We created a pooled cross-section using data from both the 2011 and 2014 Health Information National Trends Survey (HINTS). The survey is administered as repeat cross-sections by the National Cancer Institute to a national sample of noninstitutionalized adults and gathers information regarding attitudes and perceptions about health information access and use [19]. HINTS maintains a core set of questions asked in each wave of the survey, but specific topic modules are included in separate cycles. Such is the case at-hand. The questions of interest were included only in the 2011 and 2014 surveys, restricting our research to these 2 cross-sections.

Both years of the survey were mail based. HINTS employed a stratified probability sample of the US adult, civilian, noninstitutionalized population. Addresses were randomly selected from the US Postal Service’s list of residential addresses, and then an adult within a selected household was chosen to respond to the survey using the next birthday method. The next birthday method asked for the adult in the household who would next have a birthday to complete the survey and was used to eliminate bias associated with the household member most likely to receive mail. A prepaid incentive was sent at the first mailing, and multiple follow-ups were sent to recipients in order to maximize the response rate. For household with a Hispanic last name, a Spanish version of the questionnaire was delivered in addition to the English version. The total number of respondents in the 2011 and 2014 surveys were 3959 and 3677, respectively. Taking into account survey design and weighting issues, the HINTS response rates were 36.67% (3959/10796) in 2011 and 34.44% (3677/10676) in 2014. Both survey iterations yielded samples that allowed for population-level inferences after adjustment. All respondents with complete responses for all variables of interest were included in the analytic sample.

For the dependent variable, the HINTS survey asked whether the respondent had “ever kept information from (their) health care provider because (they) were concerned about the privacy and security of (their) medical record” (yes, no). This variable was used as the outcome variable for all analyses.

For our independent variables of interest, we closely followed the variable construction of related research [17,18]. Using data from the 2011 HINTS, Agaku et al evaluated the relationship between 4 indicators of privacy and security concerns and withholding behavior [17]. These 4 questions were also asked in the 2014 HINTS, making comparability possible between years. The 4 related questions about privacy and security concerns were as follows: do respondents have concerns about unauthorized access to their medical information when it is transferred electronically between providers; do respondents have concerns about unauthorized access to their medical information when it is faxed between health care providers; do they feel confident that safeguards are in place to protect their medical information from unauthorized access; and do they feel confident that they had a say in the collection, use, and sharing of their medical information. Respondents could answer each of these 4 questions in 3 levels: not at all concerned or confident, somewhat concerned or confident, or very concerned or confident. We tested differences in our model using all 3 levels and compared the results to using only 2 levels (not at all vs at least somewhat) and found no differences. Thus, for the purposes of simplicity, we dichotomized these variables.

Using the 2011 and 2012 HINTS surveys, Campos-Castillo et al identified a suppressor relationship between the perception that a provider had an EHR (“As far as you know, do any of your doctors or other health care providers maintain your medical information in a computerized system?”) and perceived global quality of care rating on withholding behavior [18]. This suppressor relationship occurred when 1 variable, the suppressor (eg, global quality of care rating), had a positive association with a covariate of interest (eg, perception of EHR use), but a negative relationship with the outcome variable (eg, withholding behavior). Accounting for the suppressor could reveal associations between the covariate and the outcome that might not have been detectable without controlling for the suppressor. Thus, we also included the perception that a provider had an EHR (yes, no) and perceived global quality of care rating as key independent variables of interest. The HINTS asked respondents who had a nonemergency department visit in the last 12 months to rate their perception of the quality of care they received using a Likert scale (poor, fair, good, very good, excellent). For comparability to the study by Campos-Castillo et al, the quality of care variable was left as a continuous variable and coded so that higher values indicated better care.

For comparability to the study by Campos-Castillo et al, our control variables aligned with their model [18]. The control variables captured respondents’ sociodemographic characteristics, health status, and health care utilization and preferences. Sociodemographic characteristics included race or ethnicity (white, black, Latino, other), gender (male, female), categorical age in years (18-35, 35-49, 50-64, 65-74, 75 or older), education level (less than high school, high-school, some college, college, graduate), annual category of household income (<US $20,000, US $20,000-$34,999, US $35,000-$49,999, US $50,000-$74,999, >US $75,000), an indicator for living in a rural area (defined as a nonmetro county), home-ownership status (homeowner, not homeowner), marital status (married, not married), any health insurance coverage (yes, no), immigration status (born in United States, immigrant), and employment status (employed, not employed). Items about patient health status included a self-rated general health measure (poor, fair, good, very good, excellent), an anxiety and depression index (none, mild, moderate, severe), and self-care self-efficacy (not confident, a little confident, somewhat confident, very confident, completely confident). Health care utilization and preferences included the number of nonemergency room visits in the year prior to the survey (1, 2-4, 5-9, 10 or more), regular health care provider (yes, no), perceived importance of personal health record access (not at all important, somewhat important, very important), and perceived importance that providers share data electronically (not at all important, somewhat important, very important).

Weighted, but unadjusted t tests or chi-squares were used to compare sample characteristics across the 2 years. Our analytic approach was designed to test the association of the independent variables of interest with withholding behavior in each year independently, as well as to test whether the relationship of the variables on withholding behavior changed between 2011 and 2014. To accomplish these tests, the 2 cross-sections were pooled together and a single fully interacted multivariate logit model, with each independent variable interacting with year, was estimated. This interacted model was solved to determine the adjusted odds ratios (OR) and confidence intervals (CI) within each year. The significance of the interaction term was used to evaluate the relative differences between years for each parameter. All results were weighted to yield US population-level inferences using a standard weighting approach developed for the HINTS dataset [20]. All analyses were conducted using Stata 14 (StataCorp LP) [21].

Public perception of the safety of their medical records is critical to not only encouraging full disclosure to their health care provider, but also supporting adoption and use of electronic modes of health communication made available by new technologies. Distrust can lead to withholding of information from providers and undermine the delivery of high-quality, efficient care. The aim of our analysis was to determine the factors that contribute to this withholding behavior, and how the effect of these factors may be shifting over time. In short, we found that the association between patient concerns and withholding information from a provider remained unchanged between 2011 and 2014.

Earlier work using 2011 HINTS data found that respondents with concerns about both faxed and electronic data, and lack of confidence that safeguards were in place to protect medical information, were more likely to withhold information from their provider [17]. Our model included a more extensive set of control variables, as well as indicators of quality and provider EHR use, and found no such relationships to exist in either 2011 or 2014. This discordance between the 2 analyses suggests that patient concerns over the safety of their medical information may not be adversely related to their disclosure of PHI to their providers. Alternatively, other factors beyond general privacy and security concerns may lead to withholding behaviors, such as lack of trust, stigma, or concerns about insurance rates [22-24].

In contrast to the lack of replicability of the earlier findings regarding the relationship between security and privacy concerns, and withholding behavior, our analysis did observe findings similar to those of Campos-Castillo et al regarding the association between quality and withholding behavior [18]. The original work did not include the privacy and security questions used in this analysis, but did include an identical set of covariates. Nonetheless, Campos-Castillo et al found that perceptions of greater quality of care reduced the odds of PHI withholding behavior [18]. This relationship was observed in our study in both 2011 and 2014, effectively reinforcing the earlier work of Campos-Castillo et al [18]. Interestingly, no difference was observed in the correlation between quality and withholding behavior between years. This latter finding suggests that despite the rise in ubiquity of EHRs alongside more public privacy breaches, high perceived quality of care may still trump any concerns that contribute to withholding behavior.

Related to this issue, our study did observe an increase in perceived EHR use between 2011 and 2014. Despite similar findings regarding quality perceptions between our study and the study by Campos-Castillo et al, our work found no relationship between perceived provider EHR use and withholding behavior, while their study did find the presence of such relationship [18]. To be specific, the study by Campos-Castillo et al found that quality acted as a suppressor variable that moderated the relationship between perceived EHR use and patient withholding behavior [18]. The variables regarding attitudes about EHR and privacy and security may be acting as confounders in the relationship, which would explain the significance in the earlier study but not this one.

Overall, our analysis suggests that in spite of the existence of security and privacy concerns, focusing resources on the delivery of high-quality care may be an effective strategy to foster patient trust. Patients may perceive quality as an indicator of a provider’s carefulness with their medical information. Quality may also help to build the patient-provider relationship [25,26]. Alternatively, the notion of privacy is evolving as more and more personal information is held on the Internet [27,28]. It may be possible that given the increasing digitization of personal information, the US population is willing to accept greater amounts of privacy risks of their personal data as a trade-off for greater convenience or better quality of care. This is a ripe area for future research, as the field of health services research should consider the role that changes in collective ideas of privacy may be playing in how patients relate to the health care system.

Our analysis faced 3 important limitations. First, the administration of HINTS combined with the weighting technique allowed for the survey to be nationally representative. However, selection bias might have remained that limited the generalizability of the sample. Furthermore, our dependent variable (withholding behavior) could be related to survey response, and respondents with complete answers to all questions might systematically differ from nonrespondents. Related to this issue, because the HINTS survey was administered to only noninstitutionalized individuals, the findings regarding withholding behavior might be biased toward the outpatient environment. Continued monitoring of the factors that contribute to patient withholding with future iterations of HINTS can help to assess the impact of this bias and evaluate the true relationship between the variables of interest and the outcome.

Second, all information in HINTS was self-reported, potentially resulting in unreliable responses. This concern was particularly relevant to the withholding behavior question that was subject to social desirability bias [29]. Third, while our analysis did compare data from 2 different years, the cross-sectional nature of HINTS made the determination of causal inference challenging. Specifically, HINTS asked whether a person had ever withheld information from their provider, leaving open the possibility that withholding behavior preceded concerns about privacy and security, or quality perceptions. This issue was further complicated by the ordering of the questions in both cycles of the HINTS survey used in this study, where the question about perception of quality of care preceded the questions about perceived EHR use and withholding behavior. Furthermore, the frame of reference for the quality of care question was “…in the past 12 months,” while for EHR use or withholding behavior question, the frame of reference was undefined or “ever.” As a result, the responses to these questions might have drawn on different experiences, and might not necessarily reference the same encounter with the health care system. Thus, our findings regarding quality of care might be biased away from the null, and our study results should be interpreted with these measurement limitations in mind. However, despite the inability to detect causality, the national representativeness of HINTS makes it useful to identify macro trends at the individual variable level. Future studies that examine the effect of privacy and security concerns on patient withholding behavior may take a more micro approach, potentially using in-depth interviews to better understand how these concerns may manifest themselves and to identify specific omitted factors.

Monitoring and assessing how technological advances may be related to patient behavior is critical to insure high-quality care and patient safety. In contrast to previous findings, the analysis presented in this study suggested minimal effects of privacy and security concerns on PHI withholding behavior, and that this relationship was constant over time. Similarly, the relationship between quality perceptions and withholding behavior was also constant over time, yet negatively correlated at both time points. Thus, our findings suggested that improving quality can buffer privacy and security concerns. While technological safeguards to protect patient health information remains important, health professionals should not forget that individual relationships remain the foundation of the patient’s experience with the health care system.