Analytics could drive the future of VistA

The Veterans Administration collects a lot of patient data. The agency data mart has about 2.5 petabytes of structured information on patient interactions -- about 900 billion rows of data covering 1.6 billion encounters and 1.5 billion pharmacy orders.

The big question for VA's experts in informatics and analytics is how to make that information available to improve the quality of care, without drowning providers in data.

"We've created a created a system where the complexity of the task at hand almost always exceeds the cognitive capabilities of providers," said Dr. Stephan Fihn, director of the Office of Analytics and Business Intelligence at the Veterans Health Administration.

To manage the firehose of data, Fihn is leading an effort to develop analytics that interpret signals from the vast stores of VA data to alert physicians when patients are at risk. The early work, which has been in place for about three-and-a-half years, assesses risk factors for death and hospitalization. Research is underway to develop analytic models to measure individual risk of kidney failure, acquiring serious infections and suicide.

The mortality and hospitalization analytics are run on 6 million patients each week, using data that is updated from VistA about every four hours. Real time data is the ultimate goal -- as is the integration of data from mobile medical devices and apps, although the lack of industry-wide data standards makes this problematic at present.

According to Fihn, about one-quarter to one-third of providers are looking at the analytic data regularly -- it's available via a web app that links through the VistA health record, and can be accessed on its own. The next step is the integration of analytic information in VistA, which has already begun in the VistA Evolution project that is modernizing the VA's home grown health record.

The models are complex, Fihn says.

The hospitalization and mortality model draws on more than 120 variables from domains across the VistA health record, from diagnoses to rates of health care utilization to demographic information. There is a push to incorporate even more data from outside sources, including the Defense Department, Health and Human Services and elsewhere.

The models are quite accurate, in terms of predicting risk, but the overall goals for analytics at VA are more ambitious.

"It's really about the ability to capture not only what we typically allude to when we talk about datasets, but the narrative of the patient, their story," said Dr. Theresa Cullen, chief medical information officer at VHA.

In an ideal system, a provider would be able to access structured and unstructured data about patients, including  diagnostic tests and personal history, and be able to query and view the information in such a way that it doesn't disrupt the personal connection between provider and patient -- a critical factor in a medical system when interactions might last only 15 or 20 minutes.

"We're nowhere near there yet," Cullen cautioned, but said that she had recently come around to thinking that such a system wasn't just "pie in the sky."

In the near term, Fihn said, it’s about getting the right information in front of the provider at the right time.

"Increasingly, what physicians are doing is acting as information managers, collating information from tests, procedures, histories, and consultants," he said.

Right now, the health record does not do a good job of managing the display of relevant information.. The goal is to redesign the interface to highlight more relevant data.

"So that when a patient comes in and says, 'my knee hurts,' I see a very different view than when the patient comes in and says, 'my head hurts,'" Fihn said.