An established proportional multistate life table (PMSLT) model was used to estimate the health impact and health system expenditure of the prescription of smartphone apps for the promotion of physical activity in primary care [23,24]. Physical activity was measured as the change in the metabolic equivalent of task (MET) minutes per week of moderate and vigorous activity. The PMSLT model simulates the entire NZ population alive in 2011 (N=4.4 million) until death or the age of 110 years. Future all-cause morbidity and mortality and incidence and case fatality rates for 5 diseases related to changes in physical activity were projected. Specifically, the model included breast cancer (women only), colorectal cancer, CHD, type 2 diabetes, and stroke.

Health gain was measured in quality-adjusted life years (QALYs) [25], whereas, for costs, a health system perspective was used, and the outputs were the difference in total health system costs between business-as-usual and the modeled intervention and included the cost of implementing the intervention. We also disaggregated the results by period and presented the impact of the intervention after 10 years and 20 years. Calibration and validation of the epidemiological aspects of the PMSLT are described in a web-based technical report [23]. A Monte Carlo simulation (2000 iterations) was used to estimate the uncertainty intervals for the key results.

QALYs and costs were discounted at 3%, with results for 0% and 6% discount rates presented as scenario analyses. We also ran the results applying an equity adjustment that set background all-cause morbidity and mortality rates for Māori to non-Māori values [26]. This technique is often used to avoid the undervaluation of health gains and identify potential health equity impacts for Māori. Scenario analyses included higher percentages of the eligible population being screened for the intervention (ie, 25% and 50%), a reversed ratio of GP to practice nurse (PN) consultation time, and maintenance of the intervention impact for 5 years.

Full details of the model are published elsewhere [23,24,27].

Rapid reviews of the literature on physical activity apps and referral schemes (ie, green prescription programs) were conducted to parameterize the intervention for modeling purposes. The modeling parameters used and the justification for their use are presented below (Tables 1 and 2).

As shown in Figure 1, during a GP visit, people with insufficient physical activity were identified using a screening question as per the one used by the NZ Health Survey [28]. For practical reasons and because it might be inappropriate to ask on some occasions, the screening question was assumed to only be asked during a proportion of such visits (10%, with 25% and 50% presented as scenario analyses). For those with insufficient levels of physical activity (defined as <300 MET minutes per week), the GP or PN then provided a printed physical activity prescription form that included specific instructions to exercise and information on how to download and use a physical activity smartphone app. The patient then chose a physical activity app, possibly with input or recommendations from the provider, from the Health Navigator app library. The Health Navigator app library is a repository of apps reviewed by experts, which contains links to a number of independently reviewed exercise apps [17]. It was assumed that individuals would use an app that was already developed and free to download. Furthermore, it was assumed that the app would be available at zero cost for as long as the participants chose to use it. However, adherence to the intervention was expected to decrease relatively quickly; therefore, long-term availability of the app would not be relevant for most participants. In addition to the prescription, GPs or PNs referred patients to service providers (eg, a regional sports trust or primary health organization in the NZ context) who provided support over the phone, including a comprehensive consultation, 2 brief follow-up calls, and technical support to use the app within the first year after the initial consultation and prescription.

We included direct costs attributable to the intervention (ie, intervention costs) and indirect costs attributable to changes in health system use resulting from the intervention [29]. Intervention costs were applied to everyone who received an initial consultation, even if they did not start using a physical activity app. In addition, it was assumed that all costs would be incurred in the first year following the initial consultation (ie, follow-up calls and technical support would occur in the same year as the intervention). Furthermore, it was assumed that those adults who were sufficiently active or who did not own a smartphone would require less than a minute of the GP’s time to screen; therefore, costs associated with such interactions were minimal and not quantified.

Current evidence indicates that physical activity interventions would likely provide minimal chronic disease reduction benefits to adults aged <40 years in NZ [24,27]. Therefore, the age range for the intervention was restricted to 40 to 79 years. Individuals aged ≥80 years were excluded because of low smartphone ownership and use, as well as the relatively high prevalence of comorbidities (eg, arthritis) that could limit participation in physical activity. Although the intervention was only applied to those aged 40 to 79 years, the model followed the entire NZ population alive in 2011 over their remaining life span. Therefore, participants could age into the intervention over time (eg, someone who was aged 38 years in 2011 could still become eligible for the intervention in 2013 when they turned 40).

On the basis of evidence in the literature (Table 1), it was assumed that 91% of eligible patients who were active smartphone users would start the intervention and that physical activity would increase by an average of 410 moderate to vigorous physical activity MET minutes per week. After 1 year, the intervention effect was assumed to be maintained by 36.5% of people, with a decay rate of 55% applied.

Tables 1 and 2 contain additional details on the base-case intervention parameters.

The potential impact of the prescription of smartphone apps for physical activity promotion in primary care settings in NZ was modeled using published estimates of uptake, effectiveness, and adherence [33,36,37]. The total health impact was modest, with 430 QALYs gained over the remaining life span of the population, albeit up to 1750 QALYs if the intervention effect was maintained for 5 years. The intervention was also likely to be cost saving for the health system and provide larger per capita health gains for Māori than for non-Māori. Given the higher per capita health gains for Māori than for non-Māori, it is plausible that this intervention could play a role in reducing health inequalities if implemented equitably. Although there is strong evidence that physical activity levels can be improved and maintained in the short term through individual-level physical activity interventions, a cumulative meta-analysis of the effects of physical activity interventions found that additional research is needed to identify interventions that are the most cost-effective [49]. This study provides additional modeling evidence of the cost-effectiveness of physical activity promotion in a primary care setting.

When compared with selected health interventions in NZ, according to methodologically compatible modeling studies, we found that the prescription of smartphone apps for physical activity promotion in primary care was likely to provide larger health gains and cost savings for the health system than a mass media campaign for physical activity apps [27], a mass media campaign for weight loss apps [50], or weight loss counseling by nurses in primary care [51] in NZ (Table 6). However, the intervention was less effective than a mass media campaign to promote a smoking cessation app in NZ [52]. It was also less effective on a per capita basis than a traditional green prescription program to promote physical activity in Australia [40], although this may be due in part to underlying differences in physical activity patterns and epidemiology across different populations. We also note that the health gains of these individual-level interventions are orders of magnitude lower than upstream interventions (eg, tobacco control endgame interventions [53] and switching driving trips to walking and cycling [24]). Implementing the prescription of smartphone apps for physical activity promotion in primary care alongside other such interventions may help maximize health gains.

A strength of this study was the use of an established PMSLT model for physical activity [24,27], which is based on high-quality disease-specific epidemiological and costing data for a whole country. The modeling framework has been widely used in Australia and NZ to assess different health interventions, including several different mHealth interventions [27,50,52], allowing for comparisons across studies. However, a limitation of the PMSLT approach is the assumption of disease independence, although the model does account for the relationship between type 2 diabetes and CHD and stroke, given that type 2 diabetes is a risk factor for these conditions. The model also does not account for potentially beneficial health impacts via the reduction of obesity, anxiety, or depression (or, conversely, health loss via an increase in these conditions because of musculoskeletal injuries associated with physical activity). Another limitation of this study was the use of a health system perspective for costs and benefits. Such a perspective does not allow for the estimation of societal-level costs and benefits outside of the health system, such as impacts on household spending or reduced greenhouse gas emissions from transport mode shifts.

There are also limitations associated with the parameter estimates used. For example, app uptake likely varies and could be increased with improved app design. In addition, adherence estimates in the literature range widely. We examined estimates from green prescription programs, stand-alone app interventions, and app+ interventions (ie, a smartphone app in addition to follow-up texts, calls, or emails). We chose an adherence estimate between those for green prescription programs and app-only interventions, which was in line with app+ interventions in which there would be some level of follow-up with the intervention participants. We assumed that all of the intervention parameters presented in Tables 1 and 2, including intervention uptake and adherence, would be the same across population groups (ie, men and women and Māori and non-Māori). However, it is possible that for certain parameters, there may be significant variations across population groups that may impact the effectiveness of the intervention. For example, it is unclear whether currently available apps adequately cater to the needs of diverse population groups.

Additional research is needed to optimize interventions that can lead to sustained increases in physical activity levels over the long term [48] and identify the forms of follow-up (eg, phone calls vs emails) that maximize adherence to physical activity interventions. App improvements may also encourage adherence. Evidence suggests that apps are most effective when they incorporate self-monitoring of physical activity, reminders for app use, and social interaction with peers [7,55]. Gamification (eg, providing point-based rewards for frequency or consistency of app use) may also improve outcomes for app-based interventions [56]. Other factors that may influence app use include simple app interfaces, easy navigation, and automatic or simplified data entry [57]. Such features help increase user engagement, which promotes positive behaviors [7]. Advancements in app design merit additional research to assess their impact on adherence.

There are indications that such a program would be best administered by PNs rather than GPs, given the larger cost savings associated with the scenario in which PNs are the dominant deliverers of the consultation. In addition, the literature suggests that GPs are often particularly time-limited [58]. Our scenario analyses suggest that larger health gains would be achievable if a higher percentage of patients were asked the screening question (ie, 25% or 50% asked the screening question). If GPs are unable to ask the screening question as frequently as PNs because of time constraints, then it may be both more effective and a better use of staff resources to routinely have PNs deliver such a program. This may also have implications for the administration of the ongoing green prescription program in NZ. However, the ratio of GP to PN consultations was only included in the model as a cost parameter, and we did not assess potential differences in effectiveness between GP and PN administration of the program. We also note that, in the context of high levels of unmet need for health care (eg, 13.3% of NZ adults did not visit a GP because of cost barriers in 2019-2020 [59]) and differential access to smartphones and the internet, additional strategies to ensure that any app-based intervention does not exacerbate existing health inequities are warranted.

With the widespread use of smartphones, mHealth interventions such as this have a large potential for scalability to a broad population [9,10]. As part of a range of interventions to address insufficient physical activity, governments should consider investing in the promotion of physical activity smartphone apps, along with additional research to improve app effectiveness and uptake. The intervention should ideally be tailored to the country context, which may include examining existing structural inequities that may influence intervention success and co-designing strategies with relevant population groups.

The promotion of smartphone apps may also complement other strategies to promote physical activity (eg, investments in walking and cycling infrastructure [60]) to support long-term behavioral changes. Although our results suggest that the promotion of physical activity smartphone apps in primary care is likely to be effective and cost-saving in NZ, these results are also likely generalizable to other high-income countries with similar chronic disease epidemiology, physical activity levels, and other population characteristics.

In this modeling study, the prescription of smartphone apps for physical activity promotion in primary care in NZ yielded modest health gains and was cost saving for the health care system. The scope for this type of mHealth intervention is expanding with the increase in smartphone ownership and the availability of easy-to-use and effective apps. This intervention should be considered by policy makers in NZ and also be considered by other high-income nations with similar characteristics.