Public health advocates have long endorsed an approach that prioritizes achieving small risk reductions for most of the population over large improvements for the minority of the population who are at high risk [1-5]. A middle-road between the population and high-risk approaches is emerging from the realization that much of the population lacking biologic cardiometabolic risk factors is not truly at low risk [6,7]; instead, they manifest equally impactful behavioral risk factors that warrant targeting for primordial prevention of disease [8-12].

The average adult reports at least two chronic disease risk behaviors; 25% report three or more; and the magnitude of behavior change needed to bring each risk factor into compliance with public health guidance is typically large [13-16]. Unhealthy diet and activity behaviors are the most prevalent lifestyle risks. Fewer than 15% of US adults eat five or more servings of fruits and vegetables daily; median intake is about half that amount [17]. Only 29% meet dietary guidelines to consume less than 10% calories from saturated fats [18-20]. Half fall short of public health recommendations for moderate to vigorous physical activity (MVPA) [21,22], and more than 50% exceed two hours per day watching television [21,23]. These four behaviors associate separately with heightened risk of cardiovascular disease and cancers, and link with others that predict premature mortality [9,10,24-30].

To date, the diet and activity changes produced by most multiple behavior change interventions have been small and poorly maintained. Conversely, when interventions have produced large initial behavior changes, long term effectiveness has been greater. However, skepticism persists about whether individuals without disease can be motivated to make and maintain large behavioral changes. We hypothesize that community dwelling adults with multiple diet and activity risk behaviors could be activated to achieve and maintain guideline levels of these behaviors by a scalable, multicomponent intervention that integrates mHealth technology, modest incentives, and remote connected coaching.

Inclusion of intervention components was guided by three principles—effectiveness, scalability, and synergy. Telephone coaching was used because the approach has demonstrated effectiveness and greater reach than in-person counseling [31,32]. On the other hand, although remote coaching is more scalable than in-person treatment, it produces smaller behavior changes [32]. Because larger magnitude behavior changes are maintained better [33], modest incentives were used that would maximize initial behavior changes [34,35]. In a prior study, we observed that incentives motivated participants to make changes of greater magnitude than they thought they could accomplish, and after being successful, most tried to maintain gains [36]. A smartphone app and accelerometer were used to provide diet and physical activity feedback synchronously to participants and their coaches, enabling connected, maximally personalized, adaptive coaching.

The Make Better Choices 2 (MBC2) trial aimed to improve upon the most effective intervention from a previous (MBC1) trial that also treated adults with four concurrent diet and activity risk behaviors: low fruit and vegetable intakes, high saturated fat intake, low MVPA, and high sedentary leisure screen time [36]. MBC1 results showed that participants incentivized for 3 weeks to increase fruits and vegetables, while decreasing leisure screen time made large sustained improvements in targeted behaviors as well as in saturated fat, which was untargeted. However, MVPA did not improve. Hence, the MBC2 trial tested whether also targeting MVPA simultaneously with or sequentially after the other diet and activity behaviors could optimize all four diet and activity behaviors, relative to a contact-control intervention.

Participants were primarily female (162/212, 76.4%), minority (125/212, 59.0%), college educated (147/212, 69.3%), with a mean age of 40.8 (11.9) years and mean BMI 34.3 (8.8). The groups did not differ in their baseline characteristics (Table 1).

Of study applicants who were web-screened, 18.4% proceeded through in-person screening and on into baseline recording to verify the presence of all four risk behaviors. Most of those listed in Figure 3 as excluded for “Other” reasons only partially completed the web screener. Of candidates who underwent baseline screening, 45.9% (212/462) were randomized. Two candidates randomly assigned to the stress and sleep contact control condition failed to receive the allocated intervention: both withdrew before receiving any treatment.

Loss to follow-up was 17.9% (38/212) and not differential across treatments (Figure 3). In the composite Z analysis, 83.5% (177/212) of participants provided a composite Z-score at two or more time points; 68.4% (145/212) provided three or more; and 50.5% (107/212) provided all four time points.

The combined simultaneous and sequential interventions produced sustained improvement, as compared to control, on the composite diet and activity score at 3, 6, and 9 months (P<.001; see Table 2 and Figure 4 A). Sequential treatment produced a small, significantly greater composite diet and activity improvement than simultaneous treatment at 6 months (P=.03); however, no differences were evident at 3 and 9 months (see Table 2 and Figure 4 B).

Figures 4 and 5 show the behavior changes over time of each treatment group. Both simultaneous and sequential interventions improved each of the four behaviors to guideline levels, exceeding control at all three follow-up assessments (See Table 2). At 9 months, intervention increased the two healthful behaviors: (1) fruit and vegetable intake by 6.5 servings per day (95% CI 6.1-6.8), mean difference from control=6.2 servings per day (95% CI 5.5-6.9), and (2) MVPA by 24.7 min per day (95% CI 20.0-29.5), mean difference from control=12.1 min per day (95% CI 5.4-18.9); and decreased the two unhealthful behaviors: (1) sedentary leisure –170.5 min per day (95% CI –183.5 to –157.5), mean difference from control=–137.7 min per day (95% CI –155.9 to –119.5), and (2) saturated fat intake –3.6% (95% CI –4.1 to –3.1), mean difference from control=–3.3% (95% CI –4.3 to –2.2).

In addition to being sustained, the improvements produced by the active interventions were generally large (0.5 to 1.7 standard deviation unit improvements compared to baseline). Although Sequential treatment reduced saturated fat intake more than simultaneous treatment at 3 months (mean difference –1.1%, 95% CI –1.8 to –0.3) and 6 months (mean difference –1.8%, 95% CI –2.6-1.0, no difference was evident at 9 months, and no other differences between simultaneous and sequential treatments were observed (Table 2).

Treatment fidelity averaged 96.8% across the 2-year study; 3 out of 20 coaches required retraining. Receipt of calls declined from 66.0% during the first half of treatment initiation to 57.7% during the second half (F[1209]=12.05, P<.001), not differing among treatment groups (61.9%, 95% CI 58.0-65.7; P=.12). Self-monitoring decreased from an average of 96.3% at baseline to 72.3% at 3 months, 63.5% at 6 months, and 54.6% at 9 months (F[3627]=95.0, P<.001), without differences across intervention groups (P=.41). Goal attainment was greater for the Active intervention groups (58.8%, 95% CI 52.2% to 65.0%) than control (33.6%, 95% CI 23.1% to 46.0%) during the first half of treatment initiation, (z=3.46, P<.001), but Active and Control groups did not differ during the last half of treatment (38.3%, 95% CI 32.6% to 44.2%; z=0.13, P=.89).

These findings support the hypothesis that adults with multiple diet and activity risk behaviors can be activated to make and maintain large improvements in diet and activity behaviors by a scalable, multicomponent intervention that integrates mHealth technology, modest incentives, and remote-connected coaching by trained paraprofessionals. Both active MBC2 interventions produced larger sustained improvements in fruit and vegetables, saturated fat, physical activity, and sedentary leisure screen time than those observed in most prior trials, including our MBC1 study [13,36,48-51], such that all four behaviors surpassed guideline recommended levels at the final study follow-up. Also, unlike the prior MBC1 interventions, both active MBC2 interventions yielded a sustained increase in MVPA documented by accelerometry [51,52]. We attribute the greater maintenance of positive diet and activity changes in the MBC2 study to the longer intervention (12 weeks versus three) and continued availability of intervention technology with some minimal coach contact.

To our knowledge, the present study is unique in testing simultaneous vs sequential versions of a multicomponent mHealth intervention to address the four most prevalent, co-occurring diet and activity risk behaviors [38,39,49,53-56]. The findings show comparable improvement from intervening to increase MVPA simultaneously with or sequentially after targeting other diet and activity risk behaviors. Even though sequential treatment produced somewhat greater improvement than Simultaneous treatment in healthy diet and activity at 6 months, the difference was small in magnitude and not sustained. These findings expand to four the number of co-occurring poor diet and activity habits that can demonstrably be changed simultaneously and support the conclusion that either a simultaneous or a sequential approach to multiple health behavior change can be expected to yield benefit [54,55].

The unusually large, well-maintained diet and activity improvements observed in this trial are likely attributable to the effectiveness of the MBC2 intervention components, including the use of appealing mHealth technology and connective coaching as vehicles to deploy effective behavior change techniques (goal-setting, self-monitoring, feedback, support, accountability). Unlike recent trials that provided digital feedback solely to patients and failed to find benefit from supplying a wearable accelerometer [57,58], MBC2 provided feedback synchronously to both participants and coaches, enabling synergistic benefit through connected counseling. Strong engagement with the intervention and study technology resulted, evidenced by participants self-monitoring on 50% of days even after 9 months.

The MBC2 study’s US $5 per week incentive for participants to meet behavioral goals during treatment initiation apparently had the intended effect of motivating participants to make large improvements. Notably, MBC2 participants made somewhat larger diet and activity improvements than those in the MBC1 study, even though the MBC2 incentive was two-thirds smaller. Moreover, no incentive to sustain healthful diet and activity changes was operative in either the MBC1 or MBC2 trial; nevertheless, behavioral improvements were maintained. Hence, these findings contradict the worry that use of incentives followed by their discontinuation inevitably undermines behavioral maintenance. Results accord with a growing body of evidence showing sustained improvements after incentives cease [34,35]. Potential scalability of modest incentives is suggested by Centers for Medicare and Medicaid reimbursement of contingency contracting for some habit disorders, and by the growing number of individuals and employers that find incentives for healthy lifestyle change cost effective [59].

Strengths of the study include a strong scientific premise grounded in the high population prevalence of individuals whose multiple diet and activity risk behaviors place them at moderately high chronic disease risk. Other strengths include a rigorous, internally valid clinical trial design comparing two active interventions with a contact-matched control, and strong initial allocation concealment and treatment fidelity procedures. In addition to objective measurement of MVPA, another strength was a 9 month follow up period allowing examination of both initiation and maintenance of behavior change.

The study also had limitations. Three of four behavioral outcomes were assessed exclusively by self-report and could have been subject to demand characteristics. Fruit and vegetable consumption may have been overestimated and time spent in leisure sedentary screen time underestimated. However, although some risk of self-report bias persists, the objective measure of MVPA derived from the accelerometer also showed large, sustained improvement following the active interventions. Fruit and vegetable intake and sedentary leisure showed the largest improvements in this study, as they had in our prior MBC1 study. Notably, in the MBC1 study, improvements in these two behaviors were also unique in being accompanied by increased self-efficacy, suggesting that changing them is both feasible and empowering [60]. Although intervention benefits persisted through nine months, longer duration follow-up remains needed. A lack of sustained superiority of sequential over simultaneous treatment could have been caused by the fact that goal progression was time-dependent, rather than mastery-based. The sequential intervention added a physical activity goal at week 7, regardless of whether participants had achieved mastery of their fruit and vegetable or sedentary leisure screen time goals. It remains possible that sequential treatment could have increased MVPA even more if the addition of this new target goal had been delayed until initial behavior targets were reached.

Treating multiple risk behaviors simultaneously is inherently more efficient than treating them sequentially, but simultaneous change might be more feasible for some population subgroups than others. In MBC2’s diverse, moderately well-educated study sample, we saw no detrimental effects of intervening on multiple diet and activity behaviors all at once, rather than one at a time. However, in a different trial, people with a greater number of risk behaviors were more likely to drop out when treated simultaneously rather than sequentially [38], an effect that could reflect the association between multiplicity of risk behaviors and social disadvantage [61,62]. If overzealous intervention on too many risk behaviors at once disproportionately drives off marginalized, resource-poor subgroups, there is risk that preventive intervention will fail to reach those in greatest need of help. Caution remains warranted before inferring that all subpopulations and contexts can accomplish unlimited behavior changes at once because: 1) simultaneous (vs sequential) intervention has yielded higher relapse and dropout in some trials; 2) the number of behaviors that can be changed at once remains unknown; and 3) some risk behaviors may be disproportionately hard to change concurrently with others [39,56,63-67].

The finding that an integrated multicomponent connective mHealth intervention produced large, sustained changes in multiple diet and activity behaviors over time is encouraging. Smartphones and wearable sensors are becoming increasingly ubiquitous, equipping consumers with real-world tools that have the potential to support healthy lifestyle changes. Interventions like the present one that combine patient facing technology with digitally connected, personalized coaching support have shown promise [36,68,69]. Including behavior change coaching as a service provided by trained paraprofessionals or artificially intelligent agents could soon make technology-supported behavioral interventions a scalable part of the health care system.

A mobile health intervention integrating a smartphone app, accelerometer, modest initial performance incentives, and remote connected coaching can produce large sustained improvements in multiple prevalent diet and activity risk behaviors, whether physical activity is targeted simultaneously with or sequentially after other risk behaviors.