Epigames Part I: What is an Epidemic Game?
This post marks the start of a series on experimental epidemic games, or “epigames” for short. Having recently published several papers with collaborators on this topic, I felt it was the right moment to provide a general introduction to the concept, offer some context around their development, and discuss potential next steps.
For reference, the publications I’m discussing include:
- An opinion piece in Nature Health titled “Understanding human behaviour for pandemic preparedness with epigames”.
- An epidemiological modeling research paper in Scientific Reports, “App-based epidemic game in a university campus reveals how risk perception and behavioral interventions shape disease transmission dynamics”.
- A behavioral analysis focused on belief-behavior mapping and cost incentives, “App-based Epidemic Game to Model Belief-Behavior Mapping and Cost Incentives in Voluntary Quarantine: A Randomized Controlled Trial”, currently available as a preprint on medRxiv while undergoing peer review.
While other papers and preprints will be linked throughout the series, this trio covers the conceptual foundation and application of epigames as a controlled experimentation approach to generate high-fidelity data useful for network, compartmental, and socio-behavioral epidemiological models.
The evolution of a concept through collaboration and discovery
This work is the culmination of long-term collaborations and reflects an evolution of ideas at the intersection of epidemiology, education, gaming, and mobile technology. While my initial work was informed by the Operation Outbreak project, I have since incorporated new elements from game theory, behavioral economics, social network science, and agent-based modeling.
Through this journey, I’ve come to view epigames as a research framework rather than a single tool. Different implementations can target specific problems, and I believe that the approach has applications beyond infectious disease epidemiology. In this first part of the series, I’ll define epigames within their original infectious disease context and share my perspective on how that definition gradually emerged. Later, in Part II, I will tackle the challenges of the approach, especially that of the external validity, ending in Part III with a look at how epigames could serve as a template for field experimentation in the social and behavioral sciences more broadly.
Defining the epigame
In our Nature Health piece, we defined epigames as “controlled situations in which participants join a simulated epidemic via a gamified smartphone app”. During the game, participants interact normally while the app uses Bluetooth signals to measure mutual proximity, contact duration, and social connection. As the simulation progresses, participants may become “infected” by a hypothetical pathogen, moving through Susceptible, Infectious, and Recovered (SIR) states.
These basic mechanics were first introduced with the Operation Outbreak app in 2018. At that time, the goal with the app was primarily to support “immersive” experiential learning simulations, with participants playing roles like “health responders” or “epidemiologists” while the digital pathogen spread via Bluetooth. However, my interest eventually shifted toward behavioral economics and social choice experiments, by realizing that the app could prompt participants to make critical decisions, such as whether to quarantine after exposure, in response to the simulated spread and the rewards or penalties built into the game.
To ensure these games remain unobtrusive yet meaningful, the idea was to introduce in-app choices that take only a few moments to be made by participants. This “light framing” would provide enough context to elicit realistic responses without biasing the participants, helping us capture real-world behavioral preferences even during long-lasting epigames. This is critical for external validity, a topic I will delve into further in Part II.
A third key element of epigames (which was precisely motivated by the need to compare attitudinal preferences with in-game behaviors and thus quantify external validity) consists in the collection of such preferences via survey questions. These surveys would be administered through the app at different points during the game, asking participants about their attitudes and beliefs regarding public health interventions, disease risk, perceived norms, and other behavioral factors.
Additionally, we can also leverage the sensing capabilities of smartphones to add additional layers of data, such as weather information, which influences contact patterns and pathogen transmission.
Taking all these elements together, we can characterize the epigame with the following “conceptual equation”:
The simulated pathogen provides the “ground truth” and mechanistic realism. Without that core factor, the data layers lose their context and the epigame model would not function. However, different layers could be added (or removed) depending on the specific needs and still remain as a “valid epigame”.
Epigames as a flexible research framework
As I mentioned at the outset, the work of several years, starting with Operation Outbreak, has led me to view epigames not as a single tool, but as a broad framework or template for experimental epidemiology using digital technology. This framework would be inherently adaptable, allowing it to be tailored to specific epidemiological questions and research designs, such as the one proposed in this protocol pre-print. While I plan to explore how this approach can be generalized even further for the socio-behavioral sciences in Part III of this series, for now, I will focus on its application within infectious disease epidemiology, where the “conceptual equation” above remains the primary guide.
The “Quarantine Game”: a hypothetical epigame study
In our recent publications, we described a hypothetical reference study called the “quarantine game” to demonstrate the flexibility of the epigame approach. In this model, physical proximity and contact network structure determine pathogen transmission, but participants are not asked to physically isolate, which would be unrealistic for a long-term study. Instead, they “quarantine” within the world of the game by selecting that behavior in the app.
To make this meaningful, the costs and benefits of the gamified quarantine must map to real-life consequences. For example, this could work through a point system where quarantining carries an immediate penalty (representing lost work or social opportunities), while getting “sick” later in the game carries a much steeper penalty (representing the health burden). To provide a tangible incentive, final scores could be linked to real-world rewards, such as gift cards, distributed via lottery.
The study protocol for such an experiment would follow distinct stages, like so:
- Recruitment: Participants install the app and complete baseline surveys.
- Calibration: Users complete surveys reframed in the context of the game, and the app calibrates Bluetooth signal strength (RSSI) to known proxemic patterns to ensure accurate distance estimation.
- Group assignment: Randomization of participants into experimental groups, which can be determined using methods such as graph-cluster randomization to minimize “cross-talk” between groups.
- Experimental manipulation: Groups are exposed to different variables, such as varying levels of information regarding peer compliance or local versus global infection risks.
- Data analysis: A mixed-methods approach evaluates how well the game data reflects real-world behavior before developing more complex models.
This is not prescriptive, but simply a theoretical reference for implementation. In fact, we have conducted a quarantine game already, which we pre-registered at OSF and whose results we reported in the recent medRxiv preprint listed at the beginning of this article. It has some differences with the reference quarantine game protocol (most notably, it did not include graph-cluster randomization and RSSI calibration), but it fits the protocol for the most part.
A prototype “tinkertoy” for research
In my own work, there is a clear duality between the “theoretical concept” of epigames and the actual Epigames app. This app represents an evolution of the proximity sensing and virtual disease simulations I first implemented for Operation Outbreak. Because of this common history, both the Operation Outbreak and Epigames apps share significant tech components and UI flows, from dependency on the Herald proximity sensing library to the use of digital avatars to inform participants of their “health” status and QR codes to simulate interventions such as masking, rapid testing, or vaccination.
However, these apps are distinct. While Operation Outbreak has been mainly focused on outbreak science education and biosecurity training, the Epigames app serves as a research “tinkertoy”, this is to say, a highly customized version of the original platform designed to test the new ideas that define the epigame concept. It is important to highlight that the concept of epigames is broader than any specific software. While the current Epigames app serves as a proof-of-concept, an epigame could be realized through a multiplicity of implementations — using wearables or even games that have nothing to do with “outbreaks,” such as multiplayer trust games for sexual health or citizen-science tools for sampling environmental data.
Exploring new ideas through rapid prototyping
The true value of a prototype tool such as the Epigames app lies in its ability to offer a set of modules that allow researchers to iterate and explore ideas quickly. For example, when I moved from using simple emojis to convey health status to more whimsical characters, I noticed that users expressed a recurrent interest in further customization (see the user survey listed at the end of this preprint article).
This observation led me to discover an entire body of work in media communications regarding emotional attachment to characters (see this article for just one example). Such psychological connection has profound implications for our central challenge: producing externally valid data. If a user feels a sense of “self-preservation” for their digital avatar, their in-game choices may be more likely to map accurately to their real-world behaviors — a topic I will explore in depth in Part II.
Ultimately, I look forward to the development of a robust “ecosystem” of digital tools for experimental epidemiology; tools that utilize meaningful gamification and simulated epidemic processes as a means to elicit truly realistic responses from epigame participants.
Historical context and the “mapping challenge”
It was important for me to recognize, both during my work on Operation Outbreak and the subsequent development of epigames, that these ideas do not exist in a vacuum. We were not the first to explore proximity sensing as a tool for measuring real-world social networks or to simulate the spread of a digital pathogen among human participants.
A legacy of participatory simulations
One of the earliest examples can be traced back to the 1990s with the Thinking Tags project at the MIT Media Lab. These “participatory simulations” allowed participants to become active agents in complex systems to build collaborative understanding, and one of their primary implementations was, fittingly, a viral outbreak simulation.
As mobile technology reached the consumer level, other researchers leveraged the availability of personal devices for participatory network science and epidemiology. A few examples of these initiatives are the following:
- FluPhone (2007-2012) and SocioPatterns (2008 — present): These projects utilized wearable RFID tags and early smartphones to measure contact networks at scale. SocioPatterns has produced a large number of network datasets since its introduction in 2008, used by the research community, while FluPhone specifically applied the technology to simulate digital pathogen transmission, very much like Operation Outbreak and Epigames now do (but with more advanced phones).
- Copenhagen Networks Study (2016): This study involved distributing loaner smartphones to a large student population to track interaction networks over several weeks, also resulting in a large dataset available for research purposes.
- Safe Blues (2021): More recently, this project used simulated outbreaks in a college campus to help forecast the real-world spread of COVID-19.
The lessons of “Corrupted Blood”
While proximity sensing provides the “how,” the concept of gamification provides the “why.” The idea of using games as proxies for real-world epidemiological behavior gained mainstream academic attention following the “Corrupted Blood” incident in the online game World of Warcraft. This accidental virtual plague suggested that virtual outbreaks could bridge the gap between static computer models and the unpredictable nature of human behavior in a controlled environment.
However, the “realism” observed in such incidents comes with a significant caveat. As communications researcher Dmitri Williams noted in 2010, many early researchers took the “mapping principle” on faith, assuming virtual behaviors would naturally translate to real-world contexts. Williams argued that we should be skeptical of this for several reasons:
- Misaligned risks: In a virtual (or game) world, the consequences of “death” are primarily social or ego-driven; there is no physical pain or permanent loss.
- Behavioral divergence: Because the stakes are lower, players in the Warcraft plague were often seen dancing, laughing, or intentionally trying to infect others — behaviors that clearly do not map to the life-or-death stakes of a real epidemic.
The core research question of epigames
It was the confluence of these elements — proximity sensing, epidemiological modeling, and the skepticism of the scientific community — that drove me to seek a more rigorous research framework. To make epidemic experiments via mobile apps useful for public health policy, we must answer a crucial research question:
What factors influence real-life mapping (or external validity) in epigames, and how can this mapping be increased?
This question of how to ensure that “game” behaviors provide meaningful data for real-world pandemic preparedness is the heart of the epigame project, and it is a topic we will revisit in much greater depth in Part II of this series.
The WKU and AUIB proof-of-concepts of a controlled experiment
As the concept of the epigame coalesced, we had the opportunity to run a large-scale simulation at Wenzhou-Kean University (WKU) in China, during the Fall semester of 2023. This was a pivotal moment in the project’s history, as it served as the development ground for the research-focused iteration of the app.
The WKU study was notable for its sheer scale and duration — nearly 1,000 participants engaging in a virtual outbreak for two weeks. While earlier Operation Outbreak simulations on U.S. campuses had involved hundreds of people, informing epidemiological modeling analyses, and the protocol at WKU indeed was built on those, this new study provided the depth of data needed for more rigorous epidemiological modeling by adjusting transmission parameters and overall research questions in advance. Perhaps the most novel element was the introduction of controllable incentives (point rewards), which established the foundation for the field experiments we conduct today.
Validating mechanistic realism
A primary goal of the WKU simulation was to validate whether a digital pathogen could truly replicate the dynamics of a biological epidemic. We parameterized the digital virus to match the known properties of SARS-CoV-2, but in a much shorter timeframe to have several infection waves in the course of two weeks, and analyzed the resulting transmission tree. Using the statistical framework introduced by Taube et al. to quantify superspreading, we found that the virtual dynamics closely mirrored reality:
- Reproduction number: We observed a marked decline in the effective reproduction number between the first and second halves of the simulation, consistent with real-world deceleration patterns in SARS and COVID-19.
- Dispersion parameter: The estimated dispersion parameter for the virtual pathogen was k = 0.35. This indicates significant transmission overdispersion, consistent with empirical estimates for diseases like SARS (k = 0.06) and COVID-19 (k = 0.14).
- Superspreader dyads: The ratio of observed to expected superspreader-superspreader dyads reached values as high as 19, comparable to the elevated ratios (>8) documented in actual COVID-19 transmission trees.
These findings proved that the epigame platform could replicate superspreading structures and realistic epidemic processes in a naturalistic university environment. Further analyses of Operation Outbreak and epigames datasets show that these experiments are also able capture the structural differences between networks measured in different settings (e.g., college campuses vs scientific conferences).
Advancing to rigorous randomized trials: the AUIB study
While WKU demonstrated feasibility and mechanistic realism, it was not a true intervention study because all participants faced the same incentive structures. We addressed this in our most recent work at the American University of Iraq — Baghdad (AUIB).
This study was designed as a pre-registered Randomized Controlled Trial (RCT) involving 567 participants. We focused on a specific public health intervention: the economic barrier to adopting voluntary quarantine. Participants were randomized into two groups:
- Group 1 (Low Barrier): Faced a nominal point penalty for choosing to quarantine.
- Group 2 (High Barrier): Faced a significant penalty (opportunity cost), which was determined by a preliminary “Willingness to Accept” (WTA) survey.
Key takeaways on external validity
The AUIB study provided the first systematic evidence for the external validity of epigames — specifically through indicator parallelism, or the extent to which in-game choices reflect real-life psychological drivers.
- Belief correlation: We found a moderate, statistically significant positive correlation between participants’ real-life health beliefs and their in-game beliefs regarding susceptibility, severity, and quarantine benefits.
- The “gatekeeper” effect: A Poisson regression model revealed that high economic barriers act as a “gatekeeper.” High costs significantly suppressed quarantine adoption among “skeptical” participants (those with low in-game motivation), while highly motivated individuals were still willing to pay the price to stay safe.
- Demographic drivers: Echoing real-world patterns, gender was a significant factor; female participants were significantly more likely to choose the quarantine option than males.
By combining the mechanistic realism of the WKU study with the rigorous experimental design of the AUIB trial, we have evidence that epigames are valid tools for network-aware behavioral epidemiology. They allow us to probe the “price” of safety and predict how populations might respond to public health interventions under varying economic and social constraints.
What epigames’ data would be useful for?
The examples above show that data generated by epigames can be analyzed using established agent-based modeling and network science techniques, as well as novel machine learning methods or GIS-aware agents. These tools allow researchers to characterize the diverse networks that form across various settings and identify the specific features that influence how they are constructed. Beyond theoretical analysis, epigames provide an adaptable mechanism for testing a wide range of interventions designed to reduce disease spread.
By implementing epigames with different incentive structures to “nudge” participants toward decisions that modify their susceptibility or transmission rates, researchers can test hypotheses regarding the individual perceptions and network factors that drive behavior. As mentioned earlier, these games can simulate choices such as wearing a mask, taking a diagnostic test, or receiving a vaccine, each with varying costs and benefits.
Network interventions and diffusion modeling
From a network intervention perspective, epigames are particularly well-suited to evaluate strategies like those proposed by Thomas Valente in his work on the diffusion of innovations. These include:
- Individual measures: Targeting messages to the most connected individuals in a network.
- Group-based strategies: Utilizing behavioral nudges that address all members of densely knit cliques.
- Induction approaches: Introducing “opinion leader” seeds to stimulate peer-to-peer diffusion of protective behaviors.
This connection to network theory is becoming increasingly formalized; for instance, the netdiffuseR R package has recently included support for epigame data structures and now features the WKU dataset for researchers to use in their own diffusion modeling.
A unified experimental framework
Ultimately, the epigames approach addresses existing gaps in epidemiological research by generating four interrelated data streams:
- High-resolution, real-life contact networks.
- Quantifiable behavioral data from in-game decisions, such as self-isolation or vaccination choices.
- Attitudinal data from integrated surveys.
- Environmental data (e.g., weather) that may influence interactions.
While this approach must account for limitations like the Hawthorne effect (behavior changes due to being studied) and sensor heterogeneity in smartphones and wearable devices, it represents a significant advance by creating a unified framework to study not only how a pathogen might spread, but why it spreads in specific patterns driven by human choice. High-resolution data from these experiments can lead to environmentally and behaviorally calibrated agent-based models (ABMs), enabling an unprecedented assessment of human behavior in pathogen transmission.
By building this infrastructure according to open and reproducible science standards, we can avoid the constant “reimplementation” of basic methods. This principled approach bridges epidemiological modeling with the biological, social, and ecological sciences to create a more data-driven future for pandemic preparedness.
To be continued…
While the individual pieces — proximity sensing, gamification, and modeling — aren’t necessarily new, the way we’ve combined them creates a systematic framework for experimental epidemiology that could result in studies with the potential to be extrapolated or mapped to real-life situations and generate actionable public and personal health insights. We have the preliminary data to show it works, but the story is far from over, and so it will continue in the next installment of this series. Stay tuned!
