The U.S. Food and Drug Administration (FDA) is increasing its capability to harness the power of big data – or more specifically of the “real world evidence” (RWE) – to assess the safety of medical products that are approved for the U.S. market. To that end, FDA is expanding its Sentinel System so that by the end of 2023 it would represent “a transformative, multi-purpose national data and scientific resource center for evidence generation that a wide array of stakeholders use to inform all aspects of healthcare decision making,” according to the recently released Sentinel System Five-Year Strategy.

The Sentinel Initiative was first started about 10 years ago, to fulfill the legislative mandate under section 905 of the FDA Amendments Act of 2007. The main goal was to establish a computerized safety surveillance system for drugs on the market, in order to augment the existing FDA post-market surveillance activities.

The full implementation of the Sentinel System was launched in 2016. Its distributed data system integrates data from 18 partner institutions, including Aetna, Blue Cross Blue Shield of Massachusetts, Duke University, Kaiser Permanente, HCA, and Optum, among others, while preserving patient privacy. The system allows FDA to evaluate safety risk signals from RWE more efficiently than before; and if such evidence is deemed sufficient to address a potential safety risk, it may even obviate the need for biopharmaceutical or medical-device companies to conduct post-market safety studies.

FDA continues to evolve the tools and methods used by the Sentinel System to identify and evaluate the risks. Important insights can be gained by combining data from different sources (e.g., distributed health networks), by refining the data analysis methods. For example, linking mother and infant data enables evaluation of the impact of fetal exposures to medications. As an example of the method-based advance, the use of machine learning and natural language processing created a way to identify instances of serious allergic reactions to medications, e.g., anaphylaxis.

At the same time, the agency still continues to utilize traditional sources of potential safety signals, such as reports generated by medical product sponsors, patients or health care providers; medical literature and lay media; clinical trials; and Adverse Event Reporting System. In contrast, Sentinel System could not only integrate the input from different sources but actively detect potential signals, by analyzing health outcomes from large populations in a large set of real-world scenarios.

In the next five years, FDA plans to expand data sources and linkages, test and improve its algorithms, increase data diversity, build up data-mining capabilities, leverage advances in data science and in safety-signal detection, broaden the Sentinel System’s user base, and encourage innovation and collaboration. For example, conceptually, FDA envisions eventual open access to the Sentinel System by patients, providers, academia, industry, other U.S. agencies, payors, and non-U.S. health authorities.

Such a broad-spanning system, once implemented, may enable qualitatively new improvements in health care. On the other hand, the challenges in designing and managing the system will mount with every additional layer of complexity. Along the way, therefore, significant investment in time, resources, and public engagement will be needed both to develop, test and maintain the sophistical technical solutions and to ensure appropriate privacy and security protections.