Digital Behavioral Biomarkers

 Quantifying cue-induced craving from implicit behavioural signals, without relying on self-report.

The measurement problem

Cue-induced craving is a central mechanism in maladaptive consumption behaviours and a strong proximal predictor of relapse. Yet outside of controlled laboratory settings, it is assessed almost exclusively through explicit self-reports, which are burdensome, slow, and biased by memory and social desirability. This limits the scalability of digital health tools and prevents the development of truly adaptive interventions that respond to craving in real time.

Our approach

We are developing the Motivational Salience Index (MSI), a digital behavioural biomarker that infers cue-induced craving implicitly from fine-grained behavioural patterns expressed during short, smartphone-based cognitive tasks. The core idea is that appetitive cues leave measurable traces in motor behaviour: when a food or substance image carries high motivational salience, it systematically modulates reaction times, error patterns, and response dynamics in ways that can be quantified without the participant ever being asked to rate their craving.

The MSI draws on three task families that our lab has studied extensively: Go/NoGo inhibitory control tasks, approach-avoidance stimulus-response compatibility tasks, and speeded forced-choice tasks. Each captures a distinct facet of motivational processing (inhibition cost, automatic approach tendency, relative incentive weighting), and their combination provides a multimodal behavioural signature of cue-specific craving. Because the measurement is embedded in brief, gamified interactions, it is low-friction and suitable for repeated, ecologically valid assessment.

Working hypothesis and preliminary evidence

Our working hypothesis is that implicit behavioural features, appropriately engineered and combined across task families, carry sufficient predictive signal to estimate cue-induced craving with clinically useful accuracy, reducing or eliminating the need for explicit self-report in deployed applications. We approach this as a machine-learning problem: behavioural features extracted from task performance are used to train supervised models that predict craving from implicit signals alone. Preliminary analyses on existing datasets from our lab (Najberg et al., 2021, 2023) support this hypothesis: task-derived features, particularly from the Go/NoGo paradigm, predict craving-related outcomes above chance and above permuted baselines across several model families, including regularised linear models, tree-based ensembles, and recurrent neural networks.

We are currently conducting a large-scale validation study (N = 1,000) at the University of Fribourg, to systematically identify the most informative behavioural features and converge toward a high-efficiency, low-burden measurement design through structured feature selection, model benchmarking, and cross-validation.

Translational context

A key property of the MSI is that it enables cue-specific closed-loop adaptivity: by continuously tracking which cues retain motivational salience and which lose it over the course of an intervention, the system can dynamically adjust treatment focus. This connects the biomarker work directly to our response training programme, where the MSI provides the objective monitoring layer needed for personalised, adaptive interventions.

The MSI is codeveloped and commercialised through BeweLab SA (bewe.com), a University of Fribourg spin-off . The technology is patented. Ongoing collaborations with insurers, clinical centres, and industry partners in the Canton of Vaud and beyond ensure that the biomarker is validated against real-world requirements. The solution is particularly relevant in the context of GLP-1 receptor agonist therapies, where objective monitoring of residual cue-induced craving is critical for relapse prevention.

Selected references from the lab

Najberg, H. et al. (2023). Effects of gamified Go/NoGo training on sugar-sweetened beverage consumption. Registered Report.

Najberg, H. et al. (2021). A gamified inhibition training platform for food cue devaluation. Registered Report.

Tapparel, M. et al. (in press). Sign-tracking bias moderates Go/NoGo training-induced food devaluation. Registered Report.