Measuring the Mind Through Play:Developing a Game-based Digital Biomarker of Acute and Chronic Stress
Year: 2026
Type: other
Abstract: Cognitive assessment holds considerable promise for predicting and understanding human health and behavior, yet the tools that deliver it tend to fall short of meeting this promise. Specifically, the computerized tasks that dominate cognitive assessment tend to bore, demotivate, and disengage participants, and their measurements are often unreliable and struggle to predict meaningful real-world outcomes. Video games are a natural remedy, given their capacity to draw sustained and self-driven cognitive effort. They therefore inspired the creation of cognitive assessment games that typically take a gamification approach, which adds game-like features to cognitive tasks while keeping their tight experimental control. However, the experiential benefits of this approach have been modest, because it neglects a central pillar of player experience in video games: players' sense of control. This tension between experimental control and player control motivates a different design strategy, which I call the inversion approach. Rather than adding game elements to a cognitive task, I start from a game and integrate cognitive assessment elements into it, balancing player and experimental control while measuring specific cognitive constructs. Building on this idea, my aim with this dissertation is to develop and evaluate the cognitive game Tunnel Runner as a game-based digital biomarker of meaningful real-world outcomes, with stress as its primary application domain. Stress suits this role well. It is prevalent and consequential, leaves a well-characterized cognitive signature, and can be both measured and experimentally induced. Its acute and chronic forms place different demands on measurement, where acute stress requires repeated cognitive assessment while chronic stress can be captured more cross-sectionally. In both cases, holistic assessment is required across multiple signals, among which cognitive performance is a promising if underutilized one. Developing and evaluating Tunnel Runner asks two things at once: that the system measure cognition well, and that its measurements detect real-world states through the cognitive signature they leave behind. I pursue this through an escalating, severity-matched evaluation framework. Each study builds on what the previous one established, and commits more resources to more demanding tests only once the easier tests have been passed. The dissertation follows this structure across three empirical Parts before closing with a general discussion. Part 1 begins by establishing the foundation. In chapter 2, I evaluated whether a cognitive game built on the inversion approach can measure cognition well while improving player experience. I found that Tunnel Runner efficiently and effectively captured multiple cognitive metrics, spanning interference control, response inhibition, response-rule switching, and decision-making under uncertainty, and it substantially improved players' motivation and engagement over cognitive tasks. It did so even though it asked players for more time and effort than either task alone. These results established the feasibility and value of the inversion approach: player freedom and effective cognitive assessment can coexist. Building on this foundation, in chapter 3 I asked whether I could enhance the data Tunnel Runner yields. Player freedom carries a visible cost, as the game evokes more incorrect initial responses that standard reaction-time measures discard. To address this, I examined whether the error-correcting responses that follow those mistakes carry cognitive signal. I found that error-correcting responses reflect the same cognitive processes as initial responses for interference control and response inhibition, though not for response-rule switching. Furthermore, I found that combining the two cognitively continuous response types can raise measurement reliability, and that error-correcting responses can measure response inhibition in their own right through a time-to-stop measure. Thus, the freedom that complicated assessment also opened a new behavioral channel for it. In chapter 4, I examined a limit of the approach: whether cognitive games serve infrequent video game players as well as frequent ones. In a large online study, infrequent gamers reported a worse experience across several facets, including mastery, ease of control, and frustration, and provided lower-quality data on the game's more behaviorally complex measures of response inhibition and response-rule switching. However, both groups were comparable on simpler measures and on facets such as focus, effort, and autonomy. Thus, the game-like freedom and complexity of the inversion approach may be most demanding for those who rarely play, which future designs will need to address. Part 2 escalates to a harder question: whether Tunnel Runner can track a transient state, acute stress, over repeated play. In chapter 5, I paired a controlled VR-based stress induction procedure with two play sessions, where the stress induction or a control procedure preceded the second play session. I found that cognitive performance improved considerably across sessions, indicating that players grew familiar with the game. Within this context, most of the stress-related impairments reported in single-session studies did not appear. The effect I expected to run the other way did: acute stress strongly improved response inhibition, whose main metric tracked stress at a level comparable to salivary cortisol, and combining the two signals detected stress better than cortisol alone. This shows that response inhibition is a promising metric for acute stress detection; however, as a single cognitive metric could be shaped by multiple factors, additional stress-sensitive metrics are required for real-world application. Alongside the cognitive question, in chapter 6 I asked whether players' motivation and engagement endure over repeated play and hold up under stress. I found that players' experience was largely resilient to a pre-game stressor, which is encouraging for a tool meant to be used when people are stressed. However, players' motivation and engagement declined substantially from the first session to the second, across curiosity, enjoyment, reward, meaning, immersion, and aesthetic appeal, while usability and frustration improved as players grew comfortable. Taken together, the findings that only one cognitive metric was stress-sensitive, and that players' experience substantially declined over repeated play, lead me to conclude that Tunnel Runner is not currently suitable for the monitoring of acute stress in real-world settings. In such settings, repeated assessment is needed, and multiple factors can affect a single cognitive indicator. Part 3 turns to stable states, which offer little within-person change and cannot be experimentally induced. In chapter 7 I first used chronological age as a proving ground for Tunnel Runner's ability to predict stable states, because it is a stable individual difference with strong cognitive correlates and a relatively unambiguous ground truth. Here the game succeeded: its cognitive metrics predicted players' age with high out-of-sample accuracy, without evident bias by gender or gamer identity. The resulting cognitive aging score aligned with external outcomes, showing greater cognitive differentiation after age 30, reduced cognitive aging in ADHD, and a negative association with gaming identity, while the depression result was inconclusive but consistent with the small differences found in neuroimaging work. This confirmed that Tunnel Runner captures stable and meaningful between-person differences. In chapter 8, I addressed the harder case of chronic stress detection, which left much to be desired. Using the same sample as in chapters 4 and 7 and self-reported chronic stress as ground truth, the game's cognitive metrics did not improve on a demographic baseline, and their predictions clustered in a narrow range. Baseline reaction time and the frequency of good decisions showed small, theoretically coherent associations, but response inhibition, the strongest marker of acute stress, showed no relationship with chronic stress. Because the game's reliability and its age-prediction success rule out a general failure to capture individual differences, this negative finding casts doubt on three assumptions behind the approach: that weak cognitive correlates of chronic stress are mainly due to poor cognitive assessment, that the cognitive profile of chronic stress mirrors that of acute stress, and that self-reported chronic stress is an adequate ground truth to be targeted by cognitive metrics. In chapter 9, the general discussion draws these escalating lines of evaluation together into a clear if asymmetric verdict. Tunnel Runner succeeds as a cognitive assessment system and as a promising biomarker of cognitive aging, and its response inhibition metric marks acute stress effectively. However, it is not yet a deployable digital biomarker of stress: only one cognitive metric detected acute stress, its appeal declined over repeated play, and its metrics did not enable chronic stress detection. Beyond these results, the dissertation makes several contributions toward future efforts at building game-based digital biomarkers: the inversion approach, the escalating severity-matched evaluation framework, a two-session paradigm for studying transient states, a modeling pipeline for extracting cognitive signal from game data, and large open datasets collected with a reusable instrument. Its main limitations, namely participant loss, self-selected and US-based samples, reliance on a single game, self-report ground truth, and declining appeal, mark where the achievements stop. Taken together, these findings, contributions, and limitations form a reusable blueprint for developing and evaluating game-based digital biomarkers.
Source: TU/e Research Portal
Author: Markovitch, Benyamin
Language: English
Subfield: Developmental and Educational Psychology
Field: Psychology
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7222852959
