JMIR Publications

ABSTRACT

Background:

Goal personalization features integrated into mobile Health (mHealth) apps have the potential to enhance physical activity as some evidence shows that the personalized goals generated by algorithms are more effective than default or fixed goals. However, it remains unclear whether goals set by users are more effective than fixed goals and which personalization strategy is more effective for different user segments.

Objective:

This field study aimed to evaluate (1) the efficacy of two step goal personalization strategies—personalized-by-you and personalized-by-the-algorithm—and (2) which strategy is more effective among users with different activity levels.

Methods:

All users of SamenGezond, a Dutch mobile health app, have a default goal of 2,000 steps per day, five days a week. For this study, two random groups were selected, totaling 5,800 users. Subsequently, an email was sent to 3,800 users in Group 1, asking whether they were satisfied with their current goal. Those who were not satisfied were offered two personalization options: to set a goal themselves or to have the algorithm integrated in the app set goals for them. In total, 1,399 users responded: 230 chose to set their own goals (Personalized-by-you group), 236 opt for setting the goal by the algorithm (Personalized-by-the-algorithm group), and 933 chose to keep the default goal (Not-changed group). The algorithm used a moving-window percentile-rank method based on step data from the previous four weeks. Users who did not personalize retained the default goal. The remaining 2,000 users in Group 2 did not receive the email and also retained the default goal. To evaluate the effectiveness of step goal personalization strategies, we employed propensity score matching and difference-in-difference analysis.

Results:

Users in the personalized-by-you group increased weekly step count by 3,793, while those in personalized-by-the-algorithm group increased by 4,315 steps, compared with no personalization group (users with default goals). The overall difference between the two interventions was not statistically significant (p > 0.5). Interestingly, users in the not-changed group also increased their weekly steps by 1,759. Furthermore, the effectiveness of each strategy varied by baseline activity level. The personalized-by-you strategy was effective for medium (increase of 5,842 steps) and high active users (increase of 4,266 steps), but not for low active users (increase of 384 steps, p = 0.815). Conversely, the personalized-by-the-algorithm intervention was effective for low (increase of 5,095 steps) and medium active users (increase of 5,278 steps), but not for high active users (increase of 1,446 steps, p = 0.512).

Conclusions:

Step goal personalization demonstrates short-term effectiveness. However, their impact varies by users' baseline activity levels, indicating the need for a tailored approach for different user segments. Future studies should examine the long-term effects of such interventions to design sustainable health behavior change strategies.