Background
Modifying the choice architecture of behavioural contexts can facilitate health behaviour change, but existing evidence builds mostly on small-scale interventions limited in duration, targets, strategies, and settings. We evaluated the effectiveness of a one-year hybrid type 2 implementation-effectiveness trial aimed at promoting healthy eating and daily physical activity with subtle modifications to the choice architecture of heterogeneous worksites. The intervention was contextualised to and integrated into the routine operations of each worksite. Effectiveness was evaluated in a quasi-experimental pre-post design.

Methods
Intervention sites (n = 21) implemented a median of two (range 1–9) intervention strategies for healthy eating and one (range 1–5) for physical activity. Questionnaires pre (n = 1126) and post (n = 943) intervention surveyed employees’ behavioural patterns at work (food consumption: vegetables/roots, fruit/berries, nuts/almonds/seeds, sweet treats, fast food, water; physical activity: restorative movement, exercise equipment use, stair use). The post-intervention questionnaire also measured employees’ perception of and response to three intervention strategies: a packed lunch recipe campaign, a fruit crew-strategy, and movement prompts. Multi- and single-level regression models evaluated effectiveness, treating intervention as a continuous predictor formed of the site-specific dose (n intervention strategies employed) and mean quality (three-point rating per strategy halfway and at the end of the intervention) of implementation relevant to each outcome.

Results
Multinomial logistic regression models found the intervention significantly associated with a favourable change in employees’ fruit and berry consumption (interaction effect of time and implementation p = 0.006) and with an unfavourable change in sweet treat consumption (p = 0.048). The evidence was strongest for the finding concerning fruit/berry consumption—an outcome that sites with greater dose and quality of implementation targeted by using strategies that reduced the physical effort required to have fruit/berries at work and by covering multiple eating-related contexts at the worksite. The quality of implementation was positively associated with the perception of (p = 0.044) and response to (p = 0.017) the packed lunch recipes, and with response to the fruit crew-strategy (p < 0.001).

Conclusions
The results suggest that a contextualised, multicomponent choice architecture intervention can positively influence eating behaviour in diverse real-world settings over a one-year period, and that higher implementation quality can enhance intervention perception and response. However, outcomes may depend on the type of intervention strategies used and the extent of their delivery.