Agricultural science has made an enormous impact on production in the industry. From hybrid seeds to chemical inputs and genetic modification, researchers have steadily made improvements in productivity and yield, developed more hearty and reliable varieties of crops, and finely tuned methods to eradicate weeds, pests, and disease. This process hasn’t been flawless, of course. We are still trying to understand how to develop reliable drought resistant crop varieties, for example. But overall, I find that most farmers and industry advocates express a bit of frustration when the general public is skeptical or disapproving of the scientific process that informs modern farming practices.
Public health science suffers from a similar problem. Despite the careful attention to the scientific process, and the use of data and evidence to inform recommendations for behavior, the general public sometimes expresses disbelief or skepticism about the trustworthiness of our results and guidance.
Understanding the data
The COVID-19 pandemic, and the process for understanding it, has raised questions for many about how to respond, as well as which experts we should listen to.
I talked with two members of the University of Iowa’s COVID-19 modeling team, Drs. Joe Cavanaugh and Grant Brown, both faculty in the Department of Biostatistics at the College of Public Health, to understand how and why public health models, which are developed to help us predict the spread of a disease and respond, are an important strategy to use in the case of a new disease outbreak.
There are a number of different approaches to these models, depending on the end goal. Some can forecast based on only what is currently known, others can quantify the effect of different conditions or seasons on virus spread, and others apply various assumptions about, say, rates of social distancing or mask use, to compare different options for response.
Building effective models
Both researchers noted predictive models are one tool among many to understand a disease, along with other strategies such as lab science that helps us figure out the characteristics of the virus itself. But predictive modeling is important, as Dr. Brown says, because the outcomes of a disease, such as infection or deaths, “at a given point in time aren’t telling the full picture. For COVID-19 specifically, there’s a very large lag time between infection and symptoms (if they appear at all), and then again between a diagnosis and hospitalization or mortality.”
He goes on to say, “ignoring that lag time means being reactive instead of proactive.” In public health, we emphasize prevention. A proactive approach that helps us predict the next stages of a disease outbreak can help us stop further infections before they occur.
Building models requires identifying and applying variables to understand which are important and how they interact. Dr. Cavanaugh, who leads the UI team, says, “any variable that could be viewed as a marker for COVID-19 transmission activity or prevalence, including confirmed positive and negative test results, hospitalizations, and deaths,” is important. In addition, “any variable that could be viewed as a measurable risk factor for infection, transmission, disease severity, or mortality, including age, ethnicity, occupation, place of residence, and preexisting health conditions,” should be considered.
These models may take into account human behaviors, such as social distancing. Dr. Cavanaugh notes that quantifying social distancing on a large scale presents a “daunting challenge.” It’s very difficult to capture actual data on social distancing. He gives the example of someone in a rural community, who may travel long distances in a given day, but not interact with many people, as compared to someone in an urban area who takes a “quick trip to the grocery store.” The urban person may not go very far, but may come into contact with many more people.
Another way to think about modeling social distancing, according to Dr. Brown, would be through both “top down” and “bottom up” approaches. In the first, you can incorporate data about how people behave because of a new statewide order about size of gatherings or other restrictions. In the second, you can “explore the possible effects of social distancing under various assumptions, using simulation.” He says this can be an especially useful approach, “because it allows us to explore differing scenarios, such as reducing contact by 10%, 20%, or 50%.” Using that approach, we can learn the potential beneficial effect of our social distancing practices.
Ultimately, to effectively build these models, researchers have to figure out the most important features of the condition they’re attempting to understand. Dr. Cavanaugh notes, “For instance, we now have very sophisticated models for predicting the onset of cardiovascular disease. Many of us are aware of the most important risk and protective factors: age, sex, ethnicity, BMI, diet, blood pressure, cholesterol level, exercise, genetics, etc. However, it took years for cardiologists and epidemiologists to identify these factors.”
Both researchers agree that it takes interdisciplinary research and collaboration, sometimes over the course of years, to correctly identify and apply the variables that will result in the best models. The current pandemic is unique in many ways, and, as Dr. Cavanaugh says, “although we’re aware of some of the most prominent risk and protective factors pertaining to infection, transmission, and disease progression, we will invariably uncover others.” In other words, we’re learning a lot and applying it to develop the best models, but there will be new information that will help us finely tune our understanding of how the disease progresses.
Both UI researchers note that statistical models are not perfect representations of a complex disease outbreak, they are subject to revision and change as we learn more. But, as Dr. Cavanaugh points out, “a model that incorporates the most important features of the phenomenon, based on currently available biomedical, epidemiological, and biostatistical principles, provides our best mechanism for learning from the past and forecasting the future.”
Agricultural and public health science have the same central goal: to improve outcomes using data. In agriculture, this means developing reliable products that will ensure high yields in the face of variable environmental conditions. In public health, it means using data to develop the best possible prevention strategies and improve health outcomes in a variety of complex scenarios. In both cases, there is also debate about different approaches, how data are used, and how we should interpret the results. Regardless, I hope that we can collectively recognize and appreciate the work that goes into developing scientific consensus, and trust our scientific experts in both fields.