Mining Medical Records to Improve Health Care

A patient’s medical record typically contains a vast amount of information, some standardized for use in computer information systems, but much written unstructured in free text.

Özlem Uzuner

How best to mine the text and make it readily accessible is the challenge being addressed by Özlem Uzuner of the Department of Information Studies.

In collaboration with Partners Healthcare in Boston, Uzuner is developing natural language processing technology to automatically extract information from narrative medical records. These records contain comprehensive information about patients’ symptoms, diseases and treatments. The technology that underlies her work is the same that underlies the IBM Watson computing system. Computing power is harnessed to rapidly process information to better understand the meaning and context of human language.

The ability to readily access the information in narrative records has a number of potential benefits. It can provide more complete information to all involved in a patient’s care. Sometimes, for example, a reaction to a prescription might lead to a hospital emergency room visit, but all the records relating to this may not be linked and accessible. For researchers, the information access can provide evidence of clinical trends, adverse events, and disease incidences.