By Jillian Kunze
The way people feel about the COVID-19 pandemic has impacted how closely they have adhered to safety measures. During a minisymposium presentation at the 2021 SIAM Conference on Applications of Dynamical Systems, which is taking place virtually this week, Folashade Agusto of the University of Kansas presented a study that aimed to find the effect of public sentiments on disease risk and spread. “We used the model to gain insights into disease transmission, rather than forecasting,” Agusto said.
COVID-19 has caused over 167 million cases and almost four million deaths across the globe since its discovery in 2019. Many countries have implemented their own measures to curtail the disease’s spread, such as social distancing, school and event closings, and travel bans. Those infected with COVID-19 were often either required to quarantine or advised to self-quarantine. But how closely people adhered to these guidelines was impacted by the risks they personally perceived. Beliefs such as religions, cultural practices, and political affiliations impacted people’s perceptions, as did previous knowledge about the disease. Public perception, sentiments, and behavior were also informed by information gathered through media like television, radio, print media, and social medias such as Twitter and Facebook.
Figure 1. Sentiments on COVID-19 in from tweets made in Australia, the U.K., Italy, South Africa, Brazil, and the U.S. over time. Blue represents more positive tweets, and red represents more negative tweets.
To begin the study, Agusto and her research team first developed a baseline COVID-19 model that explored the movement of people between various categories such as “infected,” “quarantined,” “hospitalized,” and “recovered.” There were a number of possible pathways between these categories — for example, a person who was discharged from a hospital prematurely might move back into the infected category, rather than to the recovered category. There was also a pathway for individuals to violate quarantine and move back into the infected category, rather than recovering fully before leaving quarantine.
Figure 2. Lines fitted to the change in tweet sentiment over time, with higher values representing more positive sentiments.
The researchers used a next-generation matrix method to compute the reproduction number—a quantity that represents how much a disease is expected to spread through a population—related to the model system, using contributions from both symptomatic and asymptomatic individuals. Before looking further into the impact of sentiments on COVID-19, they used the model to investigate how different rates of quarantine and hospitalizations impacted the disease dynamics. If both the quarantine and hospitalization rates were doubled, for example, the infected population was smaller than in the baseline situation and infections took place over a longer timescale. But if the rates at which people violated quarantine and hospitals discharged patients were both doubled, the infected population grew larger and the timescale of infections became shorter.
Next, Agusto delved into how people’s sentiments influenced disease transmission. "Different factors drive people’s sentiments and perceptions of risks about the virus," she said. For example, someone that thinks the pandemic is a hoax is far less likely to take precautionary safety measures. To investigate these kinds of effects, the research team downloaded over 125 million tweets about COVID-19 posted on Twitter from six countries (Australia, the U.K., Italy, South Africa, Brazil, and the U.S.) during the earlier stages of the worldwide COVID-19 pandemic in the first half of 2020.
Figure 3. The impact of public sentiments on disease transmission. In the model, a situation with more positive sentiments has fewer infections that the baseline, while a situation with more negative sentiments has more infections.
The researchers used specific keywords to find the COVID-related tweets, then used a different set of keywords to determine which tweets had more positive or negative sentiments. A positive tweet might include messages about the importance and helpfulness of safety measures or offer uplifting encouragement for people to take personal responsibility, while a negative tweet might focus on harmful or scary aspects of the virus. While this data set only captures a section of the population—since not everyone has a Twitter account—it still provides a useful sample of public sentiment.
The resulting data on tweet sentiments was divided by country (see Figure 1). Positive sentiments were more common than negative in all six countries, though their relative prevalence differed. By fitting a straight line to the change in tweet sentiments over time, the researchers approximated the overall change in sentiments by country (see Figure 2). Out of all the countries whose data was included, the U.S. had the least positive sentiments for most of the time period considered, while the U.K. had the most positive sentiments the entire time.
Agusto and her collaborators next wanted to incorporate these results on public sentiments into their COVID-19 model. The mathematical function representing these sentiments would impact other parameters such as the rates of quarantine, hospitalization, and quarantine violation. To gain some insight into the interrelation of these quantities, the researchers performed a sensitivity analysis to determine what kind of impact the various model parameters had on the reproduction number. The infection and quarantine violation rates had a positive effect on the reproduction number—they made it larger, indicating a greater spread of disease—while the quarantine and early hospitalization rates had a negative impact. “This gave us an idea about the nature of the functions to define for these respective parameters,” Agusto said. The researchers segmented the functions into two sections representing the time before and after lockdowns were imposed, as both sentiments and parameters like quarantine rate changed during that time. The linear fits to the change in sentiments over time provided relevant values for the equations.
Figure 4. The plot on the left shows simulated COVID-19 caseloads based on the relative amounts of positive and negative sentiments in each country. The plot on the right shows that actual caseloads in each country. Though the two plots are not identical, the model that incorporated sentiments was able to at least mimic reality.
After establishing the model parameters and functions, the researchers were able to explore the impact of positive and negative sentiments on the spread of COVID-19. “Overall, negative sentiments led to more infections in the model,” Agusto said (see Figure 3). Positive sentiments, on the contrary, led to fewer infections that in the case without any sentiments. And though the model did not exactly parallel the infection rates that occurred in real life, it did mimic the relative caseloads that would occur in different countries (see Figure 4). Importantly, Agusto’s COVID-19 model demonstrated the influence that human behavior and public sentiments hold on disease transmission.
Acknowledgments: This research was supported by the National Science Foundation under grant DMS 2028297.
 |
Jillian Kunze is the associate editor of SIAM News. |