This study followed a quasi-experimental 1-group pretest-posttest design in which the participants were followed up for 3 months via the app using continuous app log data and questionnaires conducted at baseline, midway through (approximately 45 days since the baseline), and at the end of the follow-up (approximately 90 days since the baseline). The effects and effect modifiers of the app were examined based on an exploratory analysis of the relationship between objectively measured app use and changes in self-reported behavioral outcomes. Owing to the correlational nature of the assessment, the reporting of the study follows the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) guideline for cohort studies [26,27] (for a completed checklist, see Checklist 1).

The research plan of the study was approved by the Institutional Review Board of the Finnish Institute for Health and Welfare (THL/5335/6.02.01/2022). The study participants gave their informed consent to participate. Participation was free of charge, and participants were not compensated.

This study was a substudy of a population-based health survey, Healthy Finland, that targeted a randomly selected and nationally representative sample of adults in Finland [28]. Participants of the Healthy Finland survey were eligible for this study if they (1) spoke Finnish fluently, (2) were aged 20 to 74 years, (3) had completed the questionnaire of the Healthy Finland survey between September and the end of the year 2022, and (4) were not randomly selected for an additional health examination component of the Healthy Finland survey. The latter criterion was to avoid influencing the health examination results and to avoid burdening the participants excessively. Participation in this study also required access to the internet via a smartphone, tablet, or computer.

Among the eligible population (N=10,207), individuals with a known phone number (n=4978, 48.8%) were sent a personal invitation to participate in this study in February 2023 via SMS text message. From the remaining eligible population, 2000 individuals (19.6% of the 10,207 eligible people) were sampled based on 5-year age groups to receive the invitation via letter. The sample size was determined for the assessment of the uptake of the app [29,30] and estimated to be sufficient to detect small effects based on logistic regression with a statistical power of 0.8 and a type I error of 0.05, assuming that 10% to 15% of the invitees would choose to participate. The invitation was valid until the beginning of March 2023. The invitation directed the participant to a website of the Finnish Institute for Health and Welfare, which provided information on the study and the app, a link to the app, and instructions on how to start using the app. Of the 6975 invitees (excluding 3 who asked for their data to be removed), 1282 (18.4%) accepted the invitation, agreed to the terms of the app, and registered for the app with their phone number and a user ID that was featured in the invitation [29,30].

The app used in this study was web-based and closed (ie, not open access) and could be used with a smartphone, tablet, or computer with no need to download the app on the device. The app was based on the BitHabit app that was developed in a T2D prevention study, Stop Diabetes (StopDia) [18,19,31], which involved the authors of this study and/or their affiliated research institutes. The BitHabit app was designed to support the adoption of lifestyle habits that promote health and aid the prevention of T2D and other lifestyle-related chronic diseases [19,20]. The BitHabit app was built on the principles of habit formation [32-34] and self-determination [35,36] theories. Habit formation approaches promote the repetition of a target behavior in the presence of consistent contextual cues until the cues begin to trigger the behavior automatically [32-34]. Self-determination theory emphasizes the role of autonomous motivation, perceived competence, and a sense of community in lifestyle changes and fosters the fulfillment of basic psychological needs—the perception of autonomy, the sense of control or self-efficacy, and the feeling of relatedness with peers or significant others [35,36]. After the StopDia study, the BitHabit app was updated by improving its visual appearance and by adding features for personalization (habit suggestions based on a background questionnaire) and for social support (a researcher-moderated discussion forum). This updated version was used in this study and is referred to as “the app” because the BitHabit brand has been out-licensed.

The app used in this study provided an “online store” or “library” of approximately 300 behavioral suggestions that translated lifestyle guidelines [37-41] into concrete actions. These actions were simple, specific, contextualized, and easily repeatable, and they were presented with a brief description of how they promote health. The actions were tied to and could be triggered by daily life activities, which made them optimal for habit formation [19]. The actions were organized into 19 habit categories related to eating and drinking (10 categories: mindful eating, meal pattern, vegetarian and fish courses, fruits and vegetables, grain products, dietary fats, sugar, salt, alcohol, and drinking water), physical activity (5 categories: aerobic exercise, muscle strengthening exercise, mobility and balance activities, stretching one’s legs, and everyday physical activity), sleep, positive mood, daily life management, and smoking.

On their first login to the app, the participants of this study gave their informed consent and completed a brief questionnaire on personal characteristics (eg, age, sex, and employment or study situation), health behavior (dietary habits, alcohol consumption, physical activity, sleep, and smoking), perceived stress, and well-being goals. On the basis of the participant’s responses to the questionnaire, the app determined the habit categories with the greatest potential for improvement and suggested the categories accordingly. After this, the participant could enter the app to browse the habit library; to inspect, select, and report actions; and to monitor and reflect on the actions performed. In addition, the app allowed setting reminders, enabled modifying the reported well-being goals, provided summary data on actions performed by other users, featured a researcher-moderated forum for discussion and questions, and provided information on health-promoting lifestyles. If the participant did not use the app for over a week, they automatically received up to 3 SMS text message reminders with 1-week intervals. No adjustments were made to the content or functionalities of the app during the study follow-up. Figure 1 illustrates the visual appearance and functionalities of the app.

The effects assessment included participants who completed the baseline questionnaire and at least one questionnaire thereafter; that is, the questionnaire midway through and/or at the end of the follow-up. We call these participants the “completers” of the study. Linear mixed effects models were used to estimate (1) the associations between app use and the slopes of the primary outcomes (diet quality, total physical activity, and BMI), that is, the interaction of elapsed time and an app use metric; and (2) the modification of these associations by the participants’ baseline characteristics. The app use metrics were evaluated at the time of the completion of each questionnaire. The models related to diet quality and BMI were fitted with individual-level random intercepts and slopes. The models related to total physical activity were fitted with individual-level random intercepts only, as the inclusion of random slopes resulted in model convergence issues.

Separate models were fitted for each outcome–app use metric pair. The app use metrics reflecting the number of reported performances were log-transformed. The model coefficients corresponding to the log-transformed numbers of performances thus reflect the expected difference in the outcome corresponding to a 2.7-fold increase in the number of performances. All models were adjusted for the following baseline characteristics potentially influencing the app use and its effects [20,23,25,29]: age, sex, education, perceived health status, BMI, perceived cognitive function, diet quality, and total physical activity, as well as the competence, barriers or concerns, and perceived benefits of using e-services. Effect modification by the baseline characteristics was assessed by estimating the 3-way interaction parameters between each characteristic, app use metric, and elapsed time. The 3-way interaction reflected how the characteristic changed the association between the use metric and the slope of the outcome.

Regarding total physical activity, values >112 hours per week (ie, on average <8 hours per day for sedentary activities and rest) were excluded as implausible. Regarding diet quality and BMI, no implausible values were detected. To assess the sensitivity against large changes in the primary outcomes due to possible misreporting, the main models were fitted also by removing individuals with the largest changes reported between 2 successive measurements. The changes between successive measurements were standardized (ie, by subtracting the mean and dividing by the SD), and the sensitivity analyses were conducted by first removing individuals with the largest standardized change >5, and then those with the largest standardized change >3.

Analyses were performed with the R software (version 4.5.0; R Foundation for Statistical Computing [56]), using the package lme4 [57] to fit the linear mixed effects models. Statistical significance was assessed with a likelihood ratio test, with a predetermined α level of .05. Data points with missing values were excluded from the analyses, and the data were assumed to be missing at random and that loss to follow-up was noninformative. As the analyses were of an exploratory nature, corrections for an increased type I error rate were not applied.

Of the 1282 individuals who accepted the study invitation and registered for the app (18.4% of invitees), 1159 (16.6%) completed the baseline questionnaire and 382 (5.5%) completed at least one questionnaire thereafter (“the completers”; Figure 2). The reasons for noncompletion are not available.

Among the completers, the proportion of men was 31% (119/382), age ranged from 20 to 74 (mean 51.3, SD 15.4) years, and the duration of full-time education ranged from 3 to 25 (mean 14.9, SD 3.4) years (Table 2). Most completers (280/382, 73%) belonged to the 3 higher income categories (household gross income approximately €2916‐€4580, or US $2953-US $4639, per month or more), and the majority were either employed or studying (218/382, 57%) or retired (134/382, 35%). A minority reported poor perceived health status (27/382, 7%), insufficient sleep (80/382, 21%), daily use of tobacco products (25/382, 7%), or poor competence to use e-services (7/382, 2%). On a scale ranging from 1 (poorer) to 5 (greater), the completers perceived their cognitive function to be on average 4.0 (SD 0.7), the extent to which they experienced barriers or concerns related to using e-services was on average 2.2 (SD 0.7), and the extent to which they perceived a benefit from e-services was on average 3.95 (SD 0.7).

Compared to the noncompleters (n=777), the completers had, on average, fewer years of education (14.9 vs 15.5 years; P=.004) and better diet quality (HDI 54.4 vs 52.5; P=.005; Table 2). The completers also differed from the noncompleters in their principal life situation (P=.03), with a greater proportion being retired (35% vs 27%) and a smaller proportion being employed or studying (57% vs 64%).

Among the completers, the percentage of active app users dropped from 97.6% (373/382) at the beginning of the follow-up to 29.1% (111/382) by the 6th week of the follow-up and to 19.9% (76/382) by the 13th week of the follow-up (Figure 3).

Over the entire follow-up, the completers used the app on a median of 5.5 (IQR 3‐10) days (Table 3). This equaled a median of 6% (IQR 3%‐11%) of the follow-up days. The completers reported performances in altogether 14 (74%) of the 19 habit categories. These 14 categories were related to physical activity (aerobic exercise, muscle strengthening exercise, mobility and balance activities, and stretching one’s legs), eating and drinking (alcohol, mindful eating, drinking water, sugar, fruits and vegetables, dietary fats, salt, and meal pattern), sleep, and positive mood. The median number of habit categories with reported performances was 2 (IQR 1‐3), while the median number of specific habits with reported performances was 5 (IQR 2‐7). The median number of total reported performances was 22 (IQR 3‐68.5). Across the app use metrics examined, the app use was significantly more frequent among the completers than among the noncompleters (P values <.001).

The completers’ mean diet quality was 54.4 (SD 11) HDI points out of 100 at baseline, rose by 1.7 points to 56.1 (SD 10.8) by the midpoint of the follow-up, and dropped by 1.5 points to 54.6 (SD 9.8) by the end of the follow-up (Table S1 in Multimedia Appendix 1). The median total physical activity was 30 (IQR 15‐50) MET hours per week at baseline, decreased by 1.25 units to 28.75 (IQR 17.5‐50) by the midpoint, and returned to 30 (IQR 17.5‐48.75) by the end of the follow-up. The mean BMI was 27.6 (SD 5.3) kg/m² at baseline, decreased by 0.1 units to 27.5 (SD 5.2) kg/m² by the midpoint, and returned to 27.6 (SD 5.7) kg/m² by the end of the follow-up.

When estimating the effect of app use on changes in the primary outcomes over the 90-day follow-up, the results differed depending on the app use metric and the outcome. A higher percentage of app use days (by 10 percentage points) was associated with a 3.09 (95% CI 0.79‐5.39) MET hours per week greater change in total physical activity (Table 4). The percentage of use days was, however, not significantly associated with changes in diet quality or BMI. In contrast, a 1-unit higher logarithmic number of total reported habit performances (ie, 2.7-fold higher on the absolute scale) was associated with a 0.51 (95% CI 0.07-0.95) point greater change in diet quality. The total number of reported performances was, however, not significantly associated with changes in total physical activity or BMI. No significant associations were found between the number of reported performances targeting each outcome and the changes observed in the respective outcomes. For unadjusted effect estimates, see Table S2 in Multimedia Appendix 1.

The results remained predominantly robust in sensitivity analyses (Table S3 in Multimedia Appendix 1). The most notable changes were a slight attenuation of the estimated effect of the percentage of use days on physical activity (β=2.33, 95% CI 0.20-4.44; P=.03) and a slight strengthening of the estimated effect of the total number of performances on physical activity (β=1.71, 95% CI 0.29-3.10; P=.02) when removing individuals with the largest standardized change >3. In addition, the effect estimate for each app use metric on BMI was slightly stronger under the sensitivity analyses.

Greater physical activity at baseline appeared to enhance the effect of the app use on total physical activity regardless of the app use metric used: the percentage of app use days (β=.13, 95% CI 0.06-0.2), the total number of reported performances (β=.05, 95% CI 0.01-0.10), or the number of reported performances targeting physical activity (β=.05, 95% CI 0.00-0.09; Tables S4-S6 in Multimedia Appendix 1). This would mean that, in increasing movement, individuals who are physically more active at baseline benefit from using the app more than individuals who are physically less active.

Additionally, a greater extent to which the participants perceived to benefit from using e-services at baseline appeared to enhance the effect of the percentage of app use days (β=4.34, 95% CI 1.11-7.60) and the number of reported performances targeting physical activity (β=2.13, 95% CI 0.10-4.13) on total physical activity (Tables S4 and S6 in Multimedia Appendix 1). This would mean that, in increasing movement, individuals with more positive attitudes toward e-services at baseline benefit from using the app more than individuals with less positive attitudes. No baseline characteristic was found to modify the effect of the app use on diet quality or BMI (Tables S4-S6 in Multimedia Appendix 1).

The participants who completed the user experience questionnaire at the end of the study gave the app a median overall acceptability score of 2.9 (IQR 2.4‐3.3) on a scale ranging from 1 to 4, where a higher value indicated greater acceptance (Table 5). The participants gave both the app and the behaviors the app suggested a median overall score of 7 (IQRs 5‐8 and 6‐8, respectively) on a scale ranging from 0 to 10. Both the median likelihood to use the app in the future and to recommend the app to others was 5 (IQR 1‐8).

We assessed the effects and effect modifiers of the use of a habit-based healthy lifestyle app on self-reported diet quality, total physical activity, and BMI in a subsample of population-based health survey participants in Finland. The assessment was exploratory and correlational in nature, building on a quasi-experimental 1-group pre-post intervention. Over 3 months, the study participants used the app on a median of 5.9% (IQR 3%-11%) of days and reported a median of 22 (IQR 3-68.5) total habit performances. The percentage of use days was associated with increased physical activity, and the number of reported performances was associated with improved diet. Some evidence also suggested that greater baseline physical activity and a positive attitude toward e-services could enhance the app use effect on physical activity. Overall, the app was accepted and received moderate ratings. However, the likelihood at which the participants considered using the app in the future and recommending the app to others indicated that there is room to improve the app. The study contributes to the scant evidence base of the effects and factors that influence the effects of digital lifestyle interventions that use a habit-based approach, in a broadly recruited sample of adults. Moreover, the work provides a real-world example of scaling up a digital lifestyle intervention previously studied among individuals at risk of chronic diseases to a more heterogeneous audience.

The strengths of the study include a behavioral theory–based digital lifestyle intervention, a web-based app that had been updated based on the learnings of a prior RCT among adults at risk of T2D [18]. In this study, the app was offered to a large group of adults sampled from the participants of a population-based health survey. As such, the study served as an attempt to scale the intervention up to a broader and more heterogeneous population. The study was rolled out in a realistic setting in which the app was simply made available for self-directed promotion of well-being and health, with no additional professional or peer support provided to boost its use. The assessments of the app use effect and effect modification incorporated participants’ objectively measured engagement with the app—a key determinant of effectiveness—and considered a wide variety of participant characteristics potentially influencing the effects. The work thus contributes to the scant evidence base on the effect modifiers of exclusively digital lifestyle interventions.

The results of the study must, however, be interpreted in the context of its limitations, particularly the lack of a control arm that precluded the assessment of causal effects and the reliance on self-reported data, both of which increase the uncertainty of our findings. To reduce the potential bias caused by self-reported data, we used measures that were validated (eg, the HDI), theory-based (the acceptability measures), and/or used in prior research (eg, the physical activity and BMI measurements); removed implausible responses from the data; and conducted sensitivity analyses that excluded the participants who reported the largest changes in the primary outcomes.

Further limitations include a multiplicity of analyses and outcomes, which increases the risk of type I error, and low participation and high attrition rates, which reduce the representativeness of our study sample to the general adult population and the statistical power to detect small effects. While the individuals to whom the app was offered represented a subsample of a population-based health survey that targeted a nationally representative random sample of adults [28], this subsample only included individuals who had chosen to participate in the survey by a certain date. Of these, 18% (1282/6975) accepted the offer and registered for the app. Of those registered, 30% (n=382) completed the measurements required for the effects assessment of this study. These individuals tended to be better off in terms of socioeconomic status, health, lifestyles, and e-service use. Our findings may thus have limited relevance to the broader adult population of Finland. Reaching and engaging more vulnerable population groups would likely require more tailored approaches.

The high attrition rate may cause bias if individuals who benefited from the app were more likely to complete the study questionnaires, leading to overoptimistic effect estimates. We compared the baseline characteristics of the completers and noncompleters of the study. The comparison showed that the completers had, on average, slightly fewer years of education and better diet quality and that they were more often retired than the noncompleters. The low participation and high attrition rate were not unusual, however, as they are widely reported challenges in digital lifestyle interventions [3,4,16,17,21], particularly when conducted among unselected target groups that may lack perceived need and motivation for lifestyle change [7,22]. In this study, the invitees had additionally recently participated in the population-based health survey, which may have reduced their motivation and resources to participate in this study as well. The findings of this study should therefore be considered preliminary, exploratory evidence that encourages further research in more robust study designs among more diverse and representative study populations with sufficient statistical power to detect small effects.

The findings of this study imply that for the development of more widely engaging habit-based digital lifestyle interventions, cocreation with the target group would be crucial. In addition, expert consensus supports a user-centered and iterative development that uses mixed methods and in-depth qualitative research to progressively refine the intervention to meet users’ context and needs [67]. Such work could also deepen our understanding of the relationship between the engagement with and the effects of the app [66], for example, by facilitating the characterization of “effective engagement”—which may not simply mean “more engagement”—to achieve the intended outcomes [67].

This exploratory, correlational study suggests that an inexpensive, low-intensity digital support for healthy habits could foster small beneficial lifestyle changes if it succeeds in sustaining the user’s engagement. However, effects may depend on the individual. Identifying those most likely to benefit from such exclusively digital interventions calls for deeper insight into the modifiers of user engagement and effects.