was to develop Iowa-specific models using data from publicly available sources that provide similar projections. Within this phase, we have formulated three modeling frameworks based on different paradigms with complementary objectives. The first framework involves a short-term prediction model designed to make accurate one-week ahead forecasts, and to evaluate the plausibility of longer-term forecasts based on their conformability with current short-term trajectories. The second framework features a medium-term prediction model that provides the potential range of the outbreak, yielding estimates of disease characteristics and prediction bounds. The disease characteristics included in this model are latent and infectious periods of COVID-19 as well as the mortality rate. The third framework is based on a model designed to evaluate various policy changes related to nonpharmaceutical interventions such as increasing or relaxing social distancing measures or implementing universal face shields. This model provides projections of how we expect the disease to spread if social distancing and other mitigation measures are implemented or relaxed at various points in time.
These first two phases of the modeling project have been largely completed and the results have been shared with the Iowa Department of Public Health (IDPH). This work is based on publicly available data sources. The two reports that summarize the first two phases of the project were made publicly available by the IDPH on April 28 (links to reports available below).
The third phase of the project underway now is to refine these models using data prepared by the IDPH, so that the models are uniquely tailored to Iowa. We are also developing an application based on the third modeling framework that will allow IDPH to provide inputs reflecting various mitigation measures (e.g., social distancing, facemasks, face shields) to assess their future impact.
In general, what can we expect out of a data modeling effort like this one? What are the limitations of modeling?
Modeling the trajectories of outcomes related to an epidemic presents a daunting challenge, one that is complicated by the unprecedented nature of the COVID-19 pandemic, which is unlike anything we’ve experienced in modern times. The most comprehensive epidemic modeling approaches attempt to characterize the spatio-temporal dynamics of infectious individuals as they network and interact with susceptible individuals, thereby exposing many of them to the virus, and subsequently increasing the incidence of infection. Infectious individuals will either recover from the disease, and presumably develop an immunity, or will succumb to the disease. In either case, they are removed from the susceptible population. With COVID-19, the dynamics are exacerbated by how easily the virus is transmitted, and by the relatively high percentage of infected individuals who will require hospitalization.
There is a well-known adage among statisticians that “All models are wrong, but some are useful.” A model cannot possibly capture all of the complexities and nuances of any naturally occurring phenomenon. However, if it appropriately characterizes the most salient features, it can be successfully used for prediction, as well as for quantifying the uncertainty associated with prediction.
What are the specific inputs that will be considered in the models?
The data provided by the IDPH includes a wealth of information on those individuals who have been tested for COVID-19, including lab results and demographic profiles (e.g., age, race, county of residence). In addition, extensive information is available for those who test positive, including comorbidity profiles and outcomes for hospitalized patients. Every record is coded by date and location, which should help in characterizing the spread of the disease both regionally and temporally. Of course, all of the data are stripped of identifying information to protect the privacy of the subjects.
What are the advantages of using the data IDPH has provided to build an Iowa-specific model as opposed to using only data from publicly available sources?
As with any state, Iowa has distinctive and unique demographic, environmental, and cultural characteristics. These characteristics along with current state policies for social distancing and mitigation influence the manner in which the pandemic has impacted the state. The data provided by the IDPH will allow us to tailor our models more specifically to Iowa.
Having said this, for the second phase of the project, the models we have formulated that rely on publicly available data sources are indeed focused on Iowa. The resulting estimates and projections may therefore be more informative than state-based results from national models that rely on a “one size fits all” approach, and assume a high degree of commonality among the epidemic curves for each state.
Will the models developed be able to account for the evolving nature of the coronavirus pandemic? How long will the modeling effort continue?
Since COVID-19 is likely to remain a problem until a vaccine is developed, we are hoping to continually adapt our models so that they will provide projections after the first wave of the pandemic has passed. Obviously, the more data that is collected, the better we will be able to refine the models. Once the state has the capacity to test large numbers of individuals, including those who are asymptomatic, the resulting data should serve to substantively improve the models. At present, the prevalence of the disease in the general state population is unknown, because most of the individuals who are tested are symptomatic.
The agreement with the Iowa Department of Public Health includes restrictions on publishing results. How will this affect the work of the team?
The restrictions pertain to the third phase of the project, based on the data that has been provided by IDPH. Our team recognizes that the immediate public health needs of the state should take precedence over our interest in ultimately disseminating and publishing our modeling methodology. We do feel that our work will provide important contributions to the fields of biostatistics and infectious diseases epidemiology, but at present, this cannot be our main priority. Our main priority is to provide the state with objective statistical and scientific information that could potentially inform policy and practice.
COVID-19 Whitepapers prepared for Iowa Department of Public Health
April 8, 2020
Critique of the IHME Model for COVID-19 Projections
April 20, 2020
Summary of Initial Modeling Efforts – COVID-19 in Iowa