Health systems across the U.S. are investing in both electronic health record systems and institutional biobanks to expand research and accelerate improvements in personalized health care. The coupling of institutional biobanks with electronic health record systems make each resource more valuable since stored biological samples can be studied along with health information collected in health records or during the course of research. Although health data has been shared between diverse organizations (public health, research, healthcare, and insurance agencies) for decades, the quantity, quality, and rate of sharing has been propelled to new levels through advances in fields such as genomics. For example, the National Institutes of Health-sponsored Electronic Medical Records and Genomics (eMERGE) Network is increasing the integration of genomic data into medical records to accelerate the translation pipeline. The potential cost-efficiency of linking electronic health record (EHR) systems with biobanks is leading to the replication and expansion of networks able to share health information among a wider range of users and for a larger proportion of the public [1
The idea that data and samples are collected for unknown future research projects strains current informed consent and data sharing models, which apply to discrete uses and emphasize the articulation of risks and benefits of specific research projects. In short, consent and data sharing operate on a one-form one-study model, while biobanking seeks to obtain consent and permission for data sharing for future research and/or for multiple projects. With a greater proportion of the population being represented in biobanks or in electronic health record systems, the risk to personal autonomy and to privacy necessarily increases in scale and scope. The policy and practice changes that biobanks and large health information systems demand transform systems of accountability and oversight as well as the terms and conditions for public trust, which is broadly recognized as critical to the ethical efficiency and practicability of highly networked systems [5
However, shifting data acquisition from a cohort-driven approach to one that leverages the convenience and cost-efficiency of electronic health record data is not without its ethical, legal, and societal challenges, which are reflected in debates about electronic security, privacy, anonymity and identifiability, and benefit sharing [6
]. Informed consent in particular struggles to meet the demands of increased data sharing and indefinite research use. Implementing “dynamic” [13
] and “durable” [1
] consent policies will require changes in how and when patients learn about research participation and their rights and responsibilities therein. To attain durable consent that meets the ethical demands of informed consent—autonomy, beneficence, and justice—mechanisms for establishing “access points” [15
] are critical for gaining and maintaining an active relationship between patient, provider, and researcher. An access point is a direct, interpersonal interaction in which one of the individuals represents a complex system or institution and can thus serve to build and maintain trust. Access points mitigate the uncertainty of being involved in large, abstract systems through the direct and human interaction that forms the basis of a trusting relationship [15
]. If an individual has the opportunity to interact with another who can represent organizational and institutional interests, the basis for trust (or mistrust) becomes tangible. Complex initiatives such as eMERGE and others that link electronic health record systems to biobanks used for public health and scientific research can generate public skepticism and doubt in both enterprises [15
] thus driving an increased need for trust.
Trust relationships in health information systems may operate one-on-one, locally to a specific organization, or across institutional boundaries. The trustor to trustee relationship may involve any combination of individuals, organizations, institutions, or systems. Examples include the doctor-patient relationship, the relationship between consumers and an organization such as Kaiser Permanente or a local research university, and the relationships defined by agreements between health care providers and public health departments to report certain types of data and information. In advancing a vision for health information infrastructure in the United States, the Institute of Medicine describes the importance of a strong “fabric of trust” between stakeholders [5
], including the general public, to assure the good intentions of data sharing for individuals, organizations, and institutions in providing health care services and conducting research. Most studies of trust in the health care context have evaluated the doctor-patient relationship. Relatively few have examined public perspectives of the system as a whole as the boundaries between research and practice become more fluid.
Empirical studies generally define trust as a cognitive expectation or willingness to impart authority and accept vulnerability to another in the fulfillment of a given set of tasks. A number of factors can influence the capacity and inclination to trust including the trustor’s past experience or willingness to trust on the one hand, and the trustee’s competency, reliability, reputation, honesty, or interestedness in the trusted relationship on the other. The trustor or trustee can be an individual, organization, or institution; the consequence of a trustee betraying confidence can mean revealing information to the wrong person, financial waste, or endangering lives [11
]. Mediators and moderators of trust include characteristics of the relationship between trustor, trustee, and the context influencing the expectations and willingness to accept vulnerability [17
]. In the present analysis, we focus on characteristics of the trustor that would influence their trust in an expanding health information system, which would underlie advances in personalized medicine.
1.1. Conceptual Model
To guide our investigation of the factors underlying the public’s trust in health information systems we created a conceptual model (Figure 1
) representing six arenas anticipated to influence System Trust. Briefly, this model extends current research (reviewed below) on trust in the health system by measuring trust at the individual and system levels, examining four key dimensions of trust: Fidelity, competency, integrity, and global trust. It evaluates the relationship between trust in the health system and: (1) knowledge of health information sharing; (2) experience with the health system; (3) attitudes and beliefs about privacy; (4) expectations of benefit; (5) psychosocial factors; and (6) demographic characteristics.
1.2. System Trust and Its Dimensions
Surveys of trust in the health system encompass several dimensions most frequently including: Communication, honesty, confidence, competence, fidelity, system trust, confidentiality and fairness [24
]. Hall et al.
developed the Wake Forest Scale that has been applied to a number of relevant dimensions of the health system at large including trust in physicians [18
], the medical profession [26
], and insurance companies [27
]. Other scales of trust in the health setting, organizations, and technology use dimensions that are consistent with the Wake Forest Scales (See e.g., LaVeist et al
. on race, trust, and health [28
]; McKnight, Choudhury, and Kacmar on technology [29
]; and Siegrist on GMOs [30
]). Our study focuses on fidelity, integrity, competency, and global trust dimensions as critical aspects of trust in information sharing necessary to implementing and scaling personalized medicine to large populations. Specifically, fidelity captures attitudes about the benevolence of the health system, i.e.
, the ability of the system to prioritize the needs and interests of the public [31
]. Integrity, defined as honesty, captures confidence in upholding the principles of non-deception. Competency refers to the ability and expertise to minimize errors and achieve goals. Global trust is an integrative concept that captures an individual’s general perception of trustworthiness. It is meant to capture aspects of trust for which a rational basis is not necessary [18
1.3. Knowledge of Health Information Sharing
Research on the public understanding of science has used qualitative and quantitative methods to evaluate the question of whether the lack of support for science is simply a question of a knowledge deficit. Evans and Durant showed that more knowledgeable individuals were more likely to support general science research, but were less likely to support controversial scientific endeavors such as human embryology [32
], suggesting that science is not to be given a carte blanche in determining the limits of acceptable fields of investigation. Bruce Wynne’s study of Cumbrian sheep farmers strongly suggested that increased knowledge among the “lay” public does not necessarily translate into increased trust in the “expert” [33
]. In expanding the networks for health information, public engagement is identified as the key mechanism for building trust and acceptance, often under the assumption that this interactional form can overcome “knowledge deficits”. In short, “It is assumed that more or better knowledge or improved communication will enhance receptivity to innovations” [36
]. These studies exemplify research that shows that knowledge impacts support for science, sometimes positively and sometimes negatively. To this end, our study looks to see whether knowledge impacts trust in data sharing and if so, whether or not it increases support. We developed a set of fact-based questions to measure an individual’s knowledge about current, common policies and practices for data sharing among health care providers, insurers, researchers, and public health.
1.4. Experience with the Health System
Trust is likely to be influenced by the amount of direct experience an individual has with the trustee [37
]. Prior experience with the actors in a complex system creates a type of awareness and understanding that helps make large and abstract systems accessible, reducing uncertainty and increasing trust [15
]. Nonetheless, prior experience or familiarity with the object of trust only offers the possibility that an individual might come to trust without actually guaranteeing it [39
]. Drawing on Luhmann’s theory that familiarity increases trust insofar as it reduces uncertainty [40
], we assessed whether or not respondents had any contact with the health system either by seeing a primary health provider or having insurance.
1.5. Trustor Expectations
In the absence of direct knowledge about information use or experience with the health system, the public can still hold expectations for what the outcomes and benefits of the system will be. Various national and international reports [41
], and direct-to-consumer (e.g., 23andMe, PatientsLikeMe) and private big data initiatives (e.g., Blue Health Intelligence) make the claim that expanding information infrastructure and making data sharing more efficient will improve the quality of health care and improve health. Understanding the public’s view of these goals and their general view of data sharing sheds light on the expectations they hold in entering into a relationship in which trust plays a central role.
1.6. Trustor Privacy Concerns
Technological advances have improved electronic data security immensely and privacy considerations have often been down-played since encryption, password protections, and firewalls reduce the risk of data resources being compromised [7
]. Survey research evaluating the public’s concerns about privacy suggests that it is a high salience issue, but that it is unclear to what extent fears about discrimination or a violation of privacy precludes participation in biobanks, a comparable arena in which data is collected and stored for future research use [44
]. There is some evidence that trust may increase if an individual is confident in the ability of a system to protect individual privacy [37
2. Experimental Section
2.1. Questionnaire Development
We developed a 119-item survey to evaluate predictors of trust in the health system, broadly defined as a web of relationships among health care providers, departments of health, insurance systems, and researchers, System Trust. We focused on including the six trustor characteristics described above in the conceptual model (Figure 1
) as well as additional questions about trust in specific institutions (health care providers, researchers, and public health), quality of experience, perceived control, and adequacy of policy oversight. Measures of the dependent variable—system trust—and the independent variables used in this paper were adapted from prior studies and contextualized for the health system [18
]. Specifically, we used the California Health Foundation’s 2005 National Consumer Health Privacy survey [48
] and methods used in risk analysis literature (see e.g., [49
]) to develop measures of knowledge, experience, and expectation of benefit. Questions from the Medical Mistrust Index [51
], and related studies of privacy of health information were adapted to assess privacy [48
]. The Health Privacy Survey also informed questions about expectation of benefit and knowledge. Additional knowledge questions were developed by the research team based on its collective experience in conducting community conversations about biobanking in the state of Michigan [13
]. Questions from the General Social Survey [55
], the General Self-efficacy Scale [56
] and the Rosenberg Self Esteem Scale [57
] were used to evaluate psychosocial factors. The complete survey as it was administered is available online [58
To estimate and control for potential bias in participant responses due to the type of scale, we measured beliefs about privacy, psychosocial factors, and System Trust using two scales. Half of the participants were asked questions on a four-point bi-polar Agree/Disagree scale (Strongly Agree, Somewhat Agree, Somewhat Disagree, and Strongly Disagree). The other half responded to these questions using a four-point uni-polar scale based on the prompt: “How true are the following statements.” The value labels that followed were: Not at all true, somewhat true, fairly true, and very true. While there were some significant mean differences in the responses depending upon which scale was utilized, there was no difference in any of the regression relationships with System Trust. Consequently, we added a regression parameter to adjust for type of scale in all models presented here to correct for this potential bias but did not need to add interaction terms with the trustor characteristics being evaluated. Uniformly, the four-point unipolar scale had slightly better statistical properties in terms of its centering in the four point scale, including less skewness and kurtosis than the bi-polar scale.
In September 2013, we conducted an online Qualtrix survey of the general public (n
= 447) using Amazon’s Mechanical Turk (MTurk) system. MTurk is an online Internet crowdsourcing marketplace that is increasingly being used for survey research and is a good source for efficiently gathering high-quality data [59
]. MTurk workers are demographically at least if not more representative of the U.S. population as traditional subject pools taken from college undergraduate and Internet samples in terms of gender, race, age and education [60
]. As compared to typical Internet convenience samples, non-response error seems to be less of a concern in MTurk samples.
2.3. Statistical Analysis
Descriptive distributional statistics were estimated on all variables to identify outliers or other distributional characteristics that may influence regression modeling. For the main outcomes of System Trust as well as four trustor characteristics (privacy, self-esteem, altruism, and self-efficacy), we used Chronbach’s alpha and principal component analysis to identify the most parsimonious set of survey questions that explained the most multivariate variation in that dimension. After removal of variables, new Chronbach’s alpha and principal components were estimated to confirm the reliability of the group of variables. Table 1
shows the Chronbach’s alpha estimations for the four dimensions of trust and four trustor characteristics (privacy, self-esteem, altruism, and self-efficacy) with the original set of variables and with the more parsimonious set derived from the principal component analysis. Supplementary Table 1, Table 2, Table 3 and Table 4
show the principal components and Eigenvectors for the variables in each of the four trust dimensions (fidelity, integrity, competency, and global trust), as well as the results of the PCA after variable removal.
Chronbach’s Alpha for Indices.
Chronbach’s Alpha for Indices.
|Trust Dimension||All Variables||Reduced Set of Variables|
|No. of Items||Chronbach’s α||No. of Items||Chronbach’s α|
|Altruism||4||0.6915||4 (no change)||0.6915 (no change)|
Descriptive statistics of survey participants (N = 447) and univariate regression relationship with system trust.
Descriptive statistics of survey participants (N = 447) and univariate regression relationship with system trust.
|Demographic Factor||Sample (%)||US Population a (%)||β' (Univariate)|
|Sex|| || || |
|Age|| || || |
|Race/ Ethnicity|| || || |
|Education|| || || |
|High School diploma or less||12.5%||43.2%||Ref|
|Some college or 2-year college||42.1%||28.6%||−0.48|
|Masters or above||8.50%||9.8%||−0.21|
|Home ownership|| || || |
|Owns home||37.6%|| ||Ref|
|Does not own home||62.4%|| ||−0.28|
|Self-rated health|| || ||−0.29**|
|Excellent||17.6%|| || |
|Very good||40.7%|| || |
|Good||28.6%|| || |
|Fair||11.4%|| || |
|Poor||1.57%|| || |
Indices were then calculated for System Trust and key trustor characteristics (e.g., privacy index, esteem index, etc.) as the sum of the participant’s responses to those survey questions divided by the number of questions.
Ordinary Least Squares (OLS) Regression analysis was used to estimate the linear relationship between overall trust in the health system and each survey question and indices, (separately) before estimating a multivariable model using stepwise selection methods (inclusion criteria (p < 0.05) and backward elimination using exclusion criteria (p ≥ 0.10). In all models we included an indicator variable to control for whether a respondent answered questions using the bipolar or unipolar scale.
This study corroborates research that shows trust to be a multi-dimensional and complex interplay between the characteristics of the trustor, the trustee, and the context in which trust is negotiated. With its application in a world where electronic health records are replacing paper-based systems, and where these data systems are being built to maximize interoperability, the concomitant technical, ethical, and policy challenges have been discussed (See, e.g., [2
]). Our study of trust sheds light on the factors that will be important in understanding the strength of the “trust fabric” of health information systems that increasingly integrate the power of biobanks with electronic health record systems. In examining characteristics of the trustor associated with trust in the health system, we found that knowledge, privacy, experience, expectations of benefit, and psychosocial factors are important in evaluating trust.
Knowledge and concerns about privacy were found to be the key factors in predicting lower levels of trust. With regard to knowledge, the results of this survey suggest that relying on the public to independently seek information or that simply informing the public of current practices may not automatically result in a more trusting environment; at present, to know information policy is not to love it, as is hypothesized frequently in the area of public support of basic science (i.e.
, “to know science is to love it”) [34
]. This finding provides an important caveat to community engagement research in the arena of biobanking and data sharing. Specifically, while engagement efforts have often revealed that a more informed public is more trusting and supportive population biobanking efforts [53
], it is more likely the process of engagement that drives the support and not the top-down bestowing of information. Understanding the importance of process in informing the public reveals the need for and value of investing in community consultation approaches that seek engagement and education via partnership models or deliberative democracy methodologies [64
Privacy has been a growing concern for Americans over the past two decades with major implications for system trust and stakeholders involved in linking electronic health records and biobanks [67
]. Indeed, balancing privacy interests with public health and health care interests in sharing health information will be an ongoing issue in seeking public trust. Addressing privacy concerns is a task will fall to the stewards and brokers of health information who can provide key access points for individuals to understand how health information is used and negotiate the terms of such use. Representatives of the health information system might be physicians, who are already known to be trusted agents, or academic researchers, who are less known, but are likely to be trusted agents given the high profile of their universities in creating local identity and communities. Engaging these professional groups to evaluate to what extent they are willing and able to take on this added brokerage role will be important in developing trusted and trustworthy systems that bridge research and health care practice.
Factors associated with increased trust include having experience with a primary care provider, expectation of benefit and psychosocial factors. Not only having a primary care provider, but also visiting that provider on one or more occasions within the year, predicts trust in the health system. This suggests that inter-personal relationships can have a positive effect on trust building in complex systems. As the health system becomes increasingly interconnected in the electronic space, it will be critical to find mechanisms and spaces in which trust can be negotiated and built person-to-person as it is in the doctor-patient relationship.
Trust is further shaped by the quality, length, and nature of the relatedness of the trustor and trustee [23
]. Notably, simply having a primary care provider, but not accessing that service, did not have any effect on system trust in our study. While a slow process, building trust at the provider level is likely to have positive spillover effects on other institutions in the health information system including research, public health, and insurance. Initiatives aimed at adding recruitment to “informed cohorts” [1
] to patient-provider interactions are likely to have a positive impact on trust building and the feasibility of “durable consent.” At the same time, psychosocial variables indicating altruism and generalized trust remained statistically significant in the final multivariable model. These are opinions and beliefs that are likely important to understanding the trustor’s proclivity toward having a positive view of the health system, its institutions and organization, and its capacity to share health information in the best interests of the patient, the research participant, or the client. Many of these factors are intuitively understood in patient-provider relationships and in one-on-one discussions. Assuring access points in which dialogue about the relationship between biobanking and electronic health records is a two-way interaction, flexible to interpersonal dynamics, will further make consent a meaningful process.
One of the strongest predictors of system trust was having a positive view of data sharing. As a part of patient engagement or biobank recruitment, being able to demonstrate that health information sharing improves health care quality will be key in building public trust. Efforts to gauge overall opinion about health information sharing will shed critical insight into the state of trust in the health system. Stakeholder engagement, which has been a key component of biobanking in the past decade and has been integral to the more recent eMERGE Network, has informed technical design solutions and has been important to facilitating organizational change efforts [67
]. These efforts should continue to identify the public’s expectations to understand how they might be met or betrayed in EHR and biobanking programs. Indeed, “durable consent,” will require trusting relationships [13
] and implementation of policies and procedures that increase transparency, assure the protection of privacy, and demonstrate trustworthiness by stating how data sharing improves health and the quality of health care.
Future studies should examine system trust and its predictors using nationally representative samples. Our results thus far suggest that decision-makers in health information sharing need to explore mechanisms and policy options that effectively build or sustain trust as they develop partnerships to work across systems. Expanding networks of health information to include electronic health record systems and research biobanks promises to accelerate the slow road of translating research into personalized medical practice, but it remains to be seen what the explicit benefits are, to whom they accrue, and what and how the public sees these benefits distributed. Finding ways to build trust can help to harmonize systems of accountability and oversight more efficiently. This is a challenging task that may require substantial short-term investment with benefits that are hard to quantify. However, a policy of no change, which assumes trust and implicitly masks many of the current health information sharing practices, may be the most risky proposition for assuring the public’s trust in the complex arena of personalized medical care.