Although HINTS is a dataset that is maintained by National Cancer Institute and does include several questions that focus on cancer, the population surveyed is random and many of the questions relate to general health communication. HINTS has 6 iterations, 5 of which include the requisite variables to extend Beckjord and colleagues’ study. The first iteration of HINTS (HINTS 1) was conducted from October 2002 through April 2003. Data collection was achieved via a random digit dial telephone survey, which generates phone numbers at random. Selected phone numbers were submitted to a telephone matching service in an effort to remove nonresidential numbers. A computer-assisted telephone interview (CATI) format was utilized to accommodate complex skip-patterns. Survey administration averaged 30 min per respondent, and data were collected from 6369 respondents.

HINTS 2 (2005) data were collected from February 2005 through August 2005. Like HINTS 1, HINTS 2 also used a random digit dialing telephone survey with a CATI format. All data collection procedures were identical to HINTS 1. Data were collected from 5586 respondents.

The HINTS 3 (2008) data were collected from January through May 2008. The HINTS 3 sample design included 2 data collection methods. One sample was drawn as a random digit dialing telephone survey, using a CATI format. The second national random sample was selected from a list of addresses from the United States Postal Service (USPS) administrative records. Data were collected from 4092 respondents via CATI and from 3582 respondents via mail. Differences between the CATI and USPS samples were tested. Despite the differences between the 2 populations in education, income, general health status, metropolitan statistical area, and internet use, all data were included to maintain fidelity to the original study.

HINTS 4, Cycles 1 and 3 data were collected from October 2011 through February 2012 and September through December 2013, respectively. The sample design for both surveys consisted of a single-mode mail survey, using the Next Birthday Method for respondent selection. The sample design for the interview portion of the survey consisted of 2 stages. In the first stage, a stratified sample of addresses was selected from a file of residential addresses. In the second stage, 1 adult was selected within each sampled household. The sampling frame consisted of a database of addresses used by Marketing Systems Group to provide random samples of addresses. Complete data were collected from 3959 respondents in 2011 and from 3185 respondents in 2013.

Sociodemographic characteristics included age (18-34, 35-49, 50-65, or >65 years), sex (male or female), education (less than high school, high school graduate, some college, or college graduate), income (<US $10,000; US $10,000-14,999; US $15,000-19,999; US $20,000-34,999; US $35,000-49,999; US $50,000-74,999; or ≥US $75,000), race and ethnicity (white, Hispanic, black, or Asian, other), metropolitan area (metro or non-metro), and health insurance status (yes or no), one’s own perceived health status (excellent, very good, good, fair, or poor), and cancer history (yes or no). Measures relating to use of technology and electronic communication were assessed in the form of questions, which asked whether they used the internet (yes or no) and whether respondents had communicated online with a provider in the past 12 months (yes or no).

There were differences in how data were categorized for the income and perceived health status variables. The categories for the income variable in HINTS changed over time, and the income categories used in this study of trends over time reflect that. Additionally, the categories for perceived health status remained stratified for a more in-depth analysis, and were not dichotomized as in the original study.

A 3-part analytic process was used to study (1) reanalysis, (2) close replication across years, and (3) trend analysis extension (see Multimedia Appendix 1). During the reanalysis stage, the publicly available HINTS 1 and 2 data were used with the goal of identifying the precise analytic methodology used in the original 2007 paper. This stage served as a precursor to replication, ensuring that the model with additional data years is done with a high degree of fidelity to the analytic framework established as valid in the literature [30]. In the close replication stage, we made minor alterations to the original analytical model to enable effective analysis of multiple data years in a manner that is as close to the original model as possible, while taking into account all available variables in each data year. Finally, we engaged in extension, that is, the analysis of additional data years, 2008, 2011, and 2013, using the same analytical approach with the purpose of identifying trends over time. The first author of the original study was invited to join the study at this stage.

Analyses were conducted using both unweighted and weighted data. All weighted analyses were conducted using jackknife variance estimation to account for the complex survey design and provide nationally representative population estimates. Mirroring the original paper [15], respondents’ sociodemographic characteristics were calculated by percentage of internet users for the years 2003, 2005, 2008, 2011, and 2013 (Table 1).

The same characteristics were then calculated by the percentage of patients who communicated online with a provider (Table 2). We employed a weighted, multivariate logistic regression for each individual data year (Multimedia Appendix 2), as well as an additional model, which pooled the data across years (Table 3) and contained the year as a predictor variable to identify potential predictors of online communication with a health care provider. Each model was adjusted to control for all sociodemographic characteristics and assessed online communication with a provider where responses of “Yes” represented a positive outcome. The original study contained a table of the results of the bivariate year-specific analyses, but trying to assess changing trends over time related to demographic factors may present spurious correlations as associations; thus, we omitted this approach in this study.

The sample characteristics of internet users for the reproduced Beckjord and colleagues [15] study using HINTS 2003 and 2005 data, and extending it to include the 2008, 2011, and 2013 HINTS iterations are presented in Table 1.

In 2003, 7% of internet users reported communicating online with a health care provider in the past 12 months, increasing to nearly 10% of internet users in 2005. These estimates are consistent with the original study, and the percentage of internet users reported communicating online with health care providers increased to 14% in 2008, 19% in 2011, and 30% in 2013.

Consistent with Beckjord and colleagues’ original study, individuals who communicated online with health care providers had significantly more years of education and were more likely to reside in a metropolitan area in 2003. In 2005, these individuals were more likely to be female and were more likely to have a personal history of cancer. The reproduction of the original study found minor differences in the means of some variables, which may be attributed to variations in the SUDAAN (Research Triangle Park, North Carolina) and STATA (College Station, Texas) jackknife algorithms. In addition, we found 2 differences between sociodemographic variables and communicating online with health care providers for 2005 when comparing results with those from the original study (see Table 2). These 2 differences were (1) communicating online with health care providers was not found to be statistically significantly associated with more years of education (P>.05) and (2) annual income was statistically significant (P=.038).

The findings of our multivariate models conflicted with the original study [15] in 2 instances in 2005. First, gender was not found to be statistically significant, and second, we found that individuals who reported poor health had higher odds of communicating online with health care providers, which was not reported in the original study. Notably, several variables were consistently found significant across multiple years. For example, gender was statistically significant for the years 2011 and 2013; having a history of cancer was statistically significant for 2005 and 2013; and geographic location was statistically significant for 2003, 2008, and 2011 (all P values were <.05). Full results of multivariate analyses by HINTS year is provided in Multimedia Appendix 2.

When exploring trends in the use of online communication with health care providers, we found a significant increase between the years 2003 and 2013. More specifically, as compared with 2003, the odds of internet users to communicate online with health care providers was found to be significantly and increasingly higher in the subsequent years, 2005 (odds ratio [OR] 1.31, 95% CI 1.03-1.68, P=.027), 2008 (OR 2.14, 95% CI 1.76-2.59, P<.001), 2011 (OR 2.92, 95% CI 2.33-3.66, P<.001), and 2013 (OR 5.77, 95% CI 4.62-7.21, P<.001).

When looking at all 5 years of HINTS data and adding in year as a covariate, the odds of communicating with a provider via the internet increased for women (OR 1.31, 95% CI 1.10-1.55, P=.002), college graduates (OR 1.88, 95% CI 1.08-3.28, P=.026), and data year (see Table 3). Conversely, the odds of communicating with a provider via the internet decreased for (1) individuals between 65 and 74 years of age (OR 0.70, 95% CI 0.52-0.94, P=.018), (2) the uninsured (OR 0.59, 95% CI 0.41-0.85, P=.005), (3) individuals with no history of cancer (OR 0.68, 95% CI 0.54-0.84, P=.001), and (4) individuals living in a nonmetro area (OR 0.63, 95% CI 0.49-0.80, P<.001).

Beckjord and colleagues [15] reported the prevalence of and changes in the use of online patient-provider communication in 2003 and 2005 and described sociodemographic and health-related factors associated with its use. In this study, we were able to use subsequent iterations of the HINTS dataset to describe the current state of online patient-provider communication and understand how its use has changed over time. The data year is shown to be statistically significant in all years with the odds of communicating with a provider online more than quadrupling from 2003 to 2013, the latter indicative of the diffusion of technology over time. This finding coincided with a recent study showing a significant increase in surgeons’ use of secure messaging; the proportion of outpatient interactions was 5.4% in 2008 compared with 15.3% in 2010 [31].

Although we were able to match the majority of the coefficients in the re-analysis stage, there were some minor differences, which may have been attributable to differences in the jackknife algorithm deployed in the software packages used. That said, the interpretation and findings of the first year (2003) did not change as a result; however, several changes were experienced in the 2005 analysis. These differences may be partially attributed to higher estimates of internet use among low-income individuals in 2005. These estimates are not surprising, given the literature reporting interest among underserved populations in using electronic and internet-based tools to communicate with their providers [32]. In addition, evidence shows increasing adoption of patient portals among providers serving underserved populations in efforts to increase patient engagement [33]. The health disparity literature brings attention to the importance of the quality of patient-provider communication when providing medical care to minority populations [34] and its contribution to disparities in medical care [35,36].

We also found that the use of online patient-provider communication steadily increased over time between the years 2003 and 2013. This upsurge in use may be largely attributed to increased interest in the use of online communication tools [32]. In addition, recent attention has been given to using online communication with providers as a means to provide patient-centered care [37] while federal policy initiatives require that physicians provide the use of secure electronic messaging to their patients through Meaningful Use Stage 2 requirements [27]. This federal push to adopt electronic health records has also led to an increase in the implementation and availability of tools to facilitate communication between patients and their providers, such as personal health records (PHRs). It is projected that 75% of adults will use PHRs by the year 2020 [38].

Beckjord and colleagues [15] also explored sociodemographic and health-related factors associated with the use of online communication with a health care provider. We found that in 2003 and 2005, women were marginally more likely to communicate online with their providers, and significantly more likely to communicate with their providers in 2011 and 2013. This is consistent with findings from a previous study identifying women as being more likely to use email to communicate with their doctors [39]. The trends may be explained by understanding gender differences in use of the internet. According to the Pew Research Center [19], men were more likely than women to use the internet in the early 2000s. However, this gap gradually decreased over the years and became equal around the year 2008. In addition, there is evidence that women are more likely than men to use the internet to access health information and to get support for health problems [40]. A recent study looking at patient portal use found that education and sex remain statistically significant when controlling for internet access and preference of communication mode (in person or over the phone vs patient portal) [41].

Being located in a metropolitan area is also shown as relatively consistent when it comes to communicating online with a health care provider. This may be partially attributed to the digital divide. There is literature that suggests patients located in urban locations are more likely to use electronic messaging to communicate with their providers [39], whereas physicians located in urban areas are more likely to provide secure messaging to their patients [42].Finally, having a history of cancer was also found to be associated with online patient-provider communication. Due to the complexity of cancer care, patients often leave visits with their physicians overwhelmed with information and confused about their condition and treatment [43]. Literature suggests that patients need multi-level communication to facilitate information exchange and foster patient-provider relationships [44]. As a result, cancer patients benefit from an asynchronous platform that allows them to access information and engage with their health care team at their convenience. Although these data do capture general health status and cancer history, they do not include other chronic conditions. We were therefore unable to explore the relationship between chronic illnesses outside of cancer and the use of online communication with a health care provider. However, the literature does suggest that these communication technologies will also help patients to more effectively manage other chronic illnesses such as diabetes and hypertension [45]. The nonsignificant finding for health status, measured by a 5-point Likert scale, may too broadly capture respondents’ health state.

HINTS is subject to the same issues all surveys and self-report instruments are, that is, low response rates, potential sampling bias, social desirability issues particularly around issues of smoking and other lifestyle choices, and item limitations. The HINTS response rate was 33% in 2003, 21% in 2005, 21% for telephone survey and 31% for mail surveys in 2007, 37% in 2011, and 35% in 2015; response rates are not a deterministic indicator of bias [46]. We note, in looking at the multivariate models by year, that the significance of gender fluctuates depending on the data year. This could have been the result of sampling dynamics and not indicative of changes in the population at large. A study asking similar questions in the same data year would have allowed for triangulation, but no such study was available. Finally, the main outcome variable is limited by its dichotomous nature and does not allow us to identify the specific type of provider respondents communicated online with within the previous 12 months. For example, a patient could have communicated electronically with a nurse or other member of the care team.

As the diffusion of this technology continues, future research related to online means of patient-provider communication should remain a focus. Future studies should try and garner a better sense of the frequency of communication and document the type of provider with whom the patient communicates to inform providers and health care organizations implementing new and additional means of online communication. Additionally, understanding use by disease type, including chronic illnesses, would be a useful addition to the literature. Although the adoption of these communication technologies is increasing among minority populations, another important area to explore is the quality of communication between minorities and their providers, and how it may influence, or be influenced by, the use of these technologies. Finally, with regards to the methodological approach utilized in this study, it is important to note that previously published studies are often not updated as new waves of data become available. HINTS data have been used in nearly 400 studies, and have yielded intriguing findings, but data updates to prior studies that use large datasets are infrequent. Perhaps this is due to instrument evolution across survey cycles that makes combining data across cycles challenging, but regardless, replication and extension of prior work is an area of research that warrants further attention.

Despite initial challenges in implementation [6], the proportion of internet users communicating online with their health care providers has significantly increased since 2003. In addition, these trends are likely to continue with the enactment of the Health Information Technology for Economic and Clinical Health Act and Meaningful Use Stage 2 core objectives, which require providers to use secure electronic messaging to communicate with their patients. However, challenges still remain pertaining to the digital divide affecting individuals residing in nonmetropolitan areas and their access to the internet, making this group less likely to communicate online with their providers. Future research should continue to investigate patient-provider communication trends, specifically to gain an understanding of successful interventions that mitigate identified barriers from both provider and patient perspectives.