Multiple modes of data sharing can facilitate secondary use of sensitive health data for research

Abstract

Evidence-based healthcare relies on health data from diverse sources to inform decision-making across different domains, including disease prevention, aetiology, diagnostics, therapeutics and prognosis. Increasing volumes of highly granular data provide opportunities to leverage the evidence base, with growing recognition that health data are highly sensitive and onward research use may create privacy issues for individuals providing data. Concerns are heightened for data without explicit informed consent for secondary research use. Additionally, researchers—especially from under-resourced environments and the global South—may wish to participate in onward analysis of resources they collected or retain oversight of onward use to ensure ethical constraints are respected. Different data-sharing approaches may be adopted according to data sensitivity and secondary use restrictions, moving beyond the traditional Open Access model of unidirectional data transfer from generator to secondary user. We describe collaborative data sharing, facilitating research by combining datasets and undertaking meta-analysis involving collaborating partners; federated data analysis, where partners undertake synchronous, harmonised analyses on their independent datasets and then combine their results in a coauthored report, and trusted research environments where data are analysed in a controlled environment and only aggregate results are exported. We review how deidentification and anonymisation methods, including data perturbation, can reduce risks specifically associated with health data secondary use. In addition, we present an innovative modularised approach for building data sharing agreements incorporating a more nuanced approach to data sharing to protect privacy, and provide a framework for building the agreements for each of these data-sharing scenarios.

Health data sharing attitudes towards primary and secondary use of data: a systematic review

“…Imposing adequate privacy-protecting tools (such as deidentification and anonymization) is found to alleviate these concerns (instead of asking participants to consent to using their identified data) which can bring positive consequences to data users (such as reduced research cost, increased recruitment rates, and decreased recruitment bias). 102 On the other hand, it can bring considerable harm to the data donor in case their identity is revealed. It was evident throughout this review that transparency was an important concern and was a need that the participants expressed.…”

Data reduction when aggregating information about harms associated with medical interventions

“…For the very same reason, it is essential to rely on a lingua franca of research on medical interventions when referring to aggregate data on harms. 9 Additionally, uniformity on characterisation and format of safety data would translate into dataset harmonisation across studies, countries and sponsors, introducing several benefits 10 : federated analyses would bypass data sharing agreement while preserving individual patient’s privacy, increasing access to data and maximising transparency 11 ; linking harmonised data to pharmacovigilance repositories would be facilitated, allowing real-time contribution of multiple data sources into a synchronous environment.…”