reprinted from Issue 22, Spring 2016 of Frontiers of Medicine (PDF)
$proxy_page= "/news/fom.html"; ?>
Digital native Alvin Rajkomar, MD, grew up in San Jose and started programming in sixth grade. In 2012, when UCSF Medical Center adopted its electronic health record (EHR) system, APeX, technology was already woven into his life. He ordered a constant stream of Clif Bars from Amazon, and relaxed by watching Netflix on his iPad.
"I assumed that APeX would similarly enhance our clinical life," said Rajkomar, imagining, for example that it would automatically suggest prescribing a laxative if he ordered a constipation-causing medication. "Instead, I realized that the EHR oftentimes made the work harder." He was also disappointed that it took months to get data for a simple research query.
Frustrated, he received training to access the APeX database himself and took software engineering courses online. One tool is leveraging the power of huge data sets to counterbalance the messiness of real-world data – the way a Google search suggests appropriate links even if the query is misspelled. "I’m focusing on a type of machine learning that analyzes large of amounts of really messy data to find ways to improve health care," said Rajkomar, who is now the principal data scientist at the Center for Digital Health Innovation. A few examples include:
- Triaging radiology images:
As a resident, Rajkomar cared for a rapidly deteriorating ICU patient with shortness of breath and low blood pressure. From an X-ray taken moments before, Rajkomar saw that air surrounding the patient’s lungs was compressing his heart. He immediately inserted a needle, decompressing the patient’s thoracic cavity and saving his life. "If I wasn’t sitting there seeing the image appear on the screen, there could have been a potentially life-threatening delay," he said. "I wondered, can’t we do initial recognition of life-threatening abnormalities the same way Facebook recognizes faces?"
He is working on just that, developing a computer program that reviews millions of X-rays to learn which features are associated with urgent conditions. "The goal is not to replace the radiologist, but to flag worrisome images for them," said Rajkomar.
- Using big data to guide care:
A computer algorithm identifies risk factors by analyzing thousands of records over time. "If your credit card account has purchases from China and Europe in the same day, that’s a strange pattern of activity – there may be fraud," said Rajkomar. He is using a similar approach to search for hidden patterns in thousands of patient records. "If you call the clinic with a fever, shouldn’t we be able to predict your risk of coming to the emergency room, and dispatch additional resources if your risk is elevated?" he asked.
"I’m very interested in collaborating with people with technical expertise and resources, particularly those in Silicon Valley," said Rajkomar. "It’s amazing to work at UCSF, because its entire mission is to help people."