Abstract
Background
The lack of effective tools available to health providers for enhancing patient physical activity prompts this study to examine the real-world impact of a physical activity reward-driven app on health outcomes, utilizing Electronic Health Records (EHR) data from Israel’s largest healthcare organization.
Methods
Conducting a retrospective cohort study, we matched app-users to non-users based on demographic and clinical characteristics.
Results
App-users have a significantly lower risk of cardiovascular disease (HR 0.95), stroke (HR 0.91), and type 2 diabetes (HR 0.82) compared to non-app users. Higher levels of physical activity among app users further reduce the incidence of cardiovascular disease (HR 0.87), stroke (HR 0.84), and type 2 diabetes (HR 0.75) compared with non-app user. However, engagement in mild physical activity, as measured by step count, does not differ from non- users in the incidence of these conditions.
Conclusions
These findings highlight the potential of app-based interventions to promote higher levels of physical activity and mitigate major vascular and metabolic illnesses.
Subject terms: Epidemiology, Medical research
Plain language summary
Regular physical activity can help prevent development of diseases such as heart disease, stroke, and type 2 diabetes. Many individuals struggle to maintain sufficient activity levels. Digital tools such as smartphone apps may help encourage healthier behaviors. This study examined the effectiveness of a smartphone app that rewards users for engaging in physical activity. We compared app users to non-users and found that those who used the app had a lower risk of developing cardiovascular disease, stroke, and type 2 diabetes. The reduction in disease risk was greater among individuals who engaged in moderate to high levels of physical activity, while those with low activity levels did not experience significant health benefits. These findings suggest that reward-based digital interventions may help promote sustained physical activity and reduce the burden of chronic disease.
Senderey et al. examine the real-world impact of a reward-driven physical activity app on cardiometabolic and cardiovascular disease incidence using electronic health records from Israel’s largest healthcare organization. App users have a lower risk of developing cardiovascular disease, stroke, and type 2 diabetes.
Introduction
The World Health Organization (WHO) has recently updated both national and international guidelines on physical activity (PA), emphasizing its crucial role in enhancing overall health and well-being. These guidelines recommend 150–300 min of moderate-intensity exercise per week or 75–150 min of vigorous-intensity exercise per week1–4.
Numerous previous studies have shown that regular physical activity reduces the risk of early mortality, as well as the development of chronic health conditions such as cardiovascular diseases, diabetes, and specific types of cancer5–9. Despite the growing numbers of evidence substantiating the multifarious health benefits associated with physical activity, nearly 80% of adults in the United States fail to meet the recommended activity levels2. Similarly, data from a national survey in Israel showed that only 29.1% of Israeli adults (aged ≥21 years) adhered to the national guidelines for physical activity10. The current guidelines emphasize that any amount of physical activity is preferable to complete inactivity, with greater benefits achieved through the practice of moderate and vigorous exercise. All of these are matters of paramount public health importance.
The widespread adoption of smartphones globally has opened up new possibilities for encouraging individuals to achieve their physical activity goals. The continuous advancement of technologies for precise monitoring of physical activity and clinical markers highlights the need for scholarly investigation. Additionally, it is noteworthy that leveraging smartphones can contribute to the remote monitoring of patient health behavior at scale11.
Few well-designed studies have explored the impact of physical activity levels on the incidence and exacerbation of chronic conditions using medical records instead of self-reported data. These studies incorporated behavioral designed interventions, such as monetary rewards and social incentives, to promote increased physical activity12–15. The use of physical activity monitoring and encouragement apps offers the opportunity to establish the association between personalized levels of physical activity and key health outcomes.
The Clalit-Active app (“the app”), launched in January 2020, was offered at no cost to all Clalit members that own Clalit’s supplementary health insurance services (SHS). The app was designed to promote healthy lifestyle, particularly by encouraging an increase in physical activity, and includes a continued monitoring of a wide range of health parameters. App users are prompted to make healthier choices, and the app has the potential to induce long-lasting changes in behavior and lifestyle. Importantly, the app adjusts over time, with personalized goals that change weekly based on user’s activity and medical recommendations.
The app enhances physical activity through its behavioral design and integration within longitudinal electronic health records (EHR) across multiple years. It provides personalized incentives, including psychological, social, and economic aspects tailored to individual physical activity levels. This combination reinforces sustained positive behavior by aligning with users’ motivations and preferences.
The current study aims to determine the real-world effectiveness of smartphone-based physical activity app and the higher levels of physical activity over time on the risk of developing chronic conditions such as cardiovascular diseases, stroke, and type 2 diabetes. Using electronic health records from Israel’s largest healthcare organization, we find that app users have a lower incidence of these conditions compared to non-users. The greatest risk reduction is observed among individuals who engage in moderate to high levels of physical activity, whereas those with low activity levels do not experience significant benefits. These findings highlight the potential of digital interventions to promote sustained physical activity and contribute to chronic disease prevention at the population level.
Methods
Population and study design
We performed a retrospective cohort study using the datasets of Clalit Health Services (CHS), the largest integrated healthcare service provider and payer system in Israel. CHS offers primary, speciality, and inpatient care to over 51% of the Israeli population, boasting 4.9 million members. CHS comprehensive healthcare data warehouse integrates hospital and community medical records and encompasses administrative and clinical data, laboratory and imaging information, pharmaceutical records, socio-demographic details, diagnoses from both community and hospital settings and biomarkers. Membership turnover within CHS is less than 1% annually, facilitating the study of population trends over time16.
The electronic health records (EHRs) used in this study are derived from Clalit Health Services, which has established rigorous validation protocols to ensure data reliability. Previous validation studies have confirmed the accuracy and comprehensiveness of Clalit’s EHR data17,18.
Eligible participants were CHS members at age ≥25 years old, with at least one year of continuous CHS membership that own CHS supplemental health services (SHS) who registered to use the app between January 2021 and June 2023. We excluded members who didn’t complete the onboarding questionnaire, CHS healthcare workers, missing health information, members without one year of CHS membership or those marked as bed-ridden.
App users were those who downloaded the app, completed the baseline questionnaire and had signed into the app at least three times within the first 30 days (i.e., app activation) emphasizing meaningful engagement during the initial month of app use. All other eligible members were marked as non-users (See Supplemental Fig. 1).
The app users were then matched 1:1 with non-app users (i.e., CHS members that meet the same baseline eligibility criteria as those listed above but did not download and activate the app). The follow-up period for both app and non-app users was defined as the initial registration date of the matched app user (i.e., index date).
The matching was performed based on socio-demographic and clinical covariates. The socio-demographic variables included gender, age at index date, place of residency (urban or rural area) and religious sector (Jewish, Arab, and Ultra-Orthodox), Gender was assigned based on electronic health records (EHR) as documented in the Clalit Health Services database, clinical variables included pre-existing chronic conditions (cancer, heart disease, sickle cell disease, asthma, stroke, hypertension, neurological disease, liver disease, thalassemia, respiratory disease, prior diagnosis of type 2 diabetes, prior diagnosis of CVD, solid organ transplant recipient or immunodeficiency), BMI category (underweight, normal weight, overweight, obese, and morbidly obese), and most recent HbA1C laboratory test value. Definitions for these variables and category thresholds are available in Supplemental Table 1.
Exact matching was performed for dichotomous and categorical variables, while the 1-nearest neighbor algorithm was employed for continuous variables to identify the closest match. App users who did not have an exact match were excluded from the analyses. We assessed covariate balance between app users and non-app users both before and after 1:1 matching (see Supplemental Fig. 2 for further details).
The intervention
The app is cellphone-based, and it was established by CHS leadership to prompt a healthy lifestyle among its members. The app is free to use for all CHS members who have supplementary health services (SHS). About 80% of CHS members are enrolled in the SHS plan. The app was initially introduced to CHS members through Clalit-funded media campaigns that were launched in January 2021. Users are prompted to enhance their engagement in physical activity by setting personalized weekly health-related goals based on their individual characteristics and previous physical activity (PA) history. An ongoing incentive system, rooted in behavioral science, offers personalized rewards aimed to increase adherence. The Clalit-Active app collects a broad range of information and since this app is embedded within Clalit, individual-level anonymized data is extracted from the app registry and its data is integrated into Clalit electronic health records (EHR) on daily basis, allowing the opportunity to assess the impact of Clalit-Active on health outcomes (for detailed information of the app features see Supplemental Table 1). Participant data was retrospectively linked with Clalit Electronic Health Records (EHR) at the time of app use, transferred and integrated with a daily granularity. This interoperability between app logs and health records facilitates the longitudinal integration of medical information, continually updated physical activity and lifestyle-related data for all Clalit-Active users. Notably, approximately 30% of app users engage with the platform on a daily basis.
Study participant classification
The study participants were categorized as follows: App users were individuals who downloaded the app, completed the baseline questionnaire, and logged into the app at least three times within the initial 30 days, indicating substantial engagement during the early phase of app utilization. Non-app users were identified as CHS members who did not download or activate the app. Among the app users, levels of physical activity were categorized based on the number of daily steps recorded during the first 30 days. The levels were consistent with the US physical activity guidelines and were defined as: mild (up to 5000 steps per day), moderate (5000-9999 steps per day), and high (10,000 steps per day or more)19. While these thresholds align with step-based categories commonly used in research, they are not directly derived from the US Physical Activity Guidelines, which do not specify step-based goals. Instead, these categories were chosen to approximate mild, moderate, and high levels of activity based on general guidelines and previous studies that correlate step counts with physical activity intensity8,20.
Outcomes
For the current study, we included three main outcomes: (1) new-onset (first) of cardiovascular disease (CVD), (2) new-onset (first) of type 2 diabetes and (3) first stroke. Additionally, we created a fourth composite outcome, encompassing any of these three diagnoses. All extracted at an individual level from the patient’s Electronic Health Records. For all four outcomes, only newly diagnosis conditions reported and documented after the index date and through the follow-up period were considered (refer to Supplemental data 1 for detailed definitions of variables). In cases where users had a past or prevalent diagnosis of the three outcomes as of the index date, these individuals were excluded from the corresponding outcome-specific analysis.
We followed all eligible members and their matched controls until the end of the study period (September 30th, 2023), outcome of interest, death, or loss to follow-up, whichever occurred first. For app users, loss of follow-up was defined as the date when they stopped using the app for at least three months.
Covariates
Adjusting for confounders of the association between physical activity, and subsequent health outcomes was performed through the matching and the cox proportional hazards regression model. Certain adjustment variables were shared across all outcome models, while others were specific to certain models only. Common confounders included BMI at baseline, number of GP visits ever recorded before the index date, and prior diagnosis of the following conditions: hypertension, type 2 diabetes, CVD, hyperlipidemia, and cancer.
All variables were extracted according to the most recently documented value reported in the EHR before the index date. Full variable definitions are presented in Supplemental Table 1.
Statistics and reproducibility
Following the matching, descriptive statistics were performed to compare the sociodemographic and clinical characteristics of app users to non-app users. We summarized the distribution of categorical variables per group and provided the average and standard deviation or median and IQR for continuous variables based on their distribution. Additionally, we characterize the levels of physical activity (mild, moderate, and high) by the type of incentives. The cumulative incidence for each outcome of interest was calculated using the Aalen-Johansen estimator21. Cox models were conducted to determine the outcome-specific hazard ratios (HR) of incidence of the three study outcomes as well as the composite outcome. We performed four different Cox proportional hazards regression models that were each adjusted for a predefined distinct set of covariates based on scientific literature and clinical judgment. All models were adjusted for the following confounders: age, gender, socioeconomic status, number of physician visits up to the index date. The models for incident CVD and stroke included the following variables at adjustment: history of hypertension, type 2 diabetes, hyperlipidemia, and latest BMI recorded prior to the index date. The model for incident type 2 diabetes included: history of hypertension, CVD, hyperlipidemia, and latest BMI recorded prior to the index date. The model for the composite outcome included: age, gender, socioeconomic status, history of hypertension, hyperlipidemia, latest BMI recorded prior to the index date and number of physician visits up to the index date. These analyses were exploratory in nature and allowed a data-driven approach to identify the incident diseases having a statistically significant association with physical activity while adjusting for relevant covariates. Additionally, a similar analysis was performed by the three levels of physical activity compared to non-app users. Based on pre-existing literature regarding the differences in physical activity levels between genders, a subgroup analysis was conducted by the levels of physical activity and gender for all the study outcomes. The proportional hazards assumption was examined using the cox.zph function22 of the survival R package and was met for all models.
Sensitivity analysis
To assess the robustness of our results about the effect of app usage on health outcomes, a sensitivity analysis was performed using a different matching approach. We employed a Propensity Score Matching (PSM). Each app user was matched with a non-user, based on the similarity of their propensity scores (1:1 matching ratio). Multivariable logistic regression was conducted to generate propensity scores.
Ethics
The study was approved by the CHS Institutional Review Board (IRB) (0068-21-COM1(. As this was a retrospective study using de-identified patient data, the requirement for informed consent was waived. However, all individuals who signed up for the Clalit-Active app provided informed consent for data collection as outlined in the app’s terms of use. The CHS database was initially established and approved for research use by the CHS IRB, which continues to oversee data governance, privacy protections, and research activities involving the dataset. All analyses in this study were conducted in accordance with institutional and ethical guidelines.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
Results
Study population and baseline characteristics
The current study included 2,689,601 participants aged 16 years and older, all of whom were covered by the SHS at the initiation of the intervention. Of these, 758,996 individuals (28.2%) downloaded the app. 354,533 app users and met all other inclusion criteria, and first onboarded between January 2021 and June 2023. We excluded 57,949 (16.3%) individuals that were not matched. App users differed from the overall Clalit patient population, being, on average, younger and with a lower prevalence of pre-existing chronic conditions (see Supplemental data 1). These differences were addressed in the analysis through careful matching with non-app users, a total of 296,972 app users were successfully matched to a non-user (1:1 ratio), as illustrated in Fig. 1. A covariate balance between app users and their matched counterparts was achieved across the considered confounding variables. Further details regarding the evaluation of covariate balance are available in Supplemental Fig. 2.
Fig. 1.
Population Flow Chart. Describes the selection process of the participants. The final study cohort includes n = 296,628 biologically independent app users and n = 296,628 matched non-app users. Participants were excluded due to missing health information, lack of continuous healthcare membership, or failure to meet study criteria.
The two groups were balanced in the demographic characteristics and the clinical characteristics. This includes gender (57% of women), sector distribution (89% Jewish, 7.6% non-Jewish, and 3.9% Ultra-Orthodox) (Table 1), and key health indicators such as BMI (mean: 26.5) and HbA1C (mean: 5.7) (Table 2).
Table 1.
Sociodemographic characteristics of app users and non-app users after matching
| Characteristic | App users | Non-app users |
|---|---|---|
| Individuals, N | 296,628 | 296,628 |
| Age, mean (SD) | 46 (14) | 46 (14) |
| Gender, n (%) | ||
| Female | 169,884 (57%) | 169,884 (57%) |
| Male | 126,744 (43%) | 126,744 (43%) |
| Sector, n (%) | ||
| Jewish | 262,709 (89%) | 262,709 (89%) |
| Ultra-orthodox | 11,478 (3.9%) | 11,478 (3.9%) |
| Non-Jewish | 22,433 (7.6%) | 22, 433 (7.6%) |
| Missing | 8 (<0.1%) | 8 (<0.1%) |
| Socio-economic status, n (%) | ||
| Low | 169,844 (57%) | 155,358 (52%) |
| Moderate | 121,455 (41%) | 135,092 (46%) |
| High | 5291 (1.8%) | 6165 (2.1%) |
| Unknown | 38 (<0.1%) | 13 (<0.1%) |
| Type of living, n (%) | ||
| Jewish cities | 278,293 (94%) | 278,293 (94%) |
| Towns | 8703 (2.9%) | 8703 (2.9%) |
| Small towns | 9084 (3.1%) | 9084 (3.1%) |
| Village | 548 (0.2%) | 548 (0.2%) |
Data represents the sociodemographic characteristics of app users and their matched non-app users. Sample size: n = 296,628 biologically independent participants per group. All categorical variables were described as N (%), while continuous variables were described as average (SD) for normal distributions or median (IQR) for skewed distributions.
SD Standard Deviation, SES Socioeconomic Status.
Table 2.
Clinical characteristics of app users and non-app users after matching
| Characteristic | App users | Non-app users |
|---|---|---|
| Individuals, N | 296,628 | 296,628 |
| BMI, mean (SD) | 26.5 (5.4) | 26.5 (5.2) |
| HbA1C, mean (SD) | 5.72 (0.85) | 5.68 (0.86) |
| Unknown | 172,862 (58%) | 172, 862 (58%) |
| Glucose (mg/dL), mean (SD) | 96 (22) | 95 (20) |
| Unknown | 37,862 (12.8%) | 36,002 (12.1%) |
| HDL (mg/dL), mean (SD) | 51 (13) | 52 (13) |
| Unknown | 48,126 (16.2%) | 45,866 (15.5%) |
| LDL (mg/dL), mean (SD) | 105 (32) | 105 (31) |
| Unknown | 50,828 (17.1%) | 48,119 (16.2%) |
| Triglycerides (mg/dL), med (IQR) | 102 (74) | 100 (70) |
| Unknown | 47,190 (15.9%) | 44,806 (15.1%) |
| Count of pre-existing chronic conditions, n (%) | ||
| 0 | 259,379 (87%) | 259,379 (87%) |
| 1–2 | 31,866 (11%) | 31,866 (11%) |
| 3–4 | 4691 (1.6%) | 4691 (1.6%) |
| 5 | 692 (0.2%) | 692 (0.2%) |
| Smoking category, n (%) | ||
| Not smoker | 177,503 (60%) | 197,044 (66%) |
| Former smoker | 58,792 (20%) | 58,147 (20%) |
| Smoker | 60,026 (20%) | 41,234 (14%) |
| Unknown | 307 (0.1%) | 203 (<0.1%) |
| Cardiovascular disease, n (%) | 18,645 (6.3%) | 18,645 (6.3%) |
| Diabetes, n (%) | 16,401 (5.5%) | 16,401 (5.5%) |
| Diabetes complications, n (%) | 5575 (1.9%) | 5575 (1.9%) |
| Hypertension, n (%) | 45,420 (15%) | 42,574 (14%) |
| Chronic renal disease, n (%) | 5769 (1.9%) | 3833 (1.3%) |
| Asthma, n (%) | 6216 (2.1%) | 5837 (2.0%) |
| Bronchiectasis, n (%) | 281 (<0.1%) | 234 (<0.1%) |
| Pulmonary fibrosis, n (%) | 24 (<0.1%) | 11 (<0.1%) |
| Chronic kidney disease, n (%) | 1346 (0.5%) | 877 (0.3%) |
| Hyperlipidemia, n (%) | 83,537 (28%) | 79,759 (27%) |
| No. GP visits in 5 yr, med (IQR) | 19 (23) | 19 (21) |
| Unknown | 3704 (<0.1%) | 2984 (<0.1%) |
Data represent clinical characteristics of app users and non-app users (n = 296,628 per group). Categorical variables are shown as n (%), continuous as mean (SD) or median (IQR).
BMI Body Mass Index, HbA1C Hemoglobin A1C, HDL High-Density Lipoprotein, LDL Low-Density Lipoprotein, IQR Interquartile Range, GP General Practitioner.
On average, app users had a lower socioeconomic status (57% vs. 52% in the low SES category) and a higher smoking rate (60% vs. 66%) than their matched non-users. Further, the prevalence of pre-existing conditions, such as a history of cardiovascular disease (6.3%) and type 2 diabetes (5.5%), was similar between the groups. The detailed comparison of the two groups presented in Table 1 and Table 2 illustrates the robust matching of patient covariates.
Further, the most frequent incentive purchases among the app users were sports equipment vouchers and smartwatches across all levels of physical activity (Supplemental data 2).
By the end of the follow-up period, among both app users and non-app users, there were 22,694 newly diagnosed CVD, stroke or type 2 diabetes cases. Of them, 20,206 newly diagnosed CVD, 2682 newly diagnosed and 5392 newly diagnosed stroke. However, across all four outcomes, the incidence rate was consistently lower among app users than compared to their matched non-users counterparts.
Cumulative incidence and disease risk by activity level
Figure 2 describes the cumulative incidence rate for stroke and CVD, and as it can be seen that the separation between the curves of app user versus non-user happened after approximately 180 days. However, the cumulative incidence for type 2 diabetes figure shows that the separation between the curves of the app users versus non-users happened earlier approximately after 90 days of becoming an app user (Fig. 2). The hazard ratio (HR) of developing type 2 diabetes (HR = 0.82; 95% CI [0.76, 0.89]), CVD (HR = 0.95; 95% CI [0.92, 0.97]), and stroke (HR = 0.91; 95% CI [0.87, 0.97]) during the follow-up was lower among app users versus non-app users (the source data for Fig. 2 is in Supplementary Data 4). The effect on the composite outcome was also statistically significant (HR = 0.90; 95% CI [0.92, 0.95]) (Table 3). In our sensitivity analysis accounting for death as a competing risk, we did not observe differences between the models. Death did not have a significant impact on the model’s outcomes (Supplemental Table 2).
Fig. 2.
Cumulative Incidence of Chronic Conditions Among App Users and Non-users. Presents the cumulative incidence of cardiovascular disease, stroke, and type 2 diabetes among app users and non-app users over the follow-up period. A shows cumulative incidence for stroke, (B) for type 2 diabetes, (C) for cardiovascular disease, and (D) for the composite outcome. Dark lines represent app users, while lighter lines represent non-users. The x-axis defines the number of days since the index date. Shaded areas represent 95% confidence intervals. Sample size: n = 296,628 biologically independent participants per group.
Table 3.
Hazard ratios for the four study outcomes among app users vs non-app users, overall and by level of physical activity
| CVD | Diabetes | Stroke | Any diagnosis | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| HR | 95% CI | HR | 95% CI | HR | 95% CI | HR | 95% CI | |||||
| Low | High | Low | High | Low | High | Low | High | |||||
| app users | 0.86a | 0.82 | 0.90 | 0.82a | 0.76 | 0.89 | 0.87a | 0.82 | 0.92 | 0.86a | 0.82 | 0.89 |
| level of activity (compared to non-app users): | ||||||||||||
| Mild | 0.92 | 0.86 | 1.00 | 1.04 | 0.92 | 1.16 | 0.91 | 0.82 | 1.01 | 0.94a | 0.88 | 1.00 |
| Moderate | 0.86a | 0.80 | 0.93 | 0.80† | 0.70 | 0.91 | 0.89a | 0.8 | 0.99 | 0.84a | 0.79 | 0.90 |
| High | 0.81a | 0.75 | 0.87 | 0.76† | 0.67 | 0.86 | 0.84a | 0.76 | 0.93 | 0.81 | 0.76 | 0.86 |
Hazard ratios (HR) and 95% confidence intervals (CI) for the study outcomes. Sample size: n = 296,628 biologically independent participants in each group. Subgroup analysis by level of physical activity includes n = 135529 for mild activity, n = 70703 for moderate activity, and n = 64824 for high activity.
The models for CVD and stroke included adjustment for the following covariates: age, gender, socioeconomic status, history of hypertension, type 2 diabetes, hyperlipidemia, latest BMI recorded prior to the index date and number of physician visits up to the index date. The model for type 2 diabetes included: age, gender, socioeconomic status, history of hypertension, CVD, hyperlipidemia, latest BMI recorded prior to the index date and number of physician visits up to the index date. The model for the composite outcome included: age, gender, socioeconomic status, history of hypertension, type 2 diabetes, latest BMI recorded prior to the index date and number of physician visits up to the index date.
CVD cardiovascular disease, first stroke episode, CI confidence interval, HR hazard ratio.
apv<0.01.
Further, our analysis of the three levels of physical activity over the follow-up period, indicates that the effect of app usage on disease incidence may be heterogeneous. Figures 3a–d shows that the cumulative incidence rate developing a chronic condition was higher among non-app users compared to individuals with moderate to high levels of physical activity (Fig. 3). Among app users, higher levels of physical activity, as measured by step count, were associated with significantly lower risk of CVD (HR 0.87, 95% CI [0.83,0.91]), stroke (HR 0.84, 95% CI[0.76,0.93]) and of type 2 diabetes (HR 0.75, 95% CI [0.65,0.86]) compared to the non- app users population (the source data for Fig. 3 is in Supplementary Data 4). However, incidence of chronic diseases among individuals with mild activity levels was not significantly different from the matched non-app users (Table 3).
Fig. 3.
Cumulative Incidence of Chronic Conditions by Physical Activity Levels. Represent the cumulative incidence of chronic conditions among participants with different levels of physical activity. A shows cumulative incidence for stroke, (B) for type 2 diabetes, (C) for cardiovascular disease, and (D) for the composite outcome. The x-axis defines the number of days since the index date and shaded areas indicate 95% confidence intervals. Sample size: mild activity (n = 135529), moderate activity (n = 70703), and high activity (n = 64824)
Gender differences in health outcomes
Additionally, our gender-based stratified analysis found that men have clearly discernible gradients across all four outcomes: as the average level of physical activity increased from mild to high, the hazard ratios have gradually decreased. Moderate and high levels of physical activity (PA), that were measured by step count, were associated with statistically significant reduction in the hazard of incident chronic disease. However, a mild level of PA was found to confer risks similar to those observed among non-users. This finding aligns with our previous analysis, which was not stratified by gender. Among women, PA was associated with a noteworthy reduction in the hazard of type 2 diabetes, across all levels of physical activity. However, among women, there were no statistically significant associations for first stroke episode or CVD (Fig. 4) (the source data for Fig. 4 is in Supplementary Data 4).
Fig. 4.
Hazard Ratios for Chronic Conditions by Physical Activity Level and Gender. Present the hazard ratio (HR) and confidence intervals (CI) type 2 diabeted, stroke, cardiovascular disease, stroke, and for the composite outcome, stratified by physical activity level and gender. The x-axis defines the HR and error bars indicate 95% confidence intervals. Sample size: n = 116,604 for males and n = 154,452 for females, with further stratification by mild, moderate, and high activity levels.
Our findings were robust in the choice of matching strategy. Sensitivity analysis using propensity score matching (PSM)—rather than our custom, two-step matching procedure—have the same results (Supplemental Fig. 3).
Discussion
The current study is the first to determine the effectiveness of embedded physical activity apps within a healthcare system in Israel and nationwide, providing a unique opportunity to glean insights into individual lifestyle behaviors and clinical data trends from a real-world data. This is made possible by the seamless integration between CHS EHRs and the data generated by the app. The current study includes 296,628 app users and an equal number of matched non-app users.
The findings highlight the importance of physical activity on the risk of developing cardiovascular disease (CVD), first stroke episode and type 2 diabetes. The app users had lower risk compared to non-app users to develop any chronic condition (i.e., CVD, stroke or type 2 diabetes). Our results showed that differences in the hazard of developing CVD or stroke occurred after approximately 180 days. However, for the separation happened earlier and can be explained by the frequency of HbA1c tests that occurred after 3 months which contribute to earlier identification of type 2 diabetic patients.
Notably, these results align with the current literature7,23. A prior study indicates that higher step counts were associated with lower risk for diabetes (OR (95% CI) = 0.69 (0.60,0.79), mirroring our findings with a Hazard ratio of 0.82 (95% CI (0.76,0.89))8. The outcomes of the sensitivity analysis employing propensity score matching are consistent with our robustness of the primary analysis, providing additional confirmation that app utilization effectively reduces the risk associated with the four observed outcomes. However, it is important to acknowledge that our study primarily demonstrates an association between physical activity levels and chronic disease risk rather than establishing a causal impact attributable directly to the app itself. Without information on changes in physical activity over time, these findings may reflect a relationship between baseline physical activity levels and health outcomes rather than the direct effect of app engagement. Therefore, while our results are promising, additional studies with longitudinal data on physical activity changes and randomized control elements are needed to confirm the app’s effectiveness in improving health behaviors and reducing disease risk.
Moreover, our study explored the effect of physical activity (PA) level on various health outcomes and found that the risk of physical activity on disease incidence may be heterogeneous. We found that being engaged in moderate to high-level PA significantly reduces the risk of developing chronic conditions compared to the non-app users. This difference was more pronounced for CVD and diabetes. However, when we compared mild PA to non-app users, we did not observe statistically significant difference. The study also examined the impact of physical activity across genders and identified a consistent directionality of the effect for both males and females20,21. Our findings regarding the gender specificity of the effect of app use emphasize the intricate interplay between gender, physical activity levels, and the risk of incident chronic diseases. Our stratified analyses highlight the importance of addressing gender-specific tailored interventions.
The current study is subject to several limitations. First, information bias that is inherited in the study by its design, where the identification of physical activity relied on data from the app. It is possible that some non-app users may engage in physical activity but are not documented in the system as they do not use the app. Further, we determined the differences between those who dropped out and those who didn’t and there were no differences between the two groups (see Supplemental data 3). It limits the potential bias regarding the physical health conditions. Second, we have confounding by indication due to potential differences between those who downloaded the app and those who did not. These potential biases could lead to a misclassification of exposure and bias the results toward the null. However, it’s crucial to note that app and non-app users were meticulously matched based on a set of clinical and demographic factors, thereby minimizing the likelihood that observed differences were solely attributed to the exposure itself.
Third, due to the retrospective nature of the dataset and its inherent limitations, the study was not designed to robustly assess a causal question. Although our exposure preceded the outcome, allowing us to examine the association between physical activity and the outcomes, we acknowledge the inability to ascertain a causal relationship between exposure and outcomes. Unfortunately, we could not fully emulate the target trial since the randomization is not feasible in real-world design. This limitation arises from the inherent variability between groups of physically active users and those who are not, which we cannot definitively discern. Furthermore, the issue of reverse causality between physical activity levels and the outcomes adds complexity to the interpretation. Since, individuals at lower risk for chronic diseases may be more likely to engage in physical activity or use the app, rather than app use directly reducing disease risk. While our matching methods and exclusion of participants with baseline chronic disease help address this issue, they cannot fully rule out reverse causation in relatively short follow-up duration of the study, which may influence the observed associations. Future research can be conducted to emulate a target trial for further investigation. Fourth, the study specifically examines the impact of physical activity on associated outcomes rather than other factors of lifestyle domains (i.e., nutrition, sleep trends, etc.). Additionally, this study did not account for potential changes in lifestyle behaviors, such as diet, alcohol consumption, or sleep patterns, that may have been influenced by app engagement. These unmeasured factors could contribute to health outcomes independently of physical activity, and thus confound the relationship between app use and the observed reductions in disease incidence. Future studies should incorporate a more comprehensive assessment of these lifestyle changes to isolate the effects of physical activity from other health behaviors influenced by app use. Lastly, an important consideration is that the adjustment for variables such as BMI and hypertension, which may act as mediators, could attenuate the observed associations between PA and CVD, stroke, and DM. While this approach was taken to reduce confounding, it may have led to an underestimation of the total effect of PA on these outcomes. Future studies should explore models that distinguish between direct and mediated effects.
In summary, our results indicate the significant association between overall physical activity on the risk of developing chronic conditions, offering valuable insights for future health promotion efforts. The adoption of physical activity habits requires the establishment of an ongoing system based on real-world data. This system integrates daily activities with the clinical trends over time, allowing for the setting of personalized goals and incentives. Our approach involves rewarding app users who meet their goals with digital health coins, enabling them to purchase various rewards in our platform marketplace, thereby turning their commitment to health into tangible benefits. This incentive system encourages users to strive for healthier behaviors.
Further, identifying non-app users suggests that health systems require specific monitoring strategies. For both app and non-app subpopulations with specific chronic health conditions, tailored interventions based on clinical status are crucial for optimizing effectiveness. These insights provide a foundation for the development of comprehensive and personalized health promotion strategies.
Supplementary information
Description of Additional Supplementary Files
Acknowledgements
This study was supported by the Ivan and Francesca Berkowitz Family Living Laboratory Collaboration at Harvard Medical School and Clalit Research Institute.
Author contributions
R.D.B., E.J. contributed equally as senior authors to this study. A.B.S., T.M., O.H., D.C., S.H., E.J. and R.D.B. conceived and designed the study. A.B.S., T.M., M.L.C., participated in data extraction and analysis. A.B.S., S.H., and R.D.B. wrote the manuscript. E.J. and R.D.B. provided clinical guidance. All authors critically reviewed the manuscript and decided to proceed with publication. R.D.B. vouches for the data and analysis.
Peer review
Peer review information
Communications Medicine thanks Mark Hamer and Michael Schmidt for their contribution to the peer review of this work. [Peer review reports are available].
Data availability
Due to national and organizational data privacy regulations, individual-level data such as those used for this study cannot be shared. Access to the data used for this study can be made available upon request, subject to an internal review to ensure that participant privacy is protected, and subject to completion of a data sharing agreement, approval from the institutional review board of CHS and institutional guidelines and in accordance with the current data sharing guidelines of CHS and Israeli law. Pending the aforementioned approvals, data sharing will be made in a secure setting, on a per-case-specific manner, as defined by the chief information security officer of CHS. Supplemental files include aggregated-level data, that was used in the study including (1) Supplemental data 1: Source data for condition definitions (2) Supplementary Data 2: Comparison of app users before matching (3) Supplementary Data 3: Comparison of users who dropped out versus those who remained in the study (4) Supplementary Data 4: Source data for figures.
Code availability
The analytic code is available at: https://github.com/clalitresearch/ActiveApp_Effectiveness.
Competing interests
The authors declare no competing interest.
Footnotes
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary information
The online version contains supplementary material available at 10.1038/s43856-025-00792-z.
References
- 1.Bull, F. C. et al. World Health Organization 2020 guidelines on physical activity and sedentary behaviour. Br. J. Sport. Med.54, 1451–1462 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Piercy, K. L. et al. The physical activity guidelines for Americans. JAMA320, 2020–2028 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Ding, D. et al. Physical activity guidelines 2020: comprehensive and inclusive recommendations to activate populations. Lancet396, 1780–1782 (2020). [DOI] [PubMed] [Google Scholar]
- 4.Feng, H. et al. Associations of timing of physical activity with all-cause and cause- specific mortality in a prospective cohort study. Nat. Commun.14, 1–10 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Stensvold, D. et al. Effect of exercise training for five years on all cause mortality in older adults-the generation 100 study: randomised controlled trial. BMJ371, m3485 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Lear, S. A. et al. The effect of physical activity on mortality and cardiovascular disease in 130 000 people from 17 high-income, middle-income, and low-income countries: the PURE study. Lancet390, 2643–2654 (2017). [DOI] [PubMed] [Google Scholar]
- 7.Lavery, J. A. et al. Association of exercise with pan-cancer incidence and overall survival. Cancer Cell42, 169–171 (2024). [DOI] [PMC free article] [PubMed]
- 8.Master, H. et al. Association of step counts over time with the risk of chronic disease in the All of Us Research Program. Nat. Med.28, 2301–2308 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Bonn, S. E., Löf, M., Östenson, C. G. & Trolle Lagerros, Y. App-technology to improve lifestyle behaviors among working adults - the Health Integrator study, a randomized controlled trial. BMC Public Health19, 273 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.The Israeli National Health Survey 2018-2020 (‘Hebrew’, 2020).
- 11.Patel, M. S. et al. Smartphones vs wearable devices for remotely monitoring physical activity after hospital discharge: a secondary analysis of a randomized clinical trial. JAMA Netw. Open3, 2020–2023 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Patel, M. S. et al. Effectiveness of behaviorally designed gamification interventions with social incentives for increasing physical activity among overweight and obese adults across the United States: the STEP UP randomized clinical trial. JAMA Intern. Med.179, 1624–1632 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Yom-Tov, E. et al. Encouraging physical activity in patients with diabetes: intervention using a reinforcement learning system. J. Med. Internet Res.19, e7994 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Patel, M. S. et al. Effect of goal-setting approaches within a gamification intervention to increase physical activity among economically disadvantaged adults at elevated risk for major adverse cardiovascular events: the ENGAGE randomized clinical trial. JAMA Cardiol.6, 1387–1396 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Arad, A., Gneezy, U. & Mograbi, E. Intermittent incentives to encourage exercising in the long run. J. Econ. Behav. Organ.205, 560–573 (2023). [Google Scholar]
- 16.Senderey, A. B. et al. It’s how you say it: systematic A/B testing of digital messaging cut hospital no-show rates. PLoS One15, e0234817 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Karpati, T. et al. Patient clusters based on HbA1c trajectories: a step toward individualized medicine in type 2 diabetes. PLoS One13, e0207096 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Dagan, N. et al. BNT162b2 mRNA COVID-19 vaccine in a nationwide mass vaccination setting. N. Engl. J. Med. NEJMoa2101765 10.1056/NEJMoa2101765 (2021). [DOI] [PMC free article] [PubMed]
- 19.How Many Steps a Day Is Considered Active? https://www.medicinenet.com/how_many_steps_a_day_is_considered_active/article.htm.
- 20.Patel, M. S. et al. Effect of behaviorally designed gamification with social incentives on lifestyle modification among adults with uncontrolled diabetes: a randomized clinical trial. JAMA Netw. Open4, 1–15 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Borgan, Ø. Aalen–Johansen Estimator. Wiley StatsRef Stat. Ref. Online10.1002/9781118445112.STAT06001 (2014).
- 22.Lumley, T., S-, R., Elizabeth, A., Cynthia, C. & Therneau, M. T. M. Package ‘ survival’. (2023).
- 23.Patel, M. S. et al. Predicting changes in glycemic control among adults with prediabetes from activity patterns collected by wearable devices. npj Digit. Med.4, 1–7 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Description of Additional Supplementary Files
Data Availability Statement
Due to national and organizational data privacy regulations, individual-level data such as those used for this study cannot be shared. Access to the data used for this study can be made available upon request, subject to an internal review to ensure that participant privacy is protected, and subject to completion of a data sharing agreement, approval from the institutional review board of CHS and institutional guidelines and in accordance with the current data sharing guidelines of CHS and Israeli law. Pending the aforementioned approvals, data sharing will be made in a secure setting, on a per-case-specific manner, as defined by the chief information security officer of CHS. Supplemental files include aggregated-level data, that was used in the study including (1) Supplemental data 1: Source data for condition definitions (2) Supplementary Data 2: Comparison of app users before matching (3) Supplementary Data 3: Comparison of users who dropped out versus those who remained in the study (4) Supplementary Data 4: Source data for figures.
The analytic code is available at: https://github.com/clalitresearch/ActiveApp_Effectiveness.