Towards Motivating Machines: Computational Modeling of the Mechanism of Actions for Effective Digital Health Behavior Change Applications

When Desiderius Erasmus said, “Prevention is better than cure,” the technologies to measure health and behavior data at any time were not omnipresent. Now, when those technologies are available, they are not fully utilized to influence human health behavior for preventing chronic diseases. Chronic diseases are the main threat to global health, with non-communicable diseases responsible for about two-thirds of deaths worldwide. Overcoming unhealthy behaviors like smoking, poor diet, physical inactivity, excessive alcohol consumption, etc., can reduce the risk of developing a chronic disease and increase the likelihood of a healthy life. Understanding and changing human behavior using digital technologies is complicated because many individual, societal, and policy factors are involved. Several theories and models explain the underlying processes of human behavior change. Still, new or updating of existing models are required to utilize the potential of modern technologies that can collect data at very detailed levels. More complex models can then be used for effective and intelligent systems that decide when, where, and how to intervene. Towards advancing the models and theories of human health behavior change, this thesis first provides a thorough insight into digital health behavior change research through bibliometric and scoping review. Second, it presents computational models of underlying determinants and processes of human behavior change based on cognitive, neuro, and behavioral sciences knowledge. For instance, how motivation is generated and regulated and how appraisal components work in an emotional episode. Lastly, it investigates the added value of using a computational model of the targeted determinants, e.g., a model of motivation as a mechanism of action in a digital health intervention. A proof of concept called MoM: Motivation to Move was designed as a digital health intervention to demonstrate its feasibility. A two-arm single-blinded study was conducted, and a special mobile app was developed to deliver an intervention. The model-based intervention has more encouraging results than the control group in personalization and efficacy. This thesis substantially contributes to the digital health behavior change field by identifying gaps, trends, and advancements in developing dynamic computational models and theories to define when and where behavioral change strategies would work effectively.