|
|
| |||||||||||||||||||||||||||||||||||
Achieving these goals requires the first large-scale, truly integrated use of the electronic data that are increasingly available in our pluralistic health care system. The first hurdles are determining how best to organize the Sentinel Network, how it should operate, and what steps are needed for its implementation.
Initially, the Sentinel Network will rely on electronic medical records and administrative data that are routinely collected by medical practices, hospitals, delivery systems, health plans, and insurers. Medicare and Medicaid databases of prescriptions and other information on the use of medical resources could be an important large-scale data source. Eventually, the network might also use information from disease registries and vital-statistics registries, as well as repositories of genomics data.
The goals of the Sentinel Initiative can be achieved without the creation of a large centralized database. In a distributed data network, participating organizations would maintain control of their data, create data files in a standard format, summarize data by running computer programs distributed by a coordinating center, and then provide consistent summarized results that could be combined to provide networkwide results. For example, the programs could identify persons within each organization who had been exposed to specific drugs, compute case-mix–adjusted rates of adverse events, and then compute comparative rates for similar populations that had received different therapies. When additional information is needed, such as the confirmation of a diagnosis through the review of full-text medical records, the individual data owners would be able to provide it.
The distributed model has several important advantages, including the fact that nearly all confidential personal health data would remain with the clinicians or other original data holders. This model also allows the people who are the most knowledgeable about the participating health care and informatics systems to ensure that their data are used and interpreted properly.
The prospects for developing a distributed network for safety surveillance that includes data from 100 million people or more appear to be good. Leaders of both public and private health care organizations have shown a high level of interest in participating. And some building blocks already exist. For example, the Vaccine Safety Datalink (VSD) project of the Centers for Disease Control and Prevention conducts near-real-time monitoring of new vaccines with the use of a distributed network that combines information from both electronic medical records and administrative databases covering nearly 9 million members of eight health plans. This system recently identified an excess risk of seizure after the administration of the measles–mumps–rubella–varicella vaccine, a finding that led to a change in the recommended use of the vaccine.4 The VSD project also conducts full epidemiologic studies to investigate concerns arising from other sources. In addition, distributed methods are being used to assess data from 50 million people belonging to five health plans in order to determine whether there is excess risk of the Guillain–Barré syndrome after receipt of the meningococcal conjugate vaccine.5
Although these and other efforts suggest that the Sentinel Initiative is feasible, there will be technical challenges. The Sentinel Network will require the capacity to apply consistent methods across multiple data sources, and it will need improved statistical methods and data-analysis tools for distributed databases. In particular, the network's signals — drawn from very large-scale observational analysis — will require further evaluation. It will therefore be essential to improve our ability to determine quickly which findings reflect genuine health risks. Making such determinations may require analyses of follow-up data on cases, as well as epidemiologic investigations like those used to evaluate disease outbreaks. In addition, further clinical studies may be required to reach conclusions about whether a medical product caused or was simply associated with the adverse event.
These technical issues suggest that the Sentinel Network may be able to achieve some objectives sooner than others. For instance, it may soon be possible to monitor rare but serious events that generally represent drug reactions, as well as certain more common, reliably coded events such as myocardial infarctions. However, more work will be required to accurately identify product exposures that are harder to detect from current electronic data — such as the use of over-the-counter drugs, drugs used in operating rooms, and many medical devices — as well as outcomes that lack uniform standards for electronic coding.
The Sentinel Network will also require governance structures and processes to ensure that the maximal public health benefits are achieved while maintaining public trust. Specific requirements include systems for prioritizing needs, for designing and implementing studies to address those needs, and for protecting personal medical data when they are used for public health purposes.
Finally, the network's findings must be communicated in a timely, transparent, and appropriate way to a range of audiences (including health care providers and the public) who are often frustrated by delays in the availability of information related to postmarketing surveillance but are also confused about what to do when new (and often not definitive) evidence is made available. Health care providers will need information about the system itself and the methods and findings of individual studies, as well as guidance on interpreting observational studies of drug safety and effective approaches for discussing them with patients. Priorities for communication with consumers and patients include finding ways of placing new information about risks in the context of current knowledge about benefits (given that no medical treatment is completely safe), specifying the degree of certainty in particular findings, addressing the limitations of observational data, and identifying the next steps for developing and communicating more definitive information.
The FDA is working with a broad range of stakeholders to further define the Sentinel Network strategy and begin its implementation, probably initially on a pilot basis. In December 2008, a public workshop on directions for the network was attended by 400 stakeholders from the FDA and other government agencies, academia, health plans, health care provider organizations, consumer- and patient-advocacy organizations, and the pharmaceutical and medical-device industries. More recently, a number of organizations in the public and private sectors have completed initial feasibility work under contracts from the FDA. With continued transparent and rigorous development, the Sentinel Network has the potential to substantially improve the quantity, quality, and timeliness of evidence regarding the safety of medical products in the United States.
The authors report advising the FDA about the design of the Sentinel Initiative and may submit proposals for contracts for its implementation. No other potential conflict of interest relevant to this article was reported.
Source Information
From Harvard Medical School and the Harvard Pilgrim Health Care Institute, Boston (R.P.); HealthCore, Wilmington, DE (M.W.); i3 Drug Safety, Waltham, MA (K.A.C.); Engelberg Center for Health Care Reform, Brookings Institution, Washington, DC (J.S.B., M.M.); and IBM, Fairfax, VA (J.M.).
This article (10.1056/NEJMp0905338) was published on July 27, 2009, at NEJM.org.
References
| |||||||||||||||||||||||||||||||||||
This article has been cited by other articles:
HOME | SUBSCRIBE | SEARCH | CURRENT ISSUE | PAST ISSUES | COLLECTIONS | PRIVACY | TERMS OF USE | HELP | beta.nejm.org Comments and questions? Please contact us. The New England Journal of Medicine is owned, published, and copyrighted © 2010 Massachusetts Medical Society. All rights reserved. |
