We recruited participants who were or had the potential to manage and share their own health data electronically. As we particularly focused on individuals who can manage health and financial tasks by the use of relevant technologies, we needed a study sample in which we can capture their daily activities of various data management beyond health care settings. To this end, we contracted with a market research company that had access to the paid panel of health consumers across 50 states in the United States for administering a cross-sectional Web-based survey in 2017. Each participant was incentivized by the completion and quality of their response, which was mainly managed by the market research firm. Our Web-based survey incorporated multiple items that measure health information–sharing behaviors, including health consumers’ intention to share their health information, information sharing contexts, demographic information, and daily technology use such as internet banking (IB) use. An institutional review board approval was obtained before survey distribution. As a result, a total of 339 responses were used for further analysis.

All survey items were sourced from the existing literature, as presented in Table 1. First, for capturing health information–sharing intention with respect to sharing contents, sharing situations, and sharing constituents, we adopted our survey items from the study by Whiddett et al [14] and Anderson and Agarwal [15]. More specifically, sharing contents contained 4 items about the types of information likely to be shared, ranging from general information to all information (including sensitive disease information). Sharing situations included four items about the instances in which health consumers are willing to transfer their health information, such as all instances or selective or emergency cases. Finally, sharing constituents involved 10 items about whom an individual is willing to share their information with such as physicians, health administrators, and insurance payers. To determine dimensionality of the items for the sharing constituents, we ran exploratory factor analysis (EFA) with varimax rotation using SPSS (IBM Corporation, Armonk, New York). Our EFA identified two factors in sharing constituents based on their direct involvement in the care: One dimension is general constituents who do not directly engage in the care (eg, community physicians, government, and health insurance companies), whereas another dimension captures care-related counterparts (eg, physicians, nurses, and pharmacists). The items for health data sharing were anchored on a 5-point Likert scale (1: strongly disagree and 5: strongly agree). Second, we measured use frequency of daily technology such as computer, internet, email, and IB (ie, “How frequently do you use internet banking?”). These items were anchored on four scales (ie, daily, weekly, monthly, and never), whose anchor was adapted from a mobile banking survey administered by the International Finance Corporation.

Finally, demographic information was captured for gender, marital status, income, education, occupation, race, ethnicity, health condition (chronic disease), and having a primary doctor within the domicile [16].

The objective of the study was to examine whether frequent use of IB affects when, what, and with whom health consumers are willing to share their own personal health data. However, in observational studies similar to this study, it is often a challenge to isolate the treatment effect because confounding factors can influence both treatment and outcome [18]. To rule out such a confounding effect, prior studies have used various statistical estimation methods such as regression and panel methods, matching estimators, instrumental variables, and regression discontinuity designs [19-21]. Among these methods, we were particularly interested in propensity score matching (PSM), as it has been widely used when randomized clinical trials are infeasible in health care research [20,21]. PSM matches the observed and possible outcomes per each object based on a propensity score—a conditional probability that each observation receives treatment based on a set of observed covariates [22,23]. After matching, average treatment effects (ATEs) can be calculated by averaging out such a difference between the observed and potential outcomes [24]. This method depends on balancing observable covariates in treatment and control groups to isolate and estimate the treatment effect in the presence of confounding effects [25].

In our data, health consumers’ use of IB and health data sharing can be confounded by known factors, that is, demographic characteristics and health status. Following a step-by-step suggestion from the study by Becker and Ichino [26], we conducted PSM (refer to the study by Becker and Ichino [26] for a more detailed explanation), including choice of confounding variables, balancing propensity score and covariates between treatment and control groups, and calculating ATE within the evaluating criteria of bias (the difference between estimated treatment and true effect) and precision of estimated treatment effects. As a first step, it is necessary to choose the correct sets of variables that affect both treatment and outcome variables to better isolate treatment effects. Next, based on the choice of confounding variables, one needs to validate whether there is an overlap in propensity scores between treatment and control groups, which is necessary for drawing an inference by comparing these two groups. After balancing propensity scores in the groups, one needs to check the distribution of covariates within the blocks of propensity score to determine if the treatment and control groups within the block have a similar covariate distribution except for the variation of treatment variable. On the basis of this, the final step is to estimate ATE of focus—either ATE among paired samples within blocks of a propensity score or average treatment effect on the treated (ATET) for the treated observations only [27]. Although there is no clear guideline for calculating sample size for PSM, matching one or two untreated subjects to each treated subject is recommended when using PSM [28].

As shown in Table 2, most of the respondents were working professionals (236/339, 69.6%), female (225/339, 66.4%), and white (269/339, 79.4%). A total of 33.3% (113/339) have chronic conditions, and interestingly, almost all participants believe they are computer literate (337/339, 99.4%). Table 2 presents the demographic characteristics of our survey respondents across three user groups—IB users with daily, weekly, or monthly use frequencies. A majority of respondents use IB weekly (170/339, 50.1%), and participants aged 25-44 years were active users of IB across three user groups (approximately 203/339, 59.9%). We did not find any statistical group difference among demographic characteristics by t test (P<.05).

Although our data were obtained from a cross-sectional Web-based survey from US health consumers, we further evaluated the representativeness of our sample compared with established benchmark. The Board of Governors of the Federal Reserve System has conducted a Web-based survey on financial consumers’ use of mobile banking in selective years [29]. As we examine the effect of IB use on PHI-sharing behavior, we compared the national profiles of IB consumers from the Federal Reserve Board in 2015 with our sample. We found that age distribution was particularly similar to our sample, as a majority of respondents were aged 25-65 years and a majority of users of mobile banking were aged 25-44 years (Multimedia Appendix 1). Furthermore, we checked age group differences among health consumers with IB frequency using a t test and found no significant group differences (P<.05).

We conducted PSM analysis using Stata version 14.2 software (College Station, Texas). In our analysis, we chose income, race, health status, and having a primary care doctor within close proximity as our confounding variables, among others, which are likely to influence both IB use and health information sharing behavior. Subsequently, the propensity score was calculated for each block using a logit model. Figure 1 presents the distribution of propensity scores between the treated and untreated groups within blocks, termed as common support (propensity score overlaps in the matched pairs for each block), with appropriate overlap across blocks of propensity scores [30].

Table 3 illustrates covariate balancing among the variables before and after matching. It indicates that standard bias (percentage of bias) was less than 5% after caliper matching, and it is reasonable to move forward with the next step of analysis. To match a treated individual with nontreated individuals using similar propensity score, we used caliper matching (0.2×standard deviation of logit of propensity score with 1:2 neighbor matching with replacement) [26]. Figure 2 demonstrates a similar distribution of information-sharing behavior (sharing content) at baseline before and after caliper matching. Thus, we proceeded with ATE and ATET.

Finally, we estimated ATE and ATET as displayed in Tables 4 and 5, respectively. Here, ATE describes the ATE of IB use on health information sharing by comparing the treated (daily use of IB) and untreated groups (nondaily use of IB) in the matched pair sample. ATET indicates the estimated average effect of daily use of IB on health information sharing among daily IB users. We defined a treatment effect of IB as the use of IB on a daily basis, as we viewed daily use of IB as health consumers’ daily habit. Prior literature noted that habit is an automatic response to be formed by more frequent use and exposure of a focal event, technologies, and tasks [31]. Thus, we can consider daily use of IB as an existing habit of managing private personal information compared with weekly or monthly use of IB in our data. As such, we proceed to examine the treatment effect of daily use of IB on health information sharing under various circumstances.

In Table 4, there are differential behaviors in sharing contents, situations, and constituents between the treated and untreated groups. Daily IB users were more willing to share general and current health information (P=.009 and P=.004, respectively) for all situations (P=.06). For sharing constituents, these individuals were prone to share their information with non–care-related personnel, such as health administrators at a hospital (P=.05). Table 5 presents the average effect of daily IB use on health information sharing among daily IB users. We found that the treated individuals were likely to share all PHI with sensitive contents (P=.04) for all situations (P=.02). Moreover, they shared more PHI with non–care-related health administrators within the hospital (P=.05). Overall, results from ATE and ATET were qualitatively similar.

Despite growing expectation of individuals’ responsibility for sharing PHI for data-driven care, there have been mixed results regarding factors influencing health consumers’ sharing intention under various circumstances. Given that management of financial information resembles that of health information, this study called for more attention on individuals’ daily use of financial information management as a proxy for health information sharing and hypothesized that the more frequently individuals managed their financial information, the more likely they were willing to share PHI under various sharing conditions. Our PSM results revealed that daily IB users were more willing to share the large extent of health information for all instances, even with personnel who were not directly involved in their care process. This is one of the first studies that explores the role of a daily habit of financial information management to predict individuals’ intention in sharing their health information.

Although we presented important findings on the role of daily habit of IB use, our results should be interpreted with caution because of their limitations. First, our study is conducted via a cross-sectional Web-based survey. Although our research question was aligned with our study design, tracking health consumers’ information management behavior over time can provide an in-depth view and further identify contextual factors in the daily living context. The cross-section time series information on health consumers’ information management in daily living would be beneficial in the future to capture granular level of measures and to control unobserved heterogeneity among individuals. Second, we conceptualized our treatment effect as habitual use of financial data management and operationalized it by the daily use of IB. Although our unidimensional, binary measure of IB use was appropriate for PSM methodology, future research can incorporate multi-item measures to capture multidimensional aspects of financial data management for health consumers in the richer research models. Finally, we acknowledge that majority of survey respondents in our study have no immediate health issues (237/339, 69.9%); therefore, the results of this study may not be generalized to patients who have various health conditions and statuses. As we assumed that the same individuals can be both health and financial consumers, the findings of this study can be a baseline information to compare individuals’ behaviors in medical situations in subsequent research.

For future research, it will be worthwhile to revisit this research model in clinical setting and explore the effect of habit on health data sharing in the clinical setting for those who have chronic conditions or medical urgency. Health data sharing is a complex and variable phenomenon, and more interdisciplinary research is indispensable. As health consumers are attaining more ownership to manage and share their own health information via websites, wearables, and mobile apps, it is important to determine whether they are capable of dealing with this volume of data [48] and whether they make informed decisions to share personal data with multiple stakeholders. Various types of structured and unstructured health information may be stored at an individual’s home or workplace or at a hospital. Given that such data need to be freed by the hands of health consumers and distributed for the optimal health care decision and outcomes [49], future research can further explore various information type and their willingness to share such data under different sharing scenarios.

This research has practical implications. First, health technology vendors may design health information management tools modeled after financial information management tools [50]. As health care and finance are both characterized by the important role of system usability for customer satisfaction, involving more consumers in the rapid process of innovation and understanding consumer behaviors and fast-moving information technology trends [51], system design can support health consumers’ willingness to share their own data and further liberate their data for consumer-centered care [52]. To better manage health information for self-health management [53], individual health consumers need to collectively as well as selectively manage either type of health information and use it for their own medical conditions and contexts. Health technology can support such sharing behavior by providing repository of health data that is similar to a consolidated bank account and allowing them to selectively share or store it [51]. Second, health consumers need to be educated on how to manage various types of health information, including access, process, and exchange of public and private health information. As health information is domain-specific and either personal [15] or widely available on the internet [54], individual health consumers should understand the difference between public and private health information and shareability of such information accordingly. As shown in our results, even with the experience of managing sensitive financial data daily, health consumers may not be ready to share PHI with other stakeholders who are not directly involved in their care. This has timely implications on the direction and content of health consumer education about sharing constituents [55].

The objective of this study was to examine the effect of daily use of financial technology on health information sharing. Considering the similarity between health and financial technology and the characteristics of such information, this study proposes that the unobserved habit of managing sensitive information daily can further affect managing and sharing another type of sensitive information—PHI. Results from PSM reveal that frequent users of financial technology are more prone to share their entire health information in all instances, even with non–care-related stakeholders. Subsequent research can explore more granular types of habits in various life domains to better understand health consumers’ readiness to manage self-health information for realizing consumer-centered care in the future.