This was a cross-sectional study in the area of digital health technology using survey data collected from June to November 2021. Participants responded to a web-based survey administered through REDCap that consisted of 172 possible questions (available upon request) to explore the use of and attitudes toward 8 common digital health tools. Common demographic questions and health care consumption questions were used as predictors of the social determinants of digital health use and were based on questions used in the National Health Interview Survey project [18]. Conditional and branching logic based on prior responses were used to pare down the questionnaire to deliver the right questions for the right participant.

Individuals were recruited from Amazon’s mTurk platform [19]. While patients were initially going to be recruited in a clinic setting, this became a safety hazard because of the COVID-19 pandemic and required the study team to use digital recruitment methods. In the context of this issue, mTurk provided access to a large heterogeneous population of willing research participants. The inclusion criteria required participants with existing mTurk worker accounts to be aged ≥18 years and be residents of the state of Georgia in the United States. Participants were compensated US $3 through mTurk’s internal platform. The data set began with 1022 records from which 630 duplicates, 29 incomplete responses, and 1 invalid survey were removed. Thus, the final analytic sample included 362 participants’ responses.

Survey questions were designed by the study team to investigate the use rates of different digital health tools and what factors may shape an individual’s use of digital health tools. The survey instrument contained primarily closed-ended question types with some opportunities to provide open-ended responses if the participant was willing. The questionnaire encompassed 172 possible questions in the English language that entailed the following domains: participant demographics and health consumption background, telehealth, digital health education, prescription management tools, digital mental health services, and doctor finder tools (available on request). The survey took approximately 15 to 20 minutes for participants to complete. Questions were presented to each participant in the same order. Demographic data and questions about health care use were also implemented to understand how social factors may impact digital health tool use.

Study data were collected and managed using REDCap electronic data capture tools hosted at Emory University [20,21]. REDCap is a secure web-based software platform designed to support data capture for research studies, providing an intuitive interface for validated data capture, audit trails for tracking data manipulation and export procedures, automated export procedures for seamless data downloads to common statistical packages, and procedures for data integration and interoperability with external sources.

The analysis included 17 independent variables. There were 2 continuous independent variables: age (as reported by the respondent) and federal poverty level (calculated using household income and household size). There were 15 self-reported categorical variables: race, gender, ethnicity, insurance status, educational attainment, living area, access to the internet, access to health care, primary care–seeking behavior in non–life-threatening events (as defined by the type of clinic visited during said events), having a routine primary care provider (PCP), amount of PCP visits in the last 12 months, amount of emergency room (ER) visits for nonurgent events in the last 12 months, amount of ER visits for urgent events in the last 12 months, amount of ER visits leading to inpatient stays in the last 12 months, and presence of chronic conditions (actual survey items available on request).

There was 1 continuous dependent variable (the number of digital health tool categories used by a patient) and 8 yes or no questions for self-reported digital health tool use in the following categories: telehealth, digital health education, prescription management tools, doctor finders, social services referral tools, digital mental health tools, digital insurance navigators, and patient portals.

First, descriptive statistics were calculated to determine the frequency of using the 8 digital health categories. Second, an adjusted linear regression was completed between the independent variables and the number of digital health tool categories adopted by a respondent. Third, separate logistic regressions were used to identify predictors of use for each category of tools. Statistical significance was determined by P values <.05. All statistical analyses were completed using R version 4.0.5 (R Foundation for Statistical Computing) with the packages tidyverse, MASS, and mice [22-25]. Multiple imputation based on mean-matching was used to impute missing data within the sample. Imputation required at least 50% or more of the existing data, and all of the included independent and dependent variables required some imputation.

This study was approved by Emory University’s institutional review board (reference STUDY00001999). Participants were recruited from Amazon’s mTurk platform and were compensated US $3 for their participation through mTurk’s internal system. All participation was voluntary, and no participant was subject to any harm. Informed consent was obtained from each participant regarding what their involvement in the study entailed and how their responses would be handled. The privacy of research participants was maintained throughout the study, and all responses were deidentified.

Given the paucity of literature surrounding the use of common digital health tools by the general population in the United States, this paper is a novel investigation on how 8 common forms of digital health management tools are adopted and used in a diverse population in Georgia. Surprisingly, there are large variations in how many and which factors predict digital health use uptake. The large variability indicates that there is likely diversity in the populations using various types of digital health tools. Moreover, this analysis’ emphasis on the variability in the adoption of digital health potentially highlights a gap between mHealth’s ideal and actual implementations; this is especially true when relationships that would logically be meaningful did not appear in the analysis. Ideally, mHealth would be a tool that allows medically indigent populations to improve their access to health care and to community-based resources that empower their care navigation efforts. Specifically, these findings highlight that digital health innovations are not always distributed equitably or to the people who would ostensibly need it the most. For example, the use of social services referral tools did not vary with education level, federal poverty level ratio, or lack of insurance. Ideally, socioeconomically disadvantaged populations—proxied for by education level, federal poverty level, and insurance status—would have significantly higher rates of adoption for social services referral tools since these tools are meant to expand this population’s access to social services. Similarly, digital prescription management tools had no statistically significant relationship with insurance status, which levies a similar concern since patients who are uninsured should be more likely to use digital tools to lower out-of-pocket prescription costs when holding the chronic condition status constant. Fundamentally, this initial investigation spotlights the need for incorporating more real-world evidence into digital health distribution, use, and outcomes beyond the controlled trial environment.

Interventions can and should be targeted to reach patients that traditionally cannot access health care options; however, these results indicate that, in practice, the patients with the greatest need may not find these technologies accessible. If this is evidence that the most socially and medically indigent patients cannot access new innovations, inequality can either become a substantial barrier to improving health outcomes at scale or even exacerbate health disparities across a digital divide. With the incorporation of on-the-ground use data, innovators and researchers can assess whether inequalities in uptake are a result of the inequalities in patient access to prerequisite infrastructure (eg, smartphones, Wi-Fi, or computers) or if some current investigations of digital health interventions may simply lack external validity, thus requiring a paradigm shift in the design, evaluation, and implementation strategies around digital health. These observations align with the fundamental shift in health intervention evaluation methodologies. More specifically, there is an increasing emphasis on generating and using real-world evidence and data to measure digital health efficacy [26].

These results functionally support the extant literature’s depiction of digital health inequality across various digital health domains and diverse populations. For example, Tappen et al [27] described a “deep digital health divide” in older populations that is associated with age, education, income, and ethnicity. Likewise, Saeed and Masters [28] noted that in psychiatric conditions, telehealth use is negatively correlated with lower socioeconomic status. Moreover, Brown et al [16] found that significant disparities exist in access to telemedical care among cardiovascular patients that are low-income, older adults, or Black or Hispanic [16]. Perhaps unsurprisingly, historic drivers of inequities in health care continue to exist in a similar manner in the digital health space. The impact of social determinants of health goes beyond access to medical care and pure health outcomes. There is an apparent impact on how the social determinants can affect access to digital health technologies, which may be a mediating variable for health outcomes and access to care. Fortunately, there is a movement toward prioritizing contextually tailored mHealth interventions that can mitigate inequalities in access to digital health [5].

The self-reported data collected from this investigation has at least three strengths. First, the large sample size provided minimizes concerns about response bias and improves statistical strength. Moreover, the use of mTurk to recruit respondents allows for a much larger pool than the original target population, focusing on patient-defined populations. As a result, this study included a more diverse panel of respondents. Second, the survey examined data across a broad array of digital health tools rather than focusing on one specific category of tools. As a result, this novel data set provides a broad view of digital health uptake across multiple subsets of patients. This also provides better comparisons across the various classes of commercialized technology. Third, the variety of independent predictors allows for better adjustment in the linear and logistic regression models to control for confounding variables’ bias.

Despite these strengths, there are at least three limitations. First, respondents were recruited and surveyed using a web-based platform (mTurk), which inherently introduces the effects of self-selection. For instance, this study analyzes how digital (ie, internet-connected and internet-enabled) health tools are used by asking questions to respondents who therefore are adept with technology and have substantial access to digital platforms before the survey. Second, the platform does not provide the opportunity to randomly sample respondents, limiting generalizability. Third, the cross-sectional nature of the data and analysis precludes causal induction. This analysis cannot link respondent behavior over time with the changes in their independent predictors’ statuses. In the future, semicausal and causal methods should be used to establish cause-and-effect relationships.

Further work to identify and reduce barriers to entry for digital health tools is vital to expanding its impact and promoting equity with the advent of new technologies. Understanding how digital health disseminates will be crucial to intervention design, implementation, and evaluation. The existing variability in the adoption of mHealth underscores the lack of a common framework to increase mHealth use across diverse patient populations, which ultimately limits an intervention’s potential for success. Innovative strategies can help improve access moving forward; however, this will require digital health innovators and regulators to regularly collect and review data regarding mHealth adoption. This can help evaluate the social and clinical returns on funding for digital health interventions, especially those aimed at improving health care access.

The large variability in adoption across digital health tools indicates that a one-size-fits-all approach to deploying these tools will likely mitigate their potential impact. Consequently, additional research is needed to better understand these patterns at larger scales and across more diverse populations, and the full range of factors that contribute to intervention uptake. Similarly, the digital health innovation community should rebuild the evaluation framework for evaluating tool distribution and adoption throughout the intervention’s lifecycle to ensure that patients are able to equitably access these services. More specifically, special consideration should be given to how new tools can contribute to the digital health divide, so when new tools are deployed, innovators should look at the characteristics of users to determine if adoption is equitably distributed.

This investigation establishes an initial portrait of how variable the use of digital health tools is across various patient demographics. Moreover, these results indicate that populations who could benefit the most from using certain tools (eg, patients who are socioeconomically disadvantaged and who would benefit from a social services referral tool) are not using these technologies. Although these tools are already publicly available, populations that could realize substantial benefits from this technology experience larger barriers to entry and sustained use (eg, information, internet, and cost obstacles). Ultimately, while these tools can be valuable, user uptake is the most important prerequisite to clinical and social utility.