Background

In 2012, the retailer Target sent advertisements for baby products to a teen who had not disclosed her pregnancy to her parents. Target had concluded the teen was pregnant after she purchased items like unscented lotion and cotton balls, which figured into algorithms predicting pregnancy [1]. The algorithm was allegedly accurate, but the tracking practices of Target were criticized after it was reported to the public [1, 2]. Individual customers reportedly complained and reported that predicting pregnancy from purchases was “creepy” [1]. After the public response Target was reported to have modified its marketing practices, instead of only sending baby supply coupons to women that their algorithm deemed pregnant, Target would send baby supply coupons with other home goods items mixed in [1]. “As long as a pregnant woman thinks she hasn’t been spied on, she’ll use the coupons” [2]. The sensitive nature of early pregnancy makes the practice of targeted marketing seem particularly invasive. There are many regulations enacted to protect traditional clinical health information, but there is less guidance for how health related digital data should be protected.

Contemporary practices to safeguard the privacy of health related data, such as HIPAA privacy rules, emerged at a time when health data were largely seen as the products of clinical encounters [3]. But health is revealed in a wide range of individual behaviors that occur outside the health care system—in purchases, communications, searches, locations—and an increasing share of those activities are captured electronically where they can be linked and analyzed. These data offer promise to advance research on individual or public health – for instance in uncovering insights on manifestations and sequelae of mental health, hospital encounters, and outbreaks [4]. Public acceptability of using these data for health purposes is, however, unknown and likely dynamic [5]. The promise of applying disparate digital data to health insights sits alongside enormous practical uncertainties about logistics, acceptability, perceived and actual value.

Prior work suggests that many individuals are willing to share substantial personal information but do not like to be surprised by how their data are used [5]. The contextual integrity theory includes the idea that perceptions of privacy are based on ethical concerns that evolve over time [6]. The use of proprietary algorithms to categorize individuals on the basis of behaviors or tendencies can be viewed as ‘creepy.’ In the context of health care, prior work has found that 85% of patients who reported using social media and who were willing to participate in research also agreed to share these data sources and have them linked to their electronic health record for health research [7]. That consent was provided in the context of active patient care, where trust, and also perhaps perceptions of information safeguards, are typically high. Beyond social media however, little is known about what other digital traces patients would willingly share with health researchers, under what circumstances, and for what reason [8].

We used a deception design to credibly evaluate participants’ willingness to share data, the health relatedness of those digital data sources, and preferences associated with data sharing (e.g. desired information to receive in return and the individuals with whom participants were willing to share).

Methods

Aim, design and setting

Patients seeking care in a high volume, urban, academic Emergency Department from July to November 2017 were approached by research assistants for study participation. Excluded were patients 1) < 18 years old, 2) non-English speaking, or 3) High acuity and Trauma Level I. Patients were asked to participate in a survey about digital data and informed that consenting participants would be eligible for a 1 in 50 chance of winning a $40 gift card.

Survey instrument

The 18-question survey (Additional file 1 of the supplementary material), administered to participants on a laptop or tablet, had 5 core components: willingness to donate data for health research (now and at death), health relatedness of digital data, prior experiences with data privacy, data sharing preferences and concerns, and demographic information.

We asked participants about 19 different types of digital data: Facebook, Twitter, Snapchat, Instagram, EMR, genetic data, prescription history, fitness trackers, credit card purchases, tax records, online purchase history, Google searches, music streaming, Yelp reviews, rideshare history, GPS data, email and text message data. These data types were chosen based on a larger project that the Center for Digital Health is conducting.

In an IRB-approved deception design, participants were asked if they would consider donating any of the 19 different data types to health researchers, and were told that if they selected “Yes” that they would be directed to do so immediately, to simulate an actual real time response. Upon finishing this question block, participants were informed that they would not actually be donating their data, and were directed to subsequent survey questions. Participants used a 5-point scale to report how strongly they felt that various types of digital data contained health-related information.

Participants were asked what data they might choose to donate to researchers, what concerns they would have about data donation (e.g. fraud, abuse, misidentification), and who (e.g. friends, family, physician) they would want to have access to their information [9].

Analysis

Descriptive statistics were used to characterize each of the components of the survey. Exploratory factor analysis (EFA) was conducted to identify clusters of different data sources grouped according to participants’ sense of health-relatedness and willingness to share. EFA was conducted in R 3.5.1 using Parallel analysis [10, 11] comparing the scree of factors of the observed data with that of a random data matrix of the same size.

Results

Of 595 people approached, 206 (35%) consented to participate. Participants were primarily young, female, African-American, and lower income (Table 1).

Table 1 Participant Characteristics
Full size table

Willingness to donate digital data for health research

Participants’ willingness to share 19 specific types of digital data with researchers at the time of the survey and after death are reported in Fig. 1. One hundred four participants (65%) agreed to share at least one digital data type listed in the survey. Participants were more willing to share digital data after death for all data types.

Fig. 1
figure 1

Percentage of Patients Agreeing to Donate Data Now & After Death. This figure indicates the proportion of patients who agreed to donate each data type when approached in the Emergency Department, and the proportion that indicated that they would be willing to donate each data type after their death

Full size image

Factor analysis revealed 6 discrete themes grouping different types of data according to patient willingness to share (Table 2). Based on the dominant content of these data, we interpreted these groupings as Health/location, Social Media, Other activities, Politics, Communication, Financial. Additional file 2: Figure S1 (in the supplement) shows the percentage of patients who reported using the indicated devices or accessing the type of data listed.

Table 2 Factor analysis ‘willingness to share’
Full size table

Health relatedness of digital data

Figure 2 reports participants’ assessments of the health relatedness of different data sources. Of note, Google search histories, data from wearables, and email were considered more health related than genetic data.

Fig. 2
figure 2

Patient Views on the Health Relatedness of Data Types. This figure represents the proportion of participants who answered either “neutral” or “agree” to whether each of the indicated data types is related to health

Full size image

Factor analysis revealed 5 discrete themes grouping different types of data according to perceived health relatedness (Table 3). Based on the dominant content of these data, we interpreted these groupings as Social Media, Health, Financial/location, Apps, Communication/commerce.

Table 3 Factor analysis ‘relatedness to health’
Full size table

Data sharing preferences and concerns

Patients were most interested in receiving feedback about potential risk factors 139 (67%) gleaned from their data; 155 patients (72%) wanted the information shared with themselves and 111 (51%) with a health care provider. Only 8 individuals (4%) said they would share health insights with their social network (Table 4).

Table 4 Preferences for data sharing and data concerns
Full size table

Patients also expressed concerns about potential data and privacy breaches; a majority were concerned that friends online might inappropriately disclose private information to others 115 (56%), that they might be defrauded online or their personal information would be abused 149 (73%), that companies might share information with third parties without consent 153 (74%), and that companies and websites might use their information in ways not stated in the privacy policies or user agreements 149 (72%).

Discussion

This study has three main findings. First, patients in this study were willing to share several non-traditional forms of data with health researchers now and even more so after they have died. Second, a non-trivial percentage of patients recognized that digital footprints left in non-health areas such as finance or commerce may reveal information about their health. Third, these patients have preferences about what health related insights they would want to learn from their digital data and with whom they would want to share this information, and potential pitfalls of digital data sharing.

Participants were willing to share many types of digital data with researchers, some revealing a willingness to share presumably sensitive data like tax records and credit card purchases. These financial data sources may be highly predictive of health and health outcomes [12]. There are many steps however between sharing and actionable information [13, 14]. Each data source provides different signal, and the extent of the potential signal is likely mediated by the amount of data shared, and how individualized that data are. A growing literature addresses correlations between digital data and health outcomes and health care utilization [15,16,17,18,19,20,21,22]. Much of this research relies on participants sharing personal data with researchers. Less is known however about patients’ perceptions about how connected these data are with their health.

The connection between many of these data sources to health is often obvious and many of those health connections were frequently recognized by study participants. And yet, regulations protecting the privacy of health information are defined not by health-relatedness, but by information source [23]. Health-related information arising in the context of clinical care is highly protected under the Health Insurance Portability and Accountability Act. Health-related information arising in the context of consumer purchases or social media use is not. And yet in some cases that latter was perceived as more health related than the former.

The emergence of direct-to-consumer genetic testing sites like 23andMe [24,25,26] can reveal predictive or suggestive information to patients that they may or may not want to know. For example, when considering feedback from genetic research, 87% of participants agree that they would want to have findings shared with them if researchers found that they had a genetic pattern linked to a life threatening condition, which was manageable or curable, 73% if the condition was not life threatening, and 72% if the condition was life threatening but not curable [27]. We found similar percentages when we asked survey respondents if they would want to know if patterns in their digital data indicate that they had a higher than average risk for a treatable disease (85%). When asked if patterns in their data indicated that they had a higher than average risk for a non-treatable disease 75% would want to know, and if patterns in their data indicated that they had a lower than average risk 74% would want to know.

In April 2018, it was revealed that the firm Cambridge Analytica had accessed the data of more than 80 million individuals’ Facebook accounts without their permission. The public response was considerable, with many concerns raised about data privacy and what companies know about individuals and the types of information they share online. This type of large-scale privacy violation has an impact on the trust people have in the security of their digital data, and some people reportedly deleted their social media accounts after this occurred [28]. Of note, a large proportion (74%) of patients surveyed in this study had expressed concern that companies might share information with third parties without consent, a full 6-months before the Cambridge Analytica activities were reported in the media.

As researchers gain greater insight into the relationship between online activity and an individual’s health, transparency of these findings is essential to maintain trust. Increasing focus on returning research findings to patients is evident in the digital era where there is a movement toward open science and better patient engagement [29].

A better understanding specifically of health-related digital footprints is important for being able to provide guidance to patients about their use of digital platforms and sharing practices. This emerging field is in its infancy as many of the most popular social media and online sites have only been available for slightly more than ten years.

While providing data back to patients would be a first step, future work would also focus on the utility of this data being provided to healthcare providers via an EMR. Less defined is how this data would be interpreted, or used, or if it would even be welcomed. Regular reports of patients’ steps walked, calories consumed, Facebook status updates, and online footprints might create overwhelming expectations of regular surveillance of questionable value and frustratingly limited opportunities to intervene even if strong signals of abnormal patterns were detected [30]. This future work could assess healthcare providers use of digital data incorporated in an EMR and focus on issues related to the accuracy, interpretability, meaning, and actionability of the data [31,32,33,34,35].

This study has limitations. The findings are exploratory and represent a small sample size from a non-representative population. Response rate may have been influenced by patients being queried in a medical environment and could vary if patients were asked in non-hospital settings. This study also has strengths. Because we told patients that we would immediately access their data should they be willing to share it, their willingness to share more likely represents true preferences, rather than merely the expressed preferences of a typical hypothetical setting.

Conclusions

Patients use a variety of digital applications that generate large amounts of data. Our work demonstrates that participants would be willing to donate some of their digital data to researchers and clinicians in pursuit of health-related insights. This work adds to the larger domain of privacy and health research by connecting various digital data with perceived health relatedness. Both the willingness to share data and the perceived relatedness of those data to health do not follow conventional divisions on which health information privacy policies are built. Future work should be directed towards understanding the contexts in which patients are most likely to donate data for research use, and how they would want insights shared with them.