External data can help make internal predictions

BOSTON--Analysis of outside information helped one hospital determine what was causing changes in its patient census.

Nephi Walton, MD, biomedical informaticist and genetics fellow at the Washington University School of Medicine, shared his organization's experience during a presentation at the Big Data & Healthcare Analytics Forum.

The hospital looked at large increases in patient census and asked a lot of questions about the cause. A lot of children had a respiratory virus which often is affected by weather and season. Viruses are more stable in cold weather, Walton said, and also depend on host susceptibility, human behavior and even air pollution.

Beyond such triggers, there are signals observable once an outbreak happens. “Lots of other things start happening. School attendance is down and there are more Google searches for treatment. People are showing up at clinics and people buy more of certain medications.” Even TV viewing patterns change, he said.

Walton said researchers found that using calendar variables and behavioral data to forecast the patient census at tertiary care children’s hospital was quite accurate. “Outside viral activity is better for predicting long-term trends and more important for predicting census than viral activity at the actual hospital,” he said.

Another case Walton discussed centered around improving the diagnostic yield and decreasing the cost of genetic testing. He said whole exome testing doesn’t really cover the whole genome, and different genetic testing companies give different results. Actual interpretation costs more than $19,000.

Walton discussed a pediatric patient who presented with seizures that did not respond to medication. Several rounds of expensive testing determined that the boy had epilepsy. This could have been determined two years earlier for less than $6,000 rather than the more than $17,000 in unnecessary testing conducted on him.

And that’s only for one patient, he noted. Some “exome” tests actually missed 100 percent of the sequence of an important gene in epilepsy. This case study shows that genetic testing changes too fast to wait for published studies.

To bring the benefits of big data and predictive analytics to the front line of medicine, faculty, residents and fellows are training to use analytics and organizations are nurturing specific projects to fit the needs of their departments of subspecialties.

Walton shared his experience with deploying a de-identified database for querying and discovery and providing a data concierge service to deliver processed clean data. They created a mechanism for crediting in publications to eliminate siloes. They also deployed an easy-to-use analysis platform which required no programming. They offered training sessions to everyone interested and provided a research assistant dedicated to analytics projects.

Part of the process is Informatics for Integrating Biology and the Bedside (i2B2) a National Institutes of Health initiative compromised of six core with multiple cells to create hives. It is free, open source software utilized to support high-level discovery queries.

“We’re looking at data we typically throw away and making use of that,” he said.

So far, they’re still breaking down the data siloes but classes have begun with a positive response and there is a large interest in predictive analytics, he said. Ideas include tracking data from ICU bed monitoring devices, predictive analytics in cardiology and more from nephrology, neurology and genetics. “If you introduce the concepts and explain how it works, people have great ideas.”