Accurate records are key to a safe, well-run health care system. But getting databases to communicate well even within the same system can be an uphill battle. And when they do talk, there must be processes in place to ensure that the entry for a 鈥淛ohn Q. Public鈥 in one database is the same as the 鈥淛ohn Q. Public鈥 in another. Otherwise, medical mistakes can happen, and John Quincy might get the drug meant for John Quentin.
That challenging problem is the focus of a new project led by , an associate professor in 911爆料's . Funded by the , the project will examine how researchers assess and report the uncertainty that arises when records from different sources are linked.
In the FDA's case, that could mean combining electronic health records from medical providers with Medicare claims, pharmacy records, and other sources. Linking those sources can give researchers richer information about a patient's experience with a drug, medical device, or treatment.
But doing that accurately can be difficult, and small details can make a big difference. For example, a person's first and middle names might appear in one database but only a first name in another. An address may change. Other identifying information may be missing or inconsistent. Two people, such as a father and son, could even share the same name.
鈥淵ou have a high probability when names appear to match, that those are the same person, but there is no way to confirm,鈥 Betancourt said. 鈥淚f important information, such as date of birth, is missing, you cannot distinguish between them. What I want to do is evaluate how the linkage quality and the uncertainty that comes with it is evaluated, reported, and incorporated into regulatory inference.鈥
Record linkage was included in an FDA announcement seeking proposals because the issue cuts across many of the agency's responsibilities, from medical devices to rare diseases. The FDA was interested in data linked by outside sponsors, such as pharmaceutical companies conducting post-market surveillance of their products.
Betancourt said the agency wants the resulting framework to help those organizations understand what information they should report and what potential problems they should consider when conducting linkage studies.
鈥淲e want to share it with sponsors to guide them through what they need to be reporting. It will tell them what to take into consideration when they are doing their linkage studies,鈥 she said. It would also provide FDA reviewers with a way to evaluate those studies and determine whether uncertainty in the linkage could affect their conclusions.
The problem extends well beyond any single type of medical data. Health care providers themselves may use different electronic record systems at different locations, making it difficult to connect information even within the same health system.
鈥淪ometimes each location even has a different electronic health record system that is not standardized,鈥 Betancourt said. 鈥淵ou can go to the same big umbrella health system, go to two different locations, and your data is not going to be synced.鈥
A universal identifier across databases would make matching records far easier, but often no such identifier exists. That leaves researchers piecing together records using information that may itself be incomplete or inconsistent.
Betancourt will begin the FDA project by examining the record-linkage methods currently being used in studies submitted by sponsors. That review will help determine the statistical approaches she uses later in the project.
Which approach proves most useful will depend on what Betancourt finds when she digs into those records. When medical decisions depend on getting John Quincy's records rather than John Quentin's, the devil, as they say, is in the details.