In this review, we defined PHR as an electronic record designed for patients to self-manage care [6]. Thus, we focused on data that were either entered by or transmitted to the patient to enable self-care management, regardless of PHR type or brand.

We considered US studies from 1950 through 2015. We limited our search to US-based studies because of variation in ontologies and legal and privacy frameworks across countries. Because we were interested in specific data content available in the PHR, rather than patients’ extent of system use, we excluded articles focusing on PHR adoption. Furthermore, we excluded articles containing data intended to be used only by health care providers, and that not to be viewed by patients. For example, articles reporting on physician use of patients’ hormone levels to assess risk factors and clinical outcomes were excluded from the analysis because this information was not intended to be used or viewed by patients. Finally, after reviewing the body of articles selected based on title and abstract, we excluded articles that focused on general concepts and did not mention specific data elements present in the PHR.

To conduct our review, and using used PubMed’s Medical Subject Headings (MeSH) database as our starting point, we identified 5 search phrases referring to the PHR: (1) personal health record, (2) personal medical record, (3) patient health record, (4) computerized patient record, and (5) personal electronic health record. A search of eligible US studies on PubMed from 1950 to 2015 and on Embase and MEDLINE from 1966 to 2015 using the previously defined phrases resulted in 1822 articles (Figure 1). The search results comprised articles containing any of the search phrases in all fields including titles and abstracts.

Title and abstract screening based on the inclusion and exclusion criteria by 2 reviewers (authors LB and AI) resulted in consideration of 334 articles. Data elements, associated data sources, and analytics techniques were described. The reviewers met after screening every 20 articles to compare results and adjudicate. Consensus was reached regarding (1) the final list of articles to be considered for full-text screening and (2) information extracted from the selected articles.

The full body screening resulted in a total of 106 articles used for data element extraction. Whenever available, reviewers LB and AI recorded the following information in an Excel 2010 (Microsoft Corporation) spreadsheet from each article reviewed in this study: (1) title, (2) author(s), (3) year of publication, (4) PHR data element(s) (ie, data collected by or shared via the PHR), (5) data type(s) (character, number, string, etc), (6) platform(s) (website, app, etc), (7) data storage (Excel database, Oracle, etc), (8) data entry (manual or electronic), (9) source, (10) receiver, (11) details regarding patient use, (12) barriers and issues, and (13) benefits.

A list of all data elements extracted from the 106 selected articles was further grouped by the reviewers into major data categories. The data categories were based on a taxonomy created in a PHR systematic review published in 2011 by Archer et al [6], which served as a foundation for this work. Categories found in our review but not included in Archer et al’s review were identified and validated by a group of clinical informatics experts.

We categorized the PHR data and refined them after consultation with an informal focus group of clinicians. In cases where different terms referred to the same data element (eg, medications, pills, and drugs), we chose 1 of the terms and grouped all synonymous data elements together under this term. Metadata pertaining to PHR functionalities were extracted from the articles and categorized based on content. For instance, articles mentioning the PHR reminder functionality were listed as references for data elements such as appointment reminders and prescription reminders, and were categorized under scheduling and treatments, respectively, as opposed to grouping all reminders under an umbrella “reminders” category. Additionally, some of the PHR data elements could have been included in different categories, depending on the user’s perspective. For example, the data elements described as prevention adherence could be viewed by the patient as part of a prevention plan but perceived by the provider as compliance with recommended health procedures and activities. We refined and ordered data categories listed in the results table based on their typical sequence of patient health care delivery. For example, scheduling data were listed before treatment data, which were listed before outcomes.

Following PHR data extraction and categorization, we performed a cross-categorical analysis of the data by percentage, source, and format. Additionally, we completed a longitudinal analysis of the time of first mention of the data element in the literature.

The literature review identified 13 major categories of PHR (Multimedia Appendix 1 [17-117]). At least one data element was included within each of the main categories, and details on the data elements and their corresponding references are provided. In addition to the data elements previously reported in Archer et al’s systematic review, this research identified 22 new data elements. Additionally, we distinguished 3 data elements from Archer et al’s review in the more recent PHR literature and separated them into more than one data element.

Patient data elements reported in the literature are available in Multimedia Appendix 1.

The comparable data elements identified in both reviews were personal information, problem lists, surgical history (procedures, hospitalizations), medical history (family history), provider information (provider list), allergies, home monitoring data, medical history, psychographics (social history, lifestyle), immunizations, prescription medications, and notes.

The data elements not previously reported in Archer et al and that we identified in this research were (1) genetic data, (2) preferences, (3) PHR settings, (4) facility information, (5) personalized search results, (6) visit preparation information, (7) compliance, (8) medical equipment and supplies, (9) self-treatment, (10) treatment plan, (11) outcomes, (12) patient-provider message, (13) incentive programs data, (14) patient health education material, (15) trainings, (16) personalized health advice, (17) environmental information, (18) assessment information, (19) personal health goals, (20) health care cost management, (21) insurance data, and (22) health status.

In our research, we were also able to separate Archer et al’s preventive health recommendations into (1) preventive care and (2) prevention adherence. We broke examinations and diagnoses down into (1) vital signs and anthropometric data, (2) physiological information, and (3) diagnosis. We further distinguished laboratory tests and appointments as (1) results, (2) imaging, and (3) appointments.

We grouped PHR data elements by source, format, and time of first mention. Analysis of data elements mentioned in the literature allowed for description of information available for analytics use. This analysis also revealed the capabilities available to patients through PHR systems.

Overall, the results indicate an increasing focus in the literature on newer types and sources of data, as well as on providing patients with access to their health data. Yet some of these may be progressing so rapidly that important related issues are somewhat neglected. Few studies, for instance, have examined the impact of user interface design on patients’ understanding of data and system use. Issues associated with the use of PHRs are mainly related to patients’ understanding of the underlying information presented. Problems related to understanding of health data may lead to stress and anxiety [63], which could outweigh the potential benefits of data access. Hence, research is needed in the area of data visualization and representation models specifically targeted for patient use. Examples of such models available in the literature are the what-if analysis, [99] brief intervention [109], and traffic-light feedback system [74]. These methods indicate the risks associated with specific health activities, along with related outcomes and recommended interventions. The traffic-light feedback system, for example, provides patients with an effective visualization tool to track their progress toward attainment of blood pressure goals.

In addition, more research is needed to investigate and improve the quality of patient-entered data. Today, more than 35,000 mobile health apps are available for the iOS and Android operating systems, generating large amounts of data [118]. Data are also increasingly entered through patient forums and portals. While new platforms allow the generation and availability of large data volumes, the wide variety of levels of expertise could lead to reliability and validity issues. Patient-entered data have been shown to be reliable for simple measures such as demographics and symptoms, but less reliable when they pertain to reporting more complicated measures such as laboratory values [5]. One method for improving accuracy could be to provide patients with standardized measures and guidelines for entering their own data, but even that needs to be part of a broader strategy to verify accuracy of data through triangulation from multiple sources.

As the variety of PHR data sources increases, special care is needed for data curation [119] and harmonization [120]. Processes need to be established to produce usable patient-reported data that can be used for research [121]. Standards need to be developed to improve interoperability between different components of the new PHR systems [122]. Data integration methods, such as entity stream mining [123], might be required to cross-reference patient data generated by different tools and devices.

In the coming years, PHR systems will create many data-related challenges, such as quality, heterogeneity, openness, security, scalability, and transparency. Abundant patient data might also trigger information overload. While potentially beneficial for improving health outcomes, streaming patient data can amount to very large volumes, creating new data quality, storage, and analysis issues. All of these challenges open doors for valuable research in health information systems.

The large amounts of data generated by sensors and devices might also require storage and analysis on the cloud [118], potentially increasing storage and analysis costs. Sharing patient data between networks may also create a risk of personal health information disclosure [124], generating additional costs for preserving patient privacy and security. This could also necessitate stronger methods for patient data protection beyond today’s practice, which opens up yet another important avenue for health informatics research.

Overall, PHR data evolution indicates a general trend toward greater patient engagement and health tracking. Moving forward, a continuation of these trends will lead to accumulation of vast amounts of rich data. If patients provide permission, research on PHR data can pave the way for patient-centered care.

The design of patient-centered decision support systems that use a combination of comprehensive individual patient information and aggregate data (collections of patient records) to provide personalized patient recommendations will be a significant area of research.

While past literature has listed patient-provider messaging as an important communication tool for patients and providers, secure message content may potentially provide a valuable patient data source for analysis. Based on their reported intended use, patient secure messages may contain information regarding health-related concerns such as new symptoms and adverse events. Among other possibilities, information retrieved from secure messages could, therefore, be used in research to identify treatment side effects and build patient risk models. However, it is important to keep in mind that terminology used by patients is likely to differ from terminology used by providers. Hence, natural language processing models traditionally used to extract patient information from provider notes may need to be adapted to fit the patient context.

Recently developed and highly effective deep learning algorithms could also be used to extract, search, sort, and analyze information from the tremendous amounts of image, voice, and video data [125] available in the PHR. Other new techniques might be needed to analyze relational data, such as from Google Maps and Google Calendars.

Also, current methods used to store, extract, and analyze EHR data are not adequate for analysis of large volumes of time-series data. Nonrelational databases might be needed to store tracking information. Stream learning algorithms [126] would also need to be applied to extract meaningful information from the terabytes of streaming data analyzed.

As patient-centered decision support systems are being implemented, it is important to ensure the validity of the generated output. Misclassification errors can be dangerous in this domain. Patient systems, which are embedded in mobile devices, need to be evaluated and approved by medical experts. Data transmitted from different sources can potentially be leveraged by providers to improve patient and population health outcomes. However, accurate measures are still needed to assess and improve the performance of such systems. In addition, these metrics need to account for biases present in patient-generated data. Prior research indicated that PHR systems are mostly used by patients who are typically more sick. Those are patients with comorbidities, such as cancer survivors [127]. Therefore, findings and models generated from analyzing these data might not be generalizable to other patient populations.

The new health care vision in the United States is characterized by automation and collaboration, creating the need for adaptation by all actors in the industry. Empowered patients today have the opportunity to leverage PHR systems data and functionalities. This, however, requires some level of technical expertise for system access and interaction, and medical knowledge in order to understand and interpret the medical information presented. Similarly, medical providers now have to learn and adopt new technologies in order to report medical data and communicate with patients. As a major actor in the health care industry, insurance companies also need to adapt to the new industry environment. Insurance firms today need to assess the value of virtual medical encounters and automated care, and process new types of patient data such as secure messages. Adaptation methods by all health industry players are yet to be assessed and optimized.

A limitation of this study is its focus on PHR data reported in the literature. The evolution of PHRs as described in this study might not necessarily reflect the state of the practice. More research is therefore needed to extract and evaluate PHR scope and the functionalities of the various PHR systems available in practice. Also, as mentioned above, this study focused on US studies, thereby limiting the scope of our analysis. Research comparing PHR systems in the United States with those used in other countries would help improve future data uses.

Digital health platforms have changed drastically in recent years. The introduction of distributed PHR systems enabled a shift toward more personalized and increasingly automated health care. The multiuser nature of PHR systems also facilitated patient-to-provider and patient-to-patient information sharing. Yet these changes generated opportunities and challenges at the user, system, and industry levels. Our assessment here of the state of the patient digital infrastructure serves as a valuable foundation for future research. Research implications identified also offer ways to significantly advance health information systems research. Identifying available PHR data also facilitates the development of intelligent health systems. Although primarily aimed at health information systems researchers, implications listed in this study can be further extended to health practitioners, insurance providers, and policy makers.