 |
 |
 |
 |
| |
|||||
|
|
|
Home > News & Events > MITRE Publications > The MITRE Digest > | |||||||||||||||||||
| Unlocking Medical Secrets While Safeguarding Patient Privacy March 2008
More patient records and other health data are now housed in electronic databases than ever before. This increased portability of electronic medical data presents a conundrum to doctors, hospitals and insurance companies. Patients are understandably nervous about protecting the privacy of their medical information, yet making certain health data more readily available to medical researchers could help in tracking the spread of contagious diseases and in unlocking the secrets to a host of other ailments. Today, because patient records contain sensitive details such as Social Security numbers, they are largely unavailable to researchers due to concerns about identity theft and other privacy issues. Removing personal identifiers like names, addresses, phone numbers, and insurance ID numbers from patient records—particularly as more and more such records become available electronically—enables them to be used by statisticians, epidemiologists, and other researchers in developing disease trend hypotheses and monitoring drug side effects. The ability to search large collections of patient records will help experts answer questions such as which drugs are most effective in treating a given ailment, and whether medical devices are more effective in certain subsets of patients, explains MITRE's John Aberdeen, an artificial intelligence engineer. He is part of a team of researchers who hope to make it easier for doctors to access anonymous electronic medical records for study purposes. Aberdeen's team has achieved significant progress in developing automated tools and techniques for removing personal data from patient records. Tag, You're It The effort is particularly timely because of a presidential directive that all Americans should have access to their medical records in electronic form by 2014. While the public policy push towards electronic health records is partly related to potential cost savings and improvements in patient care quality, an additional benefit is that electronic records would be much easier to search—provided that patient privacy concerns are addressed. MITRE's techniques for named identity identification show great promise for eliminating personal markers from electronic records in a process called "free text de-identification." MITRE has developed text mining tools based on named identity identification that find personal information, based on clues such as words beginning in capital letters that might be names. These personal identifiers can then be stripped from the document, leaving other critical information—such as details about symptoms and medical test results—intact. The centerpiece of MITRE's ongoing development effort in this area is a de-identification tool built using MITRE's Carafe toolkit. Carafe provides a framework for labeling and segmenting sequential data—in this instance sequences consisting of words within medical documents. We are at work on embedding this tool into an interactive interface, so that developers and non-developers in the medical community can install and tailor it to their needs. Developing Tools that Can Evolve with Research Goals "Our goal is to develop a flexible system that performs at a very high level of accuracy, protecting patient privacy while accommodating changing research goals and metrics," explains Ben Wellner, a MITRE artificial intelligence engineer who helped develop Carafe. MITRE's de-identification system achieved the top score in the recent American Medical Informatics Association De-identification Challenge, beating out seven other academic and commercial research teams in a competition to determine which system was most accurate and comprehensive in stripping personal identifiers from patient records. Some 1,000 medical discharge records (altered to hide any real patient information) were used in the challenge, Wellner says. MITRE and the other competitors used their de-identification systems to strip patients' and doctors' names, hospital and nursing home names and addresses, phone and fax numbers, and, in some cases, patient ages from the records. To make the task more challenging, "ambiguous" terms such as "Dr. Kidney" were added to the information in the records, he says. MITRE's de-identification system earned its winning score by masking 97 percent of personal identifiers in the records used in the challenge (recall). Of the items masked by MITRE's system, 98 percent were personal identifiers (precision). The MITRE team is working to bring these numbers up further, to minimize the need for human review of records. According to Wellner and Aberdeen, the impact of this work goes beyond medical records. MITRE's research shows promise for reducing the risk of accidental disclosure of personal information in a variety of contexts. For now, however, the team continues to focus on improving the de-identification
system's capabilities with medical records. "Having the ability to aggregate
and analyze medical information in new ways will enable our sponsors to
identify emerging health threats much more quickly than ever before,"
says Aberdeen. —by Maria S. Lee Related Information Articles and News
Technical Papers and Presentations
Websites |
||||||||||||||||||||
| Page last updated: March 31, 2008 | Top of page |
Solutions That Make a Difference.® |
|
|
