This cross-sectional study was based on survey data from the Sax Institute’s 45 and Up Study. This is a large cohort including 267,153 participants aged 45 years and older who reside in New South Wales (NSW), Australia [16]. The conduct of the 45 and Up Study was approved by the University of New South Wales Human Research Ethics Committee. The survey data were linked to data from the Pharmaceutical Benefits Scheme (PBS; data on prescribed drugs) and the NSW Admitted Patient Data Collection (APDC; hospitalization data). This study has been approved by the NSW Population & Health Services Research Ethics Committee (approval #HREC/16/CIPHS/14) and the Commonwealth Scientific and Industrial Research Organisation (CSIRO) Health and Medical Human Research Ethics Committee (approval #2021_018_RR). All cohort participants provided free and informed consent.

The recruitment for the 45 and Up Study took place between January 2006 and December 2009 and was based on a random sample from the Services Australia (formerly the Australian Government Department of Human Services) Medicare enrollment database, with oversampling in people aged 80 years and over and residents of rural and remote areas [16]. About 18% of those who were contacted consented to take part in the study which represents 11% of the NSW population aged 45 years and older [14]. A detailed description of the cohort and the study methods was published by the 45 and Up Study collaborators [14,15]. For our analyses, we used data from the second wave of follow-up collected between 2018 and 2020, specifically, the 2019 follow-up survey, which included questions about participants’ technology and mHealth use [17]. The survey was sent to 68,349 participants, of whom 31,965 responded (46.77%) [16]. In general, the questionnaires contained questions on lifestyle, medical history, family history of chronic conditions, socioeconomic status, and geographic factors [17]. The Sax Institute linked the survey data to the PBS data deterministically using a unique identifier [16]. Additionally, the Centre for Health Record Linkage linked these data to the APDC data using probabilistic techniques [18].

We first identified participants with CVD or diabetes (type 1 or 2, not including gestational diabetes) at the time of the follow-up survey and then identified those who were at risk of the conditions. These methods were similar to the methods described by Joshy et al [19] for the classification of CVD and by Comino et al [20] for the classification of diabetes in the 45 and Up Study. Individuals were classified based on information from the survey, from the APDC data, and from the PBS data. In the survey, participants were asked if they had CVD or diabetes (“Has a doctor ever told you that you have:” heart failure, atrial fibrillation, other heart disease, stroke, OR diabetes—type 1, type 2 or unsure) and if they took medication for these conditions (corresponding medication listed on the questionnaire: Cardizem/Vasocardol, warfarin, OR Diabex/Diaformin). From the APDC data, the diagnoses were classified based on related hospital admissions before the date of the survey. We searched for relevant International Classification of Diseases version 10 Australian Modification (ICD-10-AM) diagnosis codes in any of the 55 diagnostic fields or relevant Australian Classification of Health Interventions procedure codes in any of the 50 procedure code fields. These codes were based on methods described by the Australian Institute of Health and Welfare (AIHW) [21-24]. From the PBS data, people were classified with CVD or diabetes if they took medication with these indications in the past 12 months [25,26]. For CVD, the only drugs considered were those that are solely indicated for CVD. To identify a CVD diagnosis, we did not consider drugs for hypertension or dyslipidemia. People with hypertension or dyslipidemia may be at risk of CVD but have not yet developed the condition. Additionally, people may take antihypertensive drugs for other reasons than hypertension. Therefore, if people took antihypertensive or lipid-modifying drugs but did not have a diagnosis of CVD (ie, reported in the 45 and Up Study data, the APDC data, or due to taking a drug with CVD as the only indication), we did not classify them as having the condition.

For CVD, being at risk was classified as taking antihypertensive medication, lipid-modifying medication, or low-dose aspirin; reporting blood-clotting problems or hypertension; being obese (≥30 kg/m²); or having a family history of CVD (parents or siblings) and not already having CVD. For T2DM, someone was classified to be at risk if they had had gestational diabetes, were obese (≥30 kg/m²), or had a family history of T2DM (parents or siblings) but were free of type 1 or T2DM. We gathered the information to identify at-risk populations from the survey and the PBS data. Women were classified as having had gestational diabetes if they self-reported having had the condition or if they had received a diabetes diagnosis before the date of their last delivery and if there was no evidence of diabetes medication in their records for the 12 months before filling out the follow-up survey.

The primary outcome was mHealth use. We defined mHealth use through the following survey question: “How often do you use apps on your mobile phone or tablet to track the following?” The answer options were “never,” “less than once a month,” “at least once a month,” “at least once a week,” or “every day.” If any of the options were selected with “less than once a month” or more frequently, the person was classified as a mHealth user (options included the following: activity or fitness, vital signs, nutrition or weight, mood or well-being, sleep, medications). This meant that if “never” or none of the options was selected, the participant was considered a nonuser. Table 1 summarizes the technology-related questions from the 45 and Up Study questionnaire that we used for the analysis [17]. We classified all those who did not select yes for the questions about device use (computer or laptop, tablet, smartphone, fitness tracker, and smart watch) as not using them. We categorized everyone as app users who did not select “I don’t use apps/don’t know what apps are.” We dichotomized the variable to app download by putting app nonusers in one group with those who selected that they had not downloaded any apps yet and the remaining in the other group.

Other variables of interest included age, sex, income, lifestyle (smoking, alcohol, fruit or vegetables, physical activity), and physical disability. We categorized income based on the Organization for Economic Cooperation and Development (OECD) report [27] into low (less than Aus $30,000 per year [US $20,580]), middle (Aus $30,000-89,999 per year [US $20,580-61,740]), and high income (Aus $90,000 or more per year [US $61,741 or more]). For smoking, we created three categories: never, past, and current smokers. We classified physical disability through survey questions about illness or disability restricting physical activity (“Is there anything that stops you from participating in physical activity?”—ill-health; “Do you regularly need help with daily tasks because of long-term illness or disability?”—yes; “Does your health now limit you in any of the following activities?”—yes selected for any of the options). We dichotomized all other lifestyle-related variables depending on if participants met the guideline recommendations (for alcohol: maximum 10 standard drinks per week [28]; for fruits and vegetables: at least 2 servings of fruits and 5 servings of vegetables per day [29]; for physical activity: at least 150 minutes of physical activity per week with vigorous physical activity counting double [30]). For fruits and vegetables, we set servings per day to 0 if the participants selected that they did not eat any fruits or vegetables, respectively. Otherwise, we did not impute any missing values. We reported on the percentage of missingness for age, sex, BMI, smoking status, alcohol consumption, fruit and vegetable intake, physical activity, income, and physical disability. For the remaining variables, we did not report on missing values because the way we categorized these variables did not result in any missing values.

We used descriptive statistics to describe the demographics of the overall population and the various subgroups (with CVD or diabetes; at risk of CVD or T2DM; free of CVD, T2DM, and their risk factors) as well as their technology and mHealth use. Further, we used the t test and the chi-square test to check for differences between mHealth users and nonusers among those at risk of CVD or T2DM. We defined statistical significance at a level of 5%. We built multivariable logistic regression models for the total sample and those at risk of CVD or T2DM to assess the influence of age, sex, physical disability, and income (predictor variables) on mHealth use (outcome variable). From these models, we calculated adjusted odds ratios (ORs) and their 95% CIs. Further, we computed the adjusted OR of mHealth use for someone with CVD or T2DM risk compared to someone free of the condition and not at risk. We conducted the analyses in RStudio (version 1.2.5042) using the programming language R (version 4.0.0; R Foundation for Statistical Computing) within the Secured Unified Research Environment (SURE) provided by the Sax Institute [31].

The overall aim of this analysis was to understand how older Australians in general and particularly those at risk of CVD or T2DM use mHealth. To our knowledge, this is the first study of this kind in Australia. Among the at-risk population, the proportion of mHealth users was slightly higher than that of the general proportion. The multivariable logistic regression analysis showed that women, younger people, individuals without disabilities, and higher earners had higher odds of using mHealth. Among the mHealth users, there were fewer smokers and fewer people with hypertension or physical disability. On the other hand, among those who did not use mHealth, fewer people were overweight and fewer reported a family history of CVD or T2DM. According to our results, people at risk of CVD or T2DM do not have higher odds of using mHealth than do those without risk.

The results largely corresponded to the results of other researchers. Shan et al [8] compared mHealth use among US Americans with CVD or at risk with those without based on data from 2018. They concluded that those who had or were at risk of CVD were less likely to use mHealth than were the rest of the study population in an unadjusted comparison. However, when adjusted for age, race, education, household income, health insurance, and urban or rural location, the OR of having a health app on the smartphone was 1.24 (95% CI 0.85-1.81; P=.26) for women with CVD or at risk of CVD and 1.12 (95% CI 0.68-1.84; P=.65) for men with CVD or at risk of CVD compared to women and men with no history or risk factors of CVD. Seifert and Vandelanotte [32] reported on the use of wearables and mHealth apps among Swiss adults aged 65 years and older based on data from 2019. Of their 1149 participants, 43.1% owned a tablet, 68.7% owned a smartphone, 7.6% owned a fitness tracker, 3.3% owned a smart watch, and 22.9% used mHealth apps. Bhuyan et al [33] analyzed 2014 data from US American adults on mHealth use and health-oriented behavior. They found that 35.9% of the smartphone or tablet owners had mHealth apps installed. The proportion of those who had mHealth apps was much lower among those aged 55 to 64 years (10.6%) and those aged 65 years and over (4.7%). Robbins et al [9] analyzed 2015 survey data on mHealth use among US American adults who owned a mobile phone. Among the participants, daily mHealth use was more common in individuals without diseases (21.3%) than in people with high blood pressure (2.7%), with obesity (13.1%), with diabetes (12.3%), with depression (12.0%), or with high cholesterol (16.6%). Robbins et al [9] concluded that their findings suggest that those most likely to benefit from mHealth are least likely to use it.

Only 18% of those who were invited to participate in the 45 and Up Study took the baseline survey [14]. This is comparable to other large international cohort studies [34-36]. Mealing et al [37] conducted an analysis in which they compared exposure-outcome relationships in the 45 and Up Study to the NSW Population Health Survey which had a response rate of about 60% [38]. Both are based on the same population, and the analysis showed that the results from both cohorts can be generalized. Additionally, some participants were lost to follow-up, which might also raise concerns about generalizability. Wang et al [39] built a logistic regression model to assess the influence of nonresponse in the first follow-up survey of the 45 and Up Study. In their conclusion, they reported that it did not lead to substantial bias and essentially did not change the interpretation of the results [39]. Further limitations were that the data set had missing values, especially concerning fruit and vegetable intake, and was based on self-report. Ng et al [40] considered the bias through self-reported weight and height in the 45 and Up Study. They stated that the resulting BMI values were logical but underestimated being overweight or obese [40]. There is also a risk of misclassification of disease diagnosis or risk. To minimize the risk, we applied methods that have been described by the AIHW [22-24] and by other researchers who worked with the 45 and Up Study [19,20]. In our analysis, we did not classify diabetes as a risk factor for CVD because we categorized people with CVD or T2DM as a separate subgroup. There is a small risk of wrongly classifying people as at risk of CVD because they took antihypertensive drugs for an indication unrelated to an increased CVD risk.

The availability of mHealth keeps increasing, and there is little doubt that it can positively impact health promotion, risk awareness, early detection, and, in general, engagement with one’s health and well-being. However, this study and other research have shown that certain demographics are more likely to use it than are others. In our study, younger females without disabilities and with higher income had the highest odds of mHealth use. As we are in the process of developing an app-based intervention for CVD and T2DM prevention, we are interested in finding strategies that would facilitate optimal uptake by those who would be most likely to benefit from the intervention. Our findings indicate that health-related apps were less used by people with a physical disability. We think that people with disabilities would likely benefit from more personalized interventions that take into account their disability. Moreover, the results showed that it will be important to consider equity issues to ensure that people with low income and older people will not be left out. Other researchers have drawn similar conclusions and proposed solutions. For example, Foley et al [41] used a mixed methods study to explore the access, use, and benefits of digital health services in Australia and identified trust as a key factor for digital health services use. They explained that recommendations by health professionals improved trust in digital health services, which could lead to increased uptake [41]. Cheng et al [42] concluded in their systematic review that the level of electronic health literacy was often overlooked when designing such interventions, leading to a digital divide with socially disadvantaged groups being left behind. Therefore, they demanded that electronic health literacy levels need to be recognized when developing interventions [42]. In their review, Borg et al [43] identified attitudes, skills, and access as barriers to digital inclusion, and social support, education, and inclusive design as enablers. The authors highlighted the importance of user-focused and collaborative designs to ensure digital inclusion [43].

Despite most people at risk of CVD or T2DM owning a smartphone, only about a third had ever used mHealth apps. There was no difference in mHealth use between people at risk of CVD or T2DM and those not at risk. People who used mHealth apps were less likely to be male, older, with physical disability, and with a lower income. This shows that it is important to consider equity issues when implementing a mHealth intervention. For example, low income or older age should not prevent people from participating in the intervention, and, therefore, these factors should be considered when developing an implementation strategy.