Bayesian Network Analysis of Intervention-Induced Physical Activity Behavior Change: Comparative Modeling Study Across Age, Education, and Activity Impairment Subgroups.

Background: Tailoring intervention content, such as those designed to improve physical activity (PA) behavior, can enhance effectiveness. Previous Bayesian network research showed that it might be relevant to tailor PA interventions based on demographic factors such as gender, revealing differences in determinants' roles between subpopulations. In order to optimize tailoring, one needs to understand the differences between subpopulations based on different characteristics. Building on this, this study examines age, education level, and PA impairment as key moderators, as these factors might influence PA engagement and intervention responsiveness. Older adults, for example, rely more on habitual behavior, lower-educated individuals may face challenges due to lower health literacy and socioeconomic inequalities, and individuals with PA impairment, defined as functional impairments restricting PA, may face unique barriers to PA. Understanding differences based on these factors is crucial for optimizing interventions and ensuring effectiveness across diverse populations.

Objective: This study investigates, by means of Bayesian networks, differences in PA intervention mechanisms of subpopulations based on age, education level, and PA impairment.

Methods: Subpopulation-specific subsets from an integrated dataset of 5 studies are analyzed, including demographics, experimental group assignment, and PA and sociocognitive measures at baseline, short term, and long term. The relevant subpopulations are defined based on age, education level, and PA impairment. For each subpopulation, a stable Bayesian network is estimated based on the corresponding subset of data by applying a bootstrap procedure and according to a confidence threshold, relevant paths of the model are visualized in order to find indications regarding subpopulation-specific intervention mechanisms.

Results: A comparison of subpopulation-specific models unveils similarities and differences with respect to determinants' roles in PA behavior change induced by interventions. Similar structures of determinants affect short-term PA, ultimately causing effects in the long term, where intention and habit are directly related to PA for most subpopulations. With respect to age-based differences, the interventions influence PA less via attitude cons and planning for older than younger people. Looking at the level of education, planning and intrinsic motivation are less influential for low-educated participants compared with high- or medium-educated participants, whereas more influence takes place through attitude pros for this low-educated group with respect to maintaining effects in the long term. Looking at PA impairments, apart from the findings that attitude pros and planning are more relevant in the pathway of change for people without impairment, a more interesting insight is that fewer determinants are directly influenced by the intervention within the group with PA impairment.

Conclusions: Intervention mechanisms in specific demographic groups have been rarely studied so far. Initial interpretations from the derived subpopulation models in this study unveil subpopulation-specific patterns of behavioral change, which enable better tailoring of intervention content to characteristics of the target population in order to induce or enhance effects.