Stay on the Frontiers of Medicine.

Sign up for a free subscription to Proto.
mgh-logo
functionbar_help
Site Help
Font Size
Large Text
functionbar_contact
Contact Us
functionbar_aboutus
About Us
functionbar_legal
Archive
Search Results for “

Not finding what you're looking for? Articles from older issues of Proto can be found here.

SEARCH RESULTS FOR “

Sorry, no results found.

Published On Apr 02, 2015

Technology

Taming the Electronic Medical Record

The QPID system uses artificial intelligence to wade through chart data, helping doctors diagnose and treat patients.

It was just a few years ago that electronic medical records were welcomed as revolutionary. Patient information would soon flow free and fast, making paper’s cumbersome inefficiencies a distant memory. 

Yet the digital reality has underwhelmed. Electronic health record (EHR) systems, while compiling reams of patient data, have proven to be clumsy and error prone. Doctors who are already pressed for face time with their patients are now spending patients' visits staring at a screen to enter or search for information.

Enter QPID, short for Queriable Patient Inference Dossier. This artificially intelligent software system is designed to distill the sprawl of a health record—which can easily run to hundreds of arcanely written, difficult-to-search pages—into bite-size, relevant details. “There’s a rapidly increasing volume of data that clinicians are drowning in,” says Michael Zalis, a radiologist at Massachusetts General Hospital. “QPID tries to address this ‘big data’ problem.” 

Zalis conceived of QPID in 2005 while analyzing CT scans. To properly translate these images into a good clinical explanation, a radiologist needs to know what has already been observed. Yet that basic information was often difficult to glean from a patient’s electronic records. The EHRs summarized patient status in a single line, composed of medical abbreviations and often containing typos or mistakes. Zalis often found himself in a deep EHR dive, poring over dozens or hundreds of pages in search of the information he needed. 

“I said, ‘This is crazy! I’m under the gun, and this is an expensive exam I’m trying to interpret in an efficient way,' ” recalls Zalis.  He dreamed of a digital agent that could take his questions, analyze a patient’s record—making use of the structured information as well as the unstructured text—and return with a rich synopsis of the most relevant details.

Zalis, who had been developing software since grade school, had just returned to MGH after studying advanced computing at Stanford University. Working with colleague Mitch Harris, he set about writing the code for QPID. Now, a decade later, the program they developed does what Zalis envisioned: It can analyze both conversational language and structured data, make logical connections, learn from its mistakes, and organize information, guided by a library of 10,000 clinical concepts.  

QPID, which is sold commercially, has been likened to a search engine, but that’s not quite accurate. “If you’re seeing a patient and you’re worried about breast cancer, you’re not just searching her record for those two words,” says Robert Wachter, a hospitalist at the University of California, San Francisco and an advisor to QPID, who has recently published a book about health care in the digital age.  “You’re looking for family history, the words lump and mammogram, for information about certain genes. Your search is conceptual and sophisticated.”

QPID brings that conceptual richness, and can also be set up to act on the information it finds. Several years ago, when the gastroenterology department at MGH was deploying the program, it arranged to have the software look at patient records for risk factors related to sedation. It then alerted the hospital about potential problems for those patients a few days ahead of their scheduled procedures. Previously, such risk factors had been identified at the last minute, or not at all.  

Today more than 7,000 clinicians use QPID, and in 2014 it handled just under 4 million clinical encounters.  And while it doesn’t yet suggest diagnoses or treatments, that could be changing. In a new pilot project with Blue Cross Blue Shield of Massachusetts, a spinoff program called Q-Guide analyzes an individual patient’s record, then provides clinical best-practice guidelines that could fit that patient’s situation. If a physician’s treatment recommendation is in sync with what the program determines is appropriate, Blue Cross Blue Shield will automatically cover the costs, sparing the doctor the administrative hassle of seeking authorization. 

Additional incentives to use QPID come in the form of improved quality reporting. Across the United States, an increasing percentage of hospital revenue is tied to the compiling of complex quality scores, many derived from existing federal programs, such as the Physician Quality Reporting System (PQRS) and Value Based Purchasing. Hospital revenue is also tied to performance on those metrics—in 2015, 700 hospitals face fines of 1% of their Medicare and Medicaid revenue due to low scores. Current EHRs are unable to address these questions, so hospitals must employ large batteries of skilled clinicians to abstract these metrics from the EHR. Enter QPID, which can automatically abstract information to address quality scores and questions, thereby reducing labor and improving awareness of clinical problems while they are still correctable.

Ultimately, Wachter says, a program such as QPID might be able to comb through thousands of patient records, in hospital systems across the United States and beyond, to gather evidence that could aid in diagnosis and treatment. But that probably won’t happen for a while. Not only would programs need to be capable of handling such data, but they also would have to share it. And EHR companies and health organizations tend to keep their data proprietary and in-house. “We’ve gone digital,” says Wachter, “but the machines don’t talk to each other.”  For now, that’s a problem beyond QPID’s ken.

See More Actionlink-arrow