The effect of Behavioral Factors and Intervention Strategies on Pathogen Transmission: Insights from a Two-Week Epidemic Game at Wenzhou-Kean University in China

1 Introduction

The advent of new digital technologies has significantly transformed infectious disease research, including both epidemiological data collection and modeling. In particular, mobile applications (apps) can leverage the sensing and communication capabilities of smartphones and wearable devices for generating large amounts of real-time data at multiple levels. During the COVID-19 pandemic, apps were used for participatory surveillance, population-level tracking, individual risk assessment, individual screening, digital contact tracing, and education [1]. Several years before the pandemic, we started working on a proximity-based app called Operation Outbreak (OO) that facilitates immersive engagement in “mock” outbreaks that take place in real-life settings, such as schools and conferences, enhancing our understanding of outbreak dynamics and response strategies thanks to its capacity to incorporate naturalistic human behavior and controlled transmission mechanics [2, 3]. The OO mobile application is freely accessible on Google and Apple app stores [4, 5], and it is supported by a scalable cloud backend and web-based administration and analytics tools. This app enables a digital infection that spreads through participants’ mobile phones via Bluetooth, triggering a “synthetic” outbreak that participants can respond realistically to by endowing the app with appropriate reward or incentive mechanisms to various participants’ actions. The early example of the “Corrupted Blood incident” virtual pandemic in the World of Warcraft massive multiplayer online role playing game (MMORPG) back in 2007 prompted epidemiologists [6] to consider MMORPGs and other virtual games as settings where individual human behavior in response to pathogen spread could be studied experimentally rather than via modeling assumptions, issues of external validity notwithstanding [7]. The OO app was initially motivated by infectious disease education and pandemic preparedness, as the original version of the app was designed to support an immersive educational experience in middle and high schools centered around public health, outbreak response, and societal roles during health emergencies [8]. While this continues to be one of the main thrusts of the project [12], here we used a customized version of the OO app and backend to more specifically target epidemiology experimentation and research [18], by generating valuable data that informs epidemiological models and providing insights into pathogen transmission dynamics [19, 20, 21, 22].

In this study, we conducted a proximity-based “experimental epidemic game” at the campus of Wenzhou-Kean University (WKU), a Chinese-American higher education institution jointly established by Wenzhou University and Kean University of New Jersey in the Zhejiang province in China [9], which we will call the WKU game (or experiment) for short for the remainder of the paper. The aims of the WKU game were to generate a comprehensive outbreak dataset, explore spread of the simulated pathogen, and assess behavioral responses within the naturalistic environment of a college campus given a ground-truth transmission model in the app. Planning began in early fall 2023, with strategic recruitment of WKU students as organizers and facilitators. Our primary objectives were to engage at least 1,000 students over a two-week observational period and to integrate features representing individual behavioral actions within the experimentation framework. Drawing on methodologies from past (proximity-based) epidemic games at institutions like Colorado Mesa University and Brigham-Young University [3], we reached high participation rates (nearly a quarter of the enrolled students) through a comprehensive campus information campaign and incentivization strategies. In preparation for this experiment, we advanced the underlying technology in the app, including the incorporation of the open-source Herald proximity library [10] and migration to the Flutter framework [11], enhancing the app’s compatibility, adaptability and scalability [3].

Epidemiological modeling, when integrated with human behavior, serves as a foundational tool for evaluating the impact of non-pharmaceutical interventions (NPIs) on disease dynamics [3, 19, 21, 24]. The WKU proximity-based epidemic game combined open-world data with an epidemiological framework to study pathogen spread within a largely closed population. This interdisciplinary approach aims to contribute valuable insights into epidemiological modeling and public health preparedness, focusing on behavioral responses and the effectiveness of interventions in containing and mitigating disease outbreaks [3, 20, 23, 25]. Researchers have used online behavioral experiments to try to measure and model the effect of individuals’ preferences and decisions on adoption of protective measures and disease dynamics [26, 27, 28]; however, such experiments often reduce complex human behavior to simple abstract decisions that may not resonate with participants’ real-world concerns or priorities and may fail to reflect the complex social contexts and disease exposure patterns that influence behavior in realistic environments. Our approach has the advantage of providing an open-world experimentation framework with high degree of mechanistic realism for the transmission processes and a naturalistic environment able to capture inter-individual variation in real-world social settings [18]. The idea of conducting real-world “simulations” or experiments to study disease transmission has been explored before by projects like FluPhone [13] in 2010, and more recently, SafeBlues [14]. Both used Bluetooth sensing in mobile phones to spread virtual pathogens through a network of participants, but thanks to advances in mobile technology over the last decade, we were able to construct a new platform with a significantly greater potential for customization and scale than any earlier projects we are aware of.

Numerous studies have examined the impact of NPIs on pathogen transmission, providing critical insights into how measures such as social distancing, mask-wearing, and isolation can alter the course of an epidemic [29, 30]. For instance, SIRS models have demonstrated that the attributes of a pathogen, particularly its basic reproduction number (R₀), significantly influence the long-term epidemiological outcomes of transmission reduction efforts [31, 32]. Pathogens with high R₀ values may exhibit only temporary reductions in transmission rates, with minimal impact on the epidemic’s overall dynamics [41]. Additionally, other studies have highlighted the importance of seasonality and immunity in shaping medium- to long-term epidemic dynamics. These studies suggest that intermittent application of NPIs may lead to delayed but potentially larger secondary transmission peaks, underscoring the complexity of public health responses. The interaction between NPIs and pathogen characteristics necessitates ongoing evaluation and adjustment of intervention strategies to effectively address unexpected outcomes [32, 33, 34, 35].

Incorporating behavioral factors into epidemiological models significantly enhances our understanding of disease dynamics by elucidating the complex interplay between human behavior and pathogen transmission [30, 36]. Traditional models often overlook the variability in human behavior, treating responses to outbreaks as static or uniform. However, this approach does not fully account for the adaptive nature of individual and collective actions in response to evolving risk perceptions and public health interventions. Our study tries to address these challenges by integrating behavior as a dynamic variable, where individuals adjust their adherence to NPIs based on the perceived severity of the outbreak and their personal risk assessment (that is influenced by real-time factors such as disease prevalence, peer influence, and the effectiveness of public health communication) [30, 36]. Using the dataset generated from the WKU game, our model not only captures the immediate impact of NPIs such as social distancing and mask-wearing but also considers the feedback loops between behavioral adaptations and disease dynamics [37, 38]. This approach provides a more realistic and nuanced understanding of pathogen transmission, reflecting the complex interplay between human actions and epidemiological outcomes [36, 37]. While our model advances the field by incorporating these dynamic behavioral factors, we acknowledge the ongoing challenge of fully capturing the spectrum of human behavior, which underscores the need for continuous refinement and development in this area of research [39, 40].

Our modeling framework integrates these insights by utilizing data-driven approaches to simulate pathogen transmission and assess the impact of various interventions, such as quarantine, isolation, and face masks [24]. By leveraging mobile app technology, our platform allows for a detailed examination of how individual decisions—such as mask wearing, social distancing, or isolating when symptomatic—affect transmission dynamics within the controlled environment of the game, in turn embedded in the open-world setting of a college campus. The models were calibrated using data from the WKU game, enabling a precise analysis of transmission dynamics and the effectiveness of different interventions. This approach not only enhances our understanding of disease dynamics but also underscores the importance of timely interventions and compliance with preventive measures in mitigating outbreaks.

This study focuses on establishing a foundation for comprehensive models that can guide public health policies and strategies to prevent and reduce infectious disease outbreaks. The data generated from the WKU game was analyzed using a conceptual epidemiological modeling approach that integrates human attitudes, behaviors, and NPIs to disentangle and quantify the impact of these factors on pathogen spread. By fitting the model to the empirical data from our experimental game, we validated its accuracy and identified key parameters driving transmission dynamics. Sensitivity analyses and contour plots further highlight the importance of timely interventions, demonstrating how individual behavior and NPIs can significantly reduce infection rates. This research not only advances our theoretical understanding of socio-epidemiological interactions but also offers practical guidance for managing infectious diseases across diverse settings.

The paper is organized as follows: Section 2 describes the compartmental model we employed to mathematically represent disease transmission during the WKU game, which includes a variable transmission rate to capture adaptive behavioral factors and strength of NPIs. This section also calibrates the model with data from the two-week long WKU game. Section 3 provides an examination of the model that focuses on the occurrence and stability of disease-free and endemic equilibria, as well as the possibility of other complex dynamics. Results on the impact of behavior and NPIs are also presented in this section. Section 4 includes global uncertainty and parameter sensitivity analyses to evaluate the model’s robustness and identify critical parameters influencing transmission dynamics. Section 5 presents the results of numerical simulations and insights gained from global parameter sensitivity studies. Finally, Section 6 summarizes the study’s findings and discusses the implications for public health strategies and future research paths.

2 Methods

2.1 Experimental design

We designed the WKU experimental epidemic game to study human behavior and NPIs during an infectious disease outbreak. Experimental games constitute a novel approach that has been applied in fields such as public health and climate science and it is gaining recognition as a valuable source of behavioral data for research [16, 17, 15]. The app spreads the digital pathogen using an underlying “ground-truth” transmission model that assigns a probability of infection to each proximity contact between a susceptible and an infectious user (refer to the appendix for details) and informs users of their simulated health status (asymptomatic, mild and severely ill, recovered, and deceased) via an animated avatar and changes in the color of the app’s user interface (UI). A key difference with prior experimental games is the open-world, naturalistic setting of our experiment, which allows for a highly immersive experience and enables students to adopt behaviors during our experiment such as socially distancing to reduce their chances of infection, very closely mirroring real-life.

Furthermore, we considered how to gamify health-related decision-making in the app in such a way that students would need to make choices in the experiment on a regular basis, mirroring conflict between individual and group benefits, just like in a real outbreak. Based on these considerations, we implemented a point-based system in which students could collect points by deciding to ‘quarantine / isolate’ or not at the beginning of each day and then use those points to purchase virtual masks and rapid diagnostic kits through a ‘shop’ feature within the app. Of course, it would not have been reasonable to ask students to quarantine/isolate by physically restricting their movement, instead, they quarantined/isolated by selecting a button in the app, which made their avatars invisible to nearby participants (both quarantine and isolation used the same mechanics, the difference in using one term of the other refers to whether the student was in an asymptomatic state or not at the moment of their selection.) The goal of the point system was to provide a quick mechanism for students to get points without distracting them from their daily school activities, but still offering a sense of personal investment in the experiment, as their points informed a school-wide leaderboard that was used to award prizes to the top-scoring students once the experiment ended. See Fig. 1 for pictures of the students using the app and some screen captures depicting its main screen and quarantine/isolate feature.

• Download figure
• Open in new tab

Figure 1:

Students engaging with the smartphone app simulating pathogen transmission and disease progression during the epidemic game at Wenzhou-Kean University (WKU) (top). Screenshots of the app, showing (from left to right) the home page with healthy and sick avatars, a pop-up message prompting students to choose whether to quarantine/isolate for the day, and the home page interface while in quarantine/isolation (bottom). These visuals illustrate the interactive features of the app, which are integral to collecting behavioral and epidemiological data during the WKU game.

We also planned and implemented a pre-experiment stage to engage a substantial portion of the student body, employing a multifaceted strategy across various communication platforms (see Fig. 2). This strategy included the design of a dedicated informational website, the formation of WeChat groups for seamless communication, and the distribution of both digital and physical flyers to effectively disseminate information. A pivotal component of this strategy was the use of a registration form managed by the local organization team, allowing for the systematic collection of student details and facilitating insights into participation trends through regular updates. Integration with WeChat further enhanced communication channels, enabling efficient handling of inquiries and fostering active engagement among participants; this approach to data collection has proven effective [42, 43, 44]. Moreover, incentives such as extra-curricular credits and a point-based system within the app were implemented with the aim of incentivize sustained involvement and fostered a competitive environment through the school-wide leaderboard and post-game rewards. Thanks to this comprehensive approach, over 1,000 students (a quarter of WKU’s enrollment in 2023) responded to be interested in the WKU game. As expected, a fraction of those students dropped eventually, resulting in N_total = 794 students participating in the WKU game. A separate manuscript currently in preparation delves into the methodological aspects of the WKU field behavioral experiment, providing in-depth details on how it was planned, promoted, and carried out by a team of WKU students [45].

• Download figure
• Open in new tab

Figure 2:

Materials used to promote the experimental epidemic game at WKU and facilitate student enrollment into the activity. Section of an informational micro-site about WKU game with links to the registration form and WeChat group (left). Manual page detailing the use of the app and the features specific to the WKU game (center), and flier describing the final prizes to be awarded to top-scoring students (right).

2.2 Statistical analysis

The data were, first of all, analyzed using a simple statistical approach to understand the patterns of the experiment’s dynamics. Fig. 3 (a) displays the average number of infections per hour, revealing specific times of day with heightened transmission activity. Fig. 3 (b) presents the daily total infections (brown line) along with a 7-day moving average (purple line), which smooths out short-term fluctuations to highlight longer-term trends. Fig. 3 (c) shows the estimated effective reproduction number (R_eff), with gray points representing the estimated values, error bars indicating the 95% confidence intervals, and a smoothed trend black line, offering insights into the transmission potential over time. The red dashed line at R_eff = 1 serves as a critical threshold, indicating whether the infection is spreading (if R_eff > 1) or declining (if R_eff < 1). We used the serial interval information for COVID-19 from China as a proxy for the serial interval of the app-based outbreak, properly scaled to account for the two-week duration of the experiment [46]. These data provide view of infection dynamics, peak transmission periods, and the impact of interventions on outbreak control.

• Download figure
• Open in new tab

Figure 3:

Extended time series analysis of infection dynamics during the WKU game (a) Hourly average number of infections, highlighting periods of heightened transmission activity throughout the day. (b) Daily total infections (brown line) accompanied by a 7-day moving average (purple line) to smooth short-term fluctuations and reveal longer-term trends. (c) Estimated effective reproduction number (R_eff), where gray points represent the estimated R_eff values, error bars denote the 95% confidence intervals, and the black line indicates the smoothed trend. The red dashed line at R_eff = 1 marks the critical threshold, above which the infection is spreading and below which it is declining.

Even though network epidemiology is not the methodological approach we followed in this manuscript, we conducted some basic statistical analyses on the contact network and transmission tree that resulted from the WKU game. First of all, these analyses revealed that a fraction of the participants (198) did not seemingly make any contacts with anybody else. Upon closer examination, we discovered that most of those participants were using a brand of Android smartphones that was not compatible with the Bluetooth library in the app. Therefore, we removed those 198 participants, and further removed 122 participants who had interacted for less than 5 minutes in total during the 14 days of the WKU game and 2 participants who formed an isolated dyad. We used the remaining N_network = 472 = 794 − 198 − 122 − 2 in all subsequent network analyses (see Fig A4). We applied Taube et al. [47] methodology that characterizes the superspreader epidemiology in the transmission tree of any real outbreak by calculating two tree statistics, the proportion of cases causing super spreading events and the dispersion parameter. In the case of the WKU data (Fig A4), we found that it closely mirrors patterns seen in outbreaks from biological pathogens (refer to the Appendix for details). This result provides support for the external validity of the data from WKU game.

Survey responses captured basic demographics of the participating students, and their perceptions of quarantine/isolation benefits and costs, adapted from H1N1 pandemic research [50]. Detailed daily counts of within-app quarantine, isolation, mask purchases, and rapid diagnostic test use provided a detailed view of decision-making during the epidemic game. Note that only participants who downloaded the app and registered were included in the analysis, as the app tracked interactions solely among active participants.

The appendix provides a summary of the survey responses, which were not used in our present analyses, but we are planning to incorporate in follow-up work to this manuscript. This comprehensive dataset is a valuable resource for epidemiological research, health policy evaluation, and public health, enhancing our understanding of outbreak dynamics and human behavioral responses [3, 48, 49, 51, 52], and it is available on a Zenodo public repository (see Declarations section for the repository URL.)

2.3 Conceptual model framework

We employed an SEIR (Susceptible-Exposed-Infectious-Recovered) model framework [53, 54] to investigate pathogen transmission dynamics within a controlled university environment, specifically WKU through the experimentation platform and the app; see full model in Fig. 4. Our model incorporates multiple transmission pathways, accounting for both asymptomatic and symptomatic exposure, and captures the quarantine/isolation statuses of individuals. The population is assumed to be constant in size and homogeneous in structure, assumptions that are justified given the two-week duration of the experiment during a class semester and the residential nature of the WKU campus. Beyond these simplifying assumptions, the model allows for variations in vulnerability and transmissibility, thus reflecting the complex and heterogeneous nature of real-world disease spread.

• Download figure
• Open in new tab

Figure 4:

Flow chart for the model defined by equation (1) for pathogen transmission dynamics within the WKU setting. Solid arrows indicate the transitions between compartments, with the associated per capita flow rates shown next to each arrow. The model partitions the total population into ten compartments: non-quarantine susceptible (S_nq), quarantine susceptible (S_q), non-quarantine exposed (E_ni), isolated exposed (E_i), non-isolated asymptomatically infected (A_ni), isolated asymptomatically infected (A_i), non-isolated symptomatically infected (I_ni), isolated symptomatically infected (I_i), recovered (R), and deceased (D) individuals. The state variables and parameters are defined in Table A1, and parameters values in Table A2.

A key feature of our model is its integration of human behavior as a dynamic factor that influences transmission rates. We assume that during the experimental game, participants’ adherence to NPIs—such as mask-wearing, social distancing, and isolation—is driven by their perception of infection risk, which evolves in response to changes in infection severity and prevalence within the simulated environment. This feedback loop between behavior and transmission dynamics is crucial for accurately modeling the spread of infectious diseases in real-world scenarios where human decisions play a pivotal role. It is important to distinguish this post-experiment SEIR model with human behavior feedback from the model used by the app to drive the transmission during the experiment. We fit the parameters of the former using only the data gathered from the experiment, and we set the “ground-truth” parameters of the latter beforehand to ensure that the experimental game would take place within the desired time and case-count bounds.

Out app platform provides a unique experimental environment that generates rich datasets, including detailed records of contact tracing, participant behaviors, quarantine/isolation status, infection events, and (simulated) disease outcomes. By leveraging these data, we were able to construct a model to quantify the impact of various intervention strategies, such as contact tracing protocols, quarantine/isolation measures, and NPIs, on pathogen transmission. The model’s flexibility allows us to represent different NPIs mathematically, capturing how these interventions collectively reduce transmission. Additionally, our framework considers the dynamic nature of behavior, where individuals can alter their adherence to NPIs based on perceived risks and evolving conditions within the proximity-based experimental game. This approach provides a comprehensive tool for assessing the effectiveness of public health interventions and informs strategies for outbreak management and pandemic preparedness. The insights gained from integrating the experimental data into our SEIR model not only advance theoretical understanding but also offer practical guidance for controlling infectious disease outbreaks in educational settings and beyond.

3 Results: Evaluating the impact of individual perception and NPI strength

3.1 Influence of individual perception and NPIs

Our experimental data results provide critical insights into the dynamics of pathogen transmission under varying levels of individual perception and NPIs during the WKU game. Fig. 5 illustrates the daily new cases across three distinct scenarios: a naive scenario with no interventions (ɛ = 0 and ζ = 0), a scenario driven by individual perception (only ɛ = 0), and a combined scenario incorporating both individual perceptions and NPIs. In the naive scenario, represented by the orange solid curve, a rapid increase in new cases is observed, highlighting the unmitigated spread of the pathogen in the absence of any intervention. This serves as a baseline, demonstrating the potential severity of an outbreak without any behavioral change or NPIs. The second scenario, depicted by the red dashed curve, models the impact of individual behavioral responses to the perceived risk of infection. Here, participants adjusted their behavior based on the evolving outbreak situation, such as by reducing contacts or increasing protective measures like mask-wearing. The substantial reduction in new cases in this scenario underscores the significant role that individual perception plays in controlling disease spread. This finding is consistent with previous research that emphasizes the importance of individual actions in epidemic management [55, 56]. The third scenario, represented by the green solid curve, combines individual perception with structured NPIs, such as quarantine and isolation. This scenario shows the most pronounced decrease in daily new cases, illustrating the enhanced effectiveness of combining adaptive human behavior with coordinated public health interventions. The grey curve with dots, representing simulated reported cases, aligns closely with the observed data, further validating the accuracy and predictive capability of our model.

• Download figure
• Open in new tab

Figure 5:

Daily new cases across three intervention scenarios during the WKU game represented by the black shadowed curve. The orange solid curve represents the naive scenario of the simulation with no interventions, showing a rapid increase in new cases. The red curve illustrates the scenario with individual perception based on behavior change, resulting in significant reduction in new cases. The green solid curve depicts the scenario combining individual reactions with NPIs such as quarantine and isolation, showing the most pronounced decrease in daily new cases. The grey curve with dots represents the simulated reported cases, aligned with official data, validating the accuracy of our model. Effect of intervention take some time (a couple of days) to show up in the data.

3.2 Evaluation of reporting ratios and model validation

Fig. 6 (a) & (b) presents the reporting ratios between observed cases and model estimates under two scenarios: (a) the combined scenarios of individual perception and NPIs and (b) the scenario with individual perception only. The relatively stable reporting ratio observed in both scenarios indicates that the model effectively captures the dynamics of reporting behavior and the efficacy of interventions. This stability is crucial for ensuring that the model accurately reflects real-world dynamics, thereby providing reliable predictions of outbreak progression. The consistent reporting ratios also underscore the effectiveness of combining individual behavioral responses with structured NPIs in mitigating pathogen spread and ensuring reliable reporting. These findings highlight the importance of coordinated interventions that integrate both human behavior and public health strategies to achieve optimal outcomes during an epidemic. The insights gained from these simulations results set the stage for a detailed sensitivity analysis, which will be discussed in the subsequent section. This analysis aims to identify key parameters that drive transmission dynamics and intervention effectiveness, offering further guidance for optimizing public health strategies during infectious disease outbreaks.

• Download figure
• Open in new tab

Figure 6:

Reporting ratio between observed cases and model estimates under the combined scenario of individual perception (IP) and non-pharmaceutical Intervention (NPIs) (a) and IP only (b), respectively. The relatively stable ratio indicates that the model accurately captures the dynamics of reporting behavior and intervention efficacy. This stability underscores the effectiveness of coordinated interventions in mitigating the pathogen’s spread and ensuring consistent reporting accuracy.

3.3 Sensitivity analysis

We conducted sensitivity analyses to assess the influence of our model’s parameters on the transmission trajectory. The analysis underlined the varying relevance of each parameter and offered a more profound understanding of the underlying dynamics. The findings emphasize the critical role of prompt intervention and adherence to preventive measures in reducing the severity of the outbreak.

3.3.1 Global sensitivity analysis

In this sub-section, we aim to provide a paradigm that takes into account both individuals’ perception of risk (which is modelled by ζ) and NPIs’ strength (which is modelled by ɛ), as well as a time-dependent reporting rate through sensitivity analysis, and also to determine how the parameters ζ and ɛ affect the pathogen spread [55]. Our analysis demonstrates that both ζ and ɛ play vital roles in shaping transmission patterns, although ζ shows greater magnitude. Specifically, we found that ζ exerts a more pronounced effect on transmission dynamics compared to ɛ. Small increases in ɛ or ζ could result in decreased transmission, as shown in Fig. 7 (a) & (b), but the impact is relatively low for ɛ compared to ζ, which causes substantial decreases in transmission, indicating a higher sensitivity to changes in this parameter. This result stresses the role of ɛ and ζ in determining the course of an epidemic and highlights the importance of tailored measures to effectively control outbreaks. Therefore, through sensitivity analysis, we have gained valuable insights into the relative influence of ɛ and ζ, enabling us to optimize intervention strategies and enhance our understanding of the underlying transmission mechanisms. This knowledge is crucial for informing public health policies and improving pandemic preparedness.

• Download figure
• Open in new tab

Figure 7:

Sensitivity analysis illustrating the effects of parameters ɛ (a) and ζ (b) on transmission dynamics. Our analysis reveals that while both parameters significantly influence the epidemic trajectory, ζ (which measures the intensity of individual response) has a notably stronger impact than ɛ (which measures the intervention strength), which indicates that modest increases in ɛ result in a measurable reduction in transmission, but even a slight increase in ζ leads to a much more substantial decrease in transmission rates. This highlights the critical role of ζ in effective pathogen control, surpassing the influence of ɛ.

3.3.2 Local sensitivity analysis

To delve deeper into our model’s dynamics and parameter effects, we conducted a sensitivity analysis using partial rank correlation coefficients (PRCCs) to assess how model parameters influence the overall transmission [62, 67]. Utilizing PRCCs with R₀ as the response function, we identified the most effective parameters influencing the spread of the pathogen. See Fig. 8(a) and (b). Our sensitivity analysis involved sampling 5,000 random instances from uniform distributions within the specified parameter ranges. Each instance was simulated to produce biological outcomes, and PRCCs were calculated between parameters and biological values. The PRCCs for R₀ highlighted that the transition rate of exposed non-isolated individuals to the infected non-isolated class, along with the probability of transmission per contact, significantly influence R₀. These findings emphasize the importance of targeted interventions to reduce interactions among susceptible and infected individuals and decrease the probability of transmission per contact. Even minor adjustments to these critical parameters could substantially impact the infection’s spread, underscoring the necessity for effective public health measures and timely interventions to control the outbreak.

• Download figure
• Open in new tab

Figure 8:

Local sensitivity analysis using (a) R₀ and (b) the attack rate as response functions, assessed through partial rank correlation coefficients (PRCCs). The analysis identifies the transition rate from exposed non-isolated individuals to the infected non-isolated class, along with the probability of transmission per contact, as the most influential parameters on both R₀ and the attack rate. These results emphasize the critical need for targeted interventions focusing on these key parameters to effectively control the outbreak and reduce transmission rates.

3.4 Numerical simulations

In this section, we conducted numerical simulations of the simplified model to evaluate the influence of key parameters on transmission dynamics. The contour plots illustrate the dependency of the basic reproduction number (R₀) on several critical parameters. Fig. 9 (a) shows the relationship between R₀, β (probability of transmission per contact), and θ (recovery rate of symptomatically infected individuals). The plot demonstrates that higher values of β result in an increase in R₀, while higher θ values lead to a decrease in R₀, underscoring the critical importance of reducing transmission and increasing recovery rates to effectively control pathogen spread. Fig. 9 (b) presents the relationship between R₀, δ (disease-induced death rate), and α (modification parameter for decreased infectiousness). The results show that increasing both δ and α significantly lowers R₀, highlighting the importance of interventions that reduce infectiousness and disease severity in mitigating outbreaks. Finally, Fig. 9 (c) depicts the relationship between R₀, τ (recovery rate of symptomatically infected individuals), and α (modification parameter for decreased infectiousness). The plot reveals that higher values of τ and α both contribute to a reduction in R₀, reinforcing the importance of rapid recovery and lowering infectiousness as effective strategies for outbreak control. These findings emphasize the critical role of reducing transmission probability, increasing recovery rates, and mitigating disease severity in controlling infectious disease outbreaks. While other parameters could also be explored in future simulations, we focus on β, θ, δ, α, and τ due to their significant influence on transmission dynamics in the model. These key parameters offer valuable insights into how targeted interventions can alter the course of an epidemic, demonstrating the robustness and applicability of our model to real-world public health scenarios.

• Download figure
• Open in new tab

Figure 9:

Contour plots of the basic reproduction number (R₀) for the model (1), as a function of (a) transmission rate (β₀) and recovery rate of symptomatically infected individuals (θ), (b) disease-induced death rate (δ) and modification parameter for decreased infectiousness (α), and (c) recovery rate of symptomatically infected individuals (τ) and modification parameter for decreased infectiousness (α), respectively. Parameter values used are provided in Table A2.

4 Discussion

The epidemic game conducted at WKU provided a unique opportunity to pilot our app-based experimentation platform during an open-world realistic “mock” outbreak, engaging nearly 1,000 students over two weeks. This large-scale proximity-based experiment integrated behavioral experiments and generated multimodal data, including contact and transmission patterns (see Figs. 3 and A1), highlighting the platform’s potential for interdisciplinary research. WKU students gained practical experience by participating in the experiment, while a subgroup of the students were actively involved in planning and organizing the activity around campus. The resulting data are available for instructors to use during classes and further epidemiological analyses. This data are particularly significant given its realistic representation of a real-life outbreak scenario in a university campus, thereby supporting the platform’s capacity to generate rich multimodal epidemiology datasets with external validity. [3, 22, 25].

• Download figure
• Open in new tab

Figure A1:

Time series plots depicting the progression of disease cases during the experiment period. The red line represents the hourly incidence data, providing a detailed view of the outbreak’s temporal dynamics. The purple line shows the cumulative number of cases, illustrating the overall growth of the outbreak over time. Specifically, panel (a) displays the hourly number of cases, capturing the shortterm fluctuations in disease transmission, while panel (b) presents the daily number of cases, offering a broader perspective on the outbreak’s progression. The four-day gap without cases corresponds to a weekend and two consecutive holidays, during which participants had limited interactions, reducing potential transmissions.

Our in-depth study used a robust modeling approach to investigate pathogen spread during the WKU game. We designed an effective model framework (Figs. 4 & A6) to assess intervention strategies including quarantine and isolation, incorporating human behavioral perspectives and non-pharmaceutical interventions (NPIs). We conducted detailed model simulations and sensitivity analyses using the data generated from the experiment. Model fitting and parameter estimates (Fig. A7) demonstrated that our flexible transmission rate model effectively describes the transmission dynamics observed in the experiment. The inclusion of human behavioral factors, such as individual adherence to preventative actions and the assessment of risk, greatly enhanced the model’s accuracy. The predicted parameters accurately depicted the transmission situations, highlighting the significance of prompt and synchronized interventions [3, 43, 56].

Our model represents a significant step forward in the integration of dynamic behavioral factors into epidemiological modeling. Unlike traditional models that often treat behavioral responses as static or oversimplified, our approach accounts for the fluidity of human behavior during an outbreak. Individuals in our model can adjust their adherence to NPIs such as social distancing, mask-wearing, isolation and quarantine based on real-time changes in disease severity and perceived personal risk [30, 36]. This dynamic response is critical for understanding the true impact of public health interventions and for predicting the course of an epidemic [37, 38].

However, while our model incorporates a more realistic representation of behavior, it does not fully capture the entire spectrum of potential responses. Behavioral heterogeneity, influenced by factors such as cultural norms, misinformation, and varying levels of trust in public health messages, presents a complex challenge for modeling [39, 40]. Our findings suggest that while the inclusion of dynamic behavior significantly improves the accuracy of epidemiological predictions, there is still a need for further refinement. Future models should aim to incorporate more granular data on behavioral responses, potentially integrating psychological and sociological insights to better capture the diversity of human actions in the face of an epidemic. This ongoing refinement will be crucial for developing models that can more effectively inform public health strategies and interventions, particularly in diverse and rapidly changing real-world settings [30, 36].

The model simulation results presented in Fig. 5 underscore the profound impact of individual behavior and NPIs on disease transmission dynamics. The naive scenario, which operates under the assumption of no interventions, illustrated a rapid and uncontrolled spread of the pathogen, emphasizing the critical need for timely interventions to prevent widespread transmission. In contrast, scenarios incorporating individual reactions to the outbreak demonstrated a significant reduction in new cases. The red dashed curve in Fig. 5 reflects this impact, where behavioral responses such as social distancing, mask-wearing, and self-isolation markedly decrease transmission rates. Human behavior plays a pivotal role in controlling the spread of infectious diseases, as individuals’ decisions to adopt preventive measures can substantially alter the trajectory of an epidemic [55, 56, 44]. The most pronounced decrease in daily new cases was observed in the scenario combining individual reactions with NPIs, depicted by the green solid curve. This scenario incorporates measures such as quarantine, isolation, and contact tracing alongside individual behavioral responses. The synergistic effect of these combined strategies results in a substantially flatter epidemic curve, demonstrating the efficacy of coordinated intervention efforts. The alignment of simulated reported cases with official data, shown by the grey curve with dots, further validates the model’s accuracy and predictive capability.

The analysis of the reporting ratio in Fig. 6 adds another layer of validation to our model. The stability of the reporting ratio when both individual reactions and NPIs were implemented indicates that the model accurately captures the dynamics of reporting behavior and the effectiveness of interventions. This consistency is crucial for effective outbreak management, ensuring reliable data for decision-making and policy formulation. Accurate and stable reporting is essential for monitoring the outbreak and adjusting intervention strategies in real-time. These findings emphasize the critical role of NPIs and individual behavior in managing the spread of infectious diseases. Timely and decisive interventions, coupled with public compliance, are key to controlling outbreaks. Public health strategies must prioritize raising awareness about preventive measures and ensuring public cooperation. Campaigns to educate the public on the importance of these measures can significantly enhance compliance and effectiveness of interventions. The robustness of our model, validated through these simulations and sensitivity analyses, further underscores its applicability in real-world epidemiological studies and public health planning. By integrating human behavior and NPIs, the model provides a comprehensive tool for predicting outbreak trajectories and designing effective intervention strategies. This integrated approach is essential for optimizing public health responses and improving pandemic preparedness [25, 65].

Our study aligns with previous research indicating that combined individual and populationlevel actions are crucial in mitigating the spread of infectious diseases [25, 55, 56]. The insights gained from this experiment provide valuable guidance for future outbreak management and emphasize the importance of coordinated, multifaceted intervention strategies [25, 69]. Furthermore, our model simulation results demonstrate that individual behavior and NPIs are indispensable in controlling disease transmission. The effectiveness of these measures in reducing new cases and maintaining stable reporting ratios highlights the importance of a comprehensive and coordinated public health response. These findings provide a robust foundation for developing strategies to manage future outbreaks, ensuring both public compliance and effective intervention implementation.

To further highlight the effectiveness of key parameters such as ɛ (which measures the efficacy of intervention) and ζ (which measures the intensity of behavioral response), global sensitivity analysis simulations were conducted and depicted in Fig. 7. Our findings showed that minor increases in ɛ and ζ can significantly reduce transmission, with ζ having a more noticeable impact. This emphasizes the significance of robust and intensive measures in containing the outbreak. In addition, local sensitivity analysis (see Fig. 8), which has R₀ as the response function, analyzed based on the PRCC, identified the transition rate of exposed non-isolated individuals to the infected non-isolated class (σ_n) and the probability of transmission per contact (β) as the most critical parameters impacting R₀. These variables have a major impact on basic reproduction numbers, indicating their significance in control. Furthermore, contour plot simulations provided valuable visual insights into the interactions between critical model parameters and their impact on the basic reproduction number, R₀. In Fig. 9 (a), the simulations highlight the relationship between β (probability of transmission per contact) and θ (recovery rate of symptomatically infected individuals), revealing that higher values of β significantly increase R₀, while increasing θ effectively reduces R₀. This underscores the importance of both limiting transmission rates and enhancing recovery efforts in mitigating the spread of the pathogen. Similarly, Fig. 9 (b) illustrates the sensitivity of R₀ to δ (disease-induced death rate) and α (modification parameter for decreased infectiousness). The simulations show that increases in both δ and α result in a notable reduction in R₀, highlighting the critical role of interventions aimed at reducing disease severity and infectiousness in controlling outbreaks. Lastly, Fig. 9 (c) depicts the relationship between R₀, τ (recovery rate of symptomatically infected individuals), and α. The results demonstrate that higher recovery rates and reduced infectiousness both contribute significantly to lowering R₀, emphasizing the importance of rapid treatment and measures aimed at reducing the spread of infection. These contour plot simulations reinforce the notion that strategic interventions targeting key parameters—such as transmission probability, recovery rates, and infectiousness—are essential for effective epidemic control. By focusing on reducing transmission and enhancing recovery efforts, public health strategies can be optimized to limit the spread of disease and lower the reproduction number [25, 69]. This analysis underscores the critical need for coordinated, data-driven interventions tailored to the specific dynamics of each outbreak. Further numerical simulations (presented in A8) highlight the significant influence of key epidemiological parameters—transmission rate (β₀), recovery rate (θ), and progression rate (γ)—on disease dynamics. Variations in β notably impact both asymptomatic and symptomatic infections, with a 50% increase leading to substantial rises in case numbers, underscoring the need for interventions like social distancing and mask-wearing to mitigate transmission. Similarly, enhancing recovery rates (θ) by 50% reduces the infectious period for asymptomatic individuals, while slowing the progression from asymptomatic to symptomatic stages (γ) helps to alleviate the burden of symptomatic cases. These findings emphasize the critical role of reducing transmission, speeding up recovery, and delaying disease progression in controlling outbreaks effectively.

Conclusion

In conclusion, this study elucidates the significance of our platform as a powerful tool for studying pathogen transmission and evaluating the impact of human behavior and NPIs on disease dynamics via an open-world experimental game approach. The integration of a flexible, time-varying transmission rate model in the analyses significantly enhances our understanding of how behavioral factors and NPIs influence the spread of infectious diseases. By accurately capturing these complex dynamics, our model provides critical insights into effective outbreak management strategies, underscoring the importance of timely, coordinated interventions and adherence to preventive measures. Our experimentation platform’s ability to generate realistic multimodal datasets—encompassing epidemiological, behavioral, and network data streams—could prove very valuable for public health planning and pandemic preparedness. The robustness and applicability of our model, validated through rigorous simulation and sensitivity analyses, highlight its potential as an important tool for guiding public health policies and optimizing response strategies in future outbreak scenarios. Future research will aim to refine this approach by incorporating additional data streams and elements such as age structure and varying interaction patterns, thereby advancing our understanding of disease dynamics. Coupling the model with a game-theoretic approach to analyze behavior during the experiments could further enrich our insights into outbreak management, ultimately contributing to more effective public health strategies and a deeper preparedness for future health crises.

Strengths

Our study on the experimental game using the epidemic app at WKU demonstrates significant strengths. The large-scale, real-life experiment, involving nearly 1,000 students over two weeks, provided a detailed dataset closely mirroring real-world outbreaks. This rich dataset enabled rigorous analysis of pathogen transmission dynamics and the effectiveness of intervention strategies. A key strength of this research is the integration of behavioral data with epidemiological modeling. Our experimentation platform facilitated the collection of multimodal data, including epidemiological, behavioral, and network streams, offering a comprehensive view of disease dynamics. Our model, which incorporates human behavior and NPIs, accurately reflected the transmission patterns observed during the experiment. The robustness of our approach is further supported by extensive simulations and sensitivity analyses, validating the model’s accuracy and its applicability to public health planning. Moreover, the platform’s ability to generate realistic, high-quality data underscores its potential as a valuable tool for enhancing outbreak response strategies and pandemic preparedness.

Limitations

Despite the strengths of our study, we must acknowledge several limitations. The assumption of homogeneous mixing within sub-populations may oversimplify real-world interactions, potentially affecting model accuracy in diverse populations. While time-varying parameters were incorporated to reflect adaptive behaviors, some parameters were assumed constant, which may not fully capture the dynamics of real-world epidemics. The use of an exponential distribution for transitions between epidemiological states simplifies the model but may not accurately represent actual waiting time distributions. Additionally, the model does not account for demographic variations, such as age/group structure and varying interaction patterns, which could provide deeper insights into disease dynamics and intervention effects. The findings from the WKU game, while valuable, may not be fully generalizable to other settings with different population dynamics and intervention strategies. Further experiments in diverse environments are needed to validate and refine the model, which is why our long-term plans would include extending the platform to support a wide range of field experiments to study transmissible diseases.

Lastly, using smartphones’ Bluetooth radio to detect proximal contacts between individuals in physical space can introduce missing data and inaccuracies. This is due to the inherent noise in the Bluetooth signal and technical issues such as varying hardware capabilities and incomplete adherence to the Bluetooth specs from smartphone makers, particularly on Android. We are currently working together with the developers of the Herald project to address these issues, and upcoming versions of the app will likely feature more accurate and robust contact detection.

Implication for Public Health

This study underscores the importance of integrating human behavioral dynamics and NPIs into epidemiological models to improve public health responses during disease outbreaks. The experiment conducted at WKU demonstrated how real-time behavioral changes, such as adherence to social distancing, isolation and quarantine measures, can significantly alter the trajectory of an epidemic. By incorporating these adaptive behaviors into our model, we were able to provide more accurate predictions of disease transmission and identify the most effective intervention strategies. This approach emphasizes the importance of public health policies that are both responsive and adaptive, taking into account the dynamic nature of human behavior in response to changing outbreak conditions.

Also, the platform’s ability to generate comprehensive, multimodal datasets offers a valuable tool for public health planning and pandemic preparedness (for similar outbreak data generator, see [22]). The data-driven insights from this study can inform the design of targeted interventions that consider not only the biological aspects of disease spread but also the complex social and behavioral factors that influence public compliance and the effectiveness of NPIs. As public health officials navigate the challenges of controlling infectious diseases, the findings from this research provide a robust framework for developing strategies that are both evidence-based and adaptable to the changing dynamics of human behavior and pathogen transmission.

Declarations

Funding

The development of the original Operation Outbreak app and backend infrastructure has been supported by the Gordon and Betty Moore Foundation (grants GBMF9125, GBMF9125.01, and GBMF11392). The customizations of the app and backend required for the WKU game were funded by a startup grant awarded by UMass Chan Medical School to AC.

Conflict of Interests

None.

Authors’ Contributions

SSM, WM: conceptualization, data analysis, results, figures, manuscript writing; TI: figures, manuscript editing; YD, MK: user interface and graphic design; AG, HH: software development; AW, DK: planning, manuscript revision; SAN, JGY: conceptualization; AC: project lead, conceptualization, planning, manuscript writing, and funding acquisition.