Data-Sharing Efforts

Our review revealed that multiple efforts from the public and private sectors contribute to the ecosystem by collecting — either directly from individuals or from intermediaries — and distributing human data. These efforts varied, depending on how broadly they shared and whether they directly collected the shared data or simply aggregated data from other sources. Here we describe the types of data-sharing efforts we identified.

Controlled- and open-access data-sharing initiatives (n=50, 15%) collected biospecimens and data from individuals and shared them through open- or controlled-access portals. Controlled-access initiatives used a gatekeeping process that required a certain action, such as signing a data-sharing agreement, to access data. Some data-sharing efforts offered tiered access where some data were openly available and other data were controlled access. Individuals were usually recruited to share their data via these initiatives through: 1) general biomedical research; 2) specific research protocols; 3) disease-specific registries; and 4) open databases. In keeping with traditional research norms, these efforts typically operated under Institutional Review Board (IRB) or other ethics review body oversight, collected biospecimens and data, and solicited consent from participants to share data with other researchers either broadly or in keeping with a set of funder policies. Disease-specific registries typically recruited individuals to share data for general research, as opposed to a specific study protocol.

In a departure from traditional research norms, a number of data-sharing initiatives relied on individuals to contribute data received from direct-to-consumer (DTC) testing or smart-phone applications linked to wearable devices and often enrolled individuals through a website. Individuals were able to become end users, taking on the role of a researcher (i.e., citizen scientist).⁹ Open databases, such as open-SNP and Open Humans, enabled such uses since data were available for anyone to access. Open Humans also gave data contributors the option to upload a picture and a write-up about themselves, in addition to sharing data through direct, private communication with a data seeker. Open databases typically did not operate under IRB research oversight, did not fall under the purview of health data privacy laws such as the Health Insurance Portability and Accountability Act (HIPAA), and did not require formal consent for participation. Individuals assumed all risks, including the risk of personal identification. As the openSNP website was noted to state, “There is zero privacy anyway, get over it.”¹⁰

Selective data-sharing initiatives (n=9, 3%) provided data access but only to limited datasets, primarily sharing aggregated data analyses from one or a few thematically linked sources. For example, Fabulous Ladies Over Seventy (FLOSSIES) a public, web-based database indicated on their website sharing some demographic information along with allele frequencies of genes linked to breast cancer. The study population in the database biospecimens cancer-free women over the age of 70 whose biospecimens were collected as part of the Women's Health Initiative, and some data were made publicly available through the FLOSSIES genetic variant database.¹¹

Closed consortia (n=77, 24%) consisted of multi-institutional research collaborations (often between industry and non-profit organizations) that primarily shared data internally, among consortium members. For example, the website for the company Patient-sLikeMe had an online forum where patients connected with each other (researchers were also invited to access the forum). However, data sharing was noted to occur through research partnerships: “We do not participate in studies in which we simply provide data to researchers. We prefer to be active collaborators in the implementation of the research protocols for which patient data will be analyzed.”¹² Most closed consortia provided limited or no information on the terms of their collaboration agreements on their web-sites; therefore, we have limited information about their data-sharing practices. An exception was the CHARGE (Cohorts for Heart and Aging Research in Genomic Epidemiology) Consortium, which posted on their website the consortium's data-sharing policy and expectations of its members. CHARGE noted that some data were shared with external researchers by depositing datasets from published reports into dbGaP.¹³

Aggregators (n=84, 26%) combined data from multiple datasets, analyze the data, and publicly share their analyses. While some aggregators solely pooled data from published studies and/or existing datasets, others also offered mechanisms for voluntary data submissions. Both types of aggregators shared summary statistics or metadata analyses, often through a public, online browser that facilitated interrogation of the data. The University of California Santa Cruz (UCSC) Genome Browser was classified as an aggregator because the data available were combined from multiple sources, such as the laboratories participating in the Human Genome Sequencing Consortium, and the UCSC Genome Browser provided a publicly and freely accessible visualization of the integrated, analyzed data.¹⁴

Repositories (n=20, 6%) served primarily to house datasets from multiple sources and engage in data curation to ensure datasets were adequately de-identified and met data submission criteria. Aggregators shared certain features with repositories, but repositories typically did not take the additional step of combining and analyzing the data. Repositories stored and shared individual datasets contributed voluntarily or in compliance with a funding agency mandate and offered data through controlled-, open-, or tiered-access (i.e., some data open, other data through controlled access with a gatekeeper). A few efforts did not fit neatly into a single category. For example, the NIH's GenBank DNA sequence database had both aggregator and repository features, which included archiving data, synchronizing data daily with other U.S. and international databases, offering a mechanism for researchers to submit data, and publicly sharing aggregated outputs.¹⁵

Data-Sharing Facilitators

The following sections describe efforts that facilitated the flow and use of data in an MIC ecosystem by connecting researchers and other third parties to existing data sources, providing critical infrastructure, and developing standards.

Brokers (n=16, 5%) connected individuals to specific kinds of data. Some brokers were structured as federated networks, where data contributed by multiple efforts were funneled to a centralized database for the purposes of data interrogation, and the individual data contributors retained control over access to the datasets. We classified the Global Alliance for Genomics and Health (GA4GH) Beacon Network as such a broker because the Beacon resource had the ability to query the multiple participating databases to find out if any contained information on a particular genetic variant.¹⁶ Taking the brokering function further was the Global Alzheimer's Association Interactive Network (GAAIN). GAAIN described its data resource as providing output of analyses on data from various U.S. and international research efforts, including the Canadian Longitudinal Study on Aging and the Japanese Alzheimer's Disease Neuroimaging Initiative, that GAAIN formatted based on standards established by the Clinical Data Interchange Standards Consortium for the purpose of data exploration and hypothesis development, as the contributing data initiatives themselves granted access to the original datasets.¹⁷

Other brokers connected researchers with individuals who may want to share data or participate in research. MyGene2 was classified as this type of broker. Through its website, MyGene2 indicated that clinicians and researchers who have contributed information on individuals with a rare condition are given access to phenotypic information for other similar individuals and connected to families of those affected, as well as to other clinicians and researchers.¹⁸ MyGene2 also noted its participation in the GA4GH Beacon Network. We also classified we classified Portable Genomics' NuMe as a broker. NuMe, a for-profit enterprise, that offered a mechanism to connect researchers to data from NuMe customers. NuMe is noteworthy because the company indicated that customers would receive monetary compensation if their data were to be shared.¹⁹ This contrasts with traditional research norms, according to which participants are not invited to share in any profits from the use of their data.

The repositories described above conveniently provided a “one-stop shop” for data from multiple sources. This abundance of data spurred the development of independent cataloging efforts, or indexers (n=9, 3%). Indexers functioned as a directory, typically offering a publicly-accessible, web-based query through which users could identify which existing resource, such as ClinVar and dbGaP, to access for information of interest. CLINVITAE, for example, a free and public resource developed by the genetic testing company Invitae Corp, directed users to public databases containing information on genetic variants of interest.²⁰

Data analysis tool providers (n=33, 10%) offered discrete products to help researchers and others not specialized in data analysis to interpret genomic and health data. Their tools were intended to help researchers manage, analyze, and understand large datasets with simple, user-friendly software, often through data visualizations. An example is the company Seven Bridges, with its portfolio of products including the Seven Bridges Platform. The Seven Bridges Platform, available to U.S. and international research projects, was recorded as a data analysis tool provider because of its provision of bioinformatics analyses of large genomic datasets via Amazon Web services and Google Cloud.²¹ The National Cancer Institute's Cancer Genomics Cloud has implemented the Seven Bridges Platform.²²

The sharing and storage of vast amounts of data was supported by infrastructure providers (n=17, 5%), who provided a digital space to host data-sharing projects. Synapse Commons, developed by Sage Bio-networks, a non-profit organization, was one as such provider. Synapse has established general community guidelines concerning data privacy and security, while each research team using Synapse has the discretion to determine whether to offer data access to all or some registered users.²³ Other similar projects, such as the Platform for Engaging Everyone Responsibly (PEER) created through a partnership between a non-profit patient advocacy organization, Genetic Alliance, and the company Private Access, Inc., provided platform-enabling technology that can be used by research efforts. PEER has been recognized for its user-friendly interface designed to facilitate self-governance by giving individuals dynamic and granular control over data sharing.^{Reference Shelton24} The HEROIC Registry for individuals with hereditary cancer, which we classified as a data-sharing initiative, integrated the PEER platform to achieve its mission of providing a “patient-centric genetic database.”²⁵

Finally, data-sharing requirements imposed by governmental funding agencies, such as the NIH via its GDS Policy, and guidelines developed by professional organizations, such as the International Committee of Medical Journal Editors (ICMJE) proposal on the sharing of clinical trial data, are (or in the cases of the ICMJE proposal, could become) standard policies shaping and regulating the flow of biomedical research data.^{Reference Taichman, Sahni, Pinborg and Peiperl26} Beyond these organizations, we identified work by policy and guideline developers (n=12, 4%), frequently international, who facilitated the development and establishment of stakeholder-informed policies and guidelines that promote data sharing. The GA4GH, Human Variome Project, and Public Population Project in Genomics and Society have developed data-sharing guidelines informed by different global perspectives on challenges and best practices.^{Reference Knoppers27}

Data Sources and End Users

While we did not attempt to quantify the data contributors and end users, we recognize their important role in sustaining the flow of data and creating value from data. As described in the sections above, the data-sharing ecosystem relied on four main contributors of data: 1) individuals, where data entered the ecosystem directly from the human beings the data describe, for example, through a DTC testing consumer's contribution of her data to an open-access data-sharing initiative, or a patient's use of her HIPAA access right to make a similar direct contribution of data from her medical record; 2) healthcare providers and health plans, where, for example, a healthcare system contributed data from medical records including DNA-derived data, or medical records were linked to DNA-derived data generated for research purposes, or a clinical laboratory contributed genetic testing results and related information, such as the basis for a variant call; 3) researchers, where data entered the ecosystem from datasets maintained by individual investigators who led laboratories or research organizations such as pharmaceutical companies and biorepositories that generated data from biospecimens; and 4) publishers, where datasets corresponding to manuscripts were made available to actors in the ecosystem, including aggregators and indexers.

The main end users of the data-sharing ecosystem were: 1) researchers accessing data for use in a scientific study who possessed the conventional bona fides for access to data maintained by controlled-access data-sharing initiatives including an advanced degree in a relevant scientific discipline, a research protocol, and an institutional affiliation entailing some degree of oversight (which could encompass IRB approval); 2) clinicians accessing data for patient care purposes; and 3) citizen scientists, here meaning members of the general public with an interest in engaging in genomic research, as well as researchers who would not be considered qualified to access data according to common controlled-access criteria (e.g., laboratory-based researchers interested in variant information but without a specific research protocol, researchers with limited resources outside institutional settings and lacking access to an oversight mechanism).^{Reference Villanueva, Cook-Deegan, Robinson and McGuire28}