Analytics 50: How big data innovators reap results

Five winners of the 2016 and Drexel University Analytics 50 awards share details of their projects, lessons learned and advice.

1 2 3 4 5 6 Page 2
Page 2 of 6
john martin online Children’s Hospital of Philadelphia

John Martin, senior director of enterprise
analytics at Children’s Hospital of Philadelphia.

Children’s Hospital of Philadelphia: Detecting and preventing venous thromboembolism

Children’s Hospital of Philadelphia (CHOP) has been on a mission to use data and advanced analytics to improve the quality of its care and patient outcomes. To that end, it has launched an initiative to improve detection of venous thromboembolism (VTE) by using text analytics tools to glean insights from unstructured data in physicians’ reports.

“We’ve actually been executing a road map and strategy that we started in 2008,” says John Martin, senior director of enterprise analytics at CHOP. “It started pretty typically with ‘Let’s build a data warehouse based on use cases, with a long-term vision of precision medicine and analytics.’ We started with nothing. We had to build up to it.”

VTE is a condition that involves the formation of blood clots within a deep vein (deep vein thrombosis) that break loose and travel to the lungs (pulmonary embolism).

According to the U.S. Department of Health and Human Services, there are about 350,000 to 600,000 new cases of VTE in the U.S. annually; recurrent cases bring that number up to about 1 million. Nearly two-thirds of the people who experience VTE are hospitalized or were recently hospitalized, and about 300,000 of them die each year.

Children at risk

Martin notes that hospital-acquired VTE is currently the second-most common cause of harm to hospitalized pediatric patients, after central line-associated bloodstream infections. It’s currently the focus of a nationwide prevention campaign. The overall mortality rate associated with pediatric VTE is estimated at 2.2 percent, Martin says. Additionally, pediatric patients diagnosed with hospital-acquired VTE stay in the hospital an average of 8.1 days longer than other children and cost $25,000 more to treat.

As dangerous as VTE is, preventive measures, including early detection, can dramatically reduce the incidence of the malady. And much of the data needed for that sort of prevention can be found in physicians’ notes. The current mechanisms used to identify VTE events depend on manually generated clinical lists and a post-discharge review. Martin says both processes are time-consuming and error-prone and don’t result in immediate detection.

A faster process

To speed up the process, CHOP decided to create a decision support tool for physicians. The hospital applies natural language processing (NLP) to radiologists’ reports, creating a fully automated solution that quickly analyzes complex batches of physician notes and offers a high level of accuracy in identifying and tracking patients with hospital-acquired VTE.

Clinical documentation stored in the electronic health records (EHR) is backed up to a reporting database on a daily basis and then transferred to the CHOP data warehouse.

When the backup is done, identifying information is removed from radiology reports and the reports are transferred to the NLP engine via a secure cloud service. The NLP engine produces results in an XML document that includes both a semantic translation of the notes into discrete data and application of a classification model created by CHOP for deep vein thrombosis. The document then goes to the data warehouse and the patient’s identification is restored.

The data is then converted to Hadoop structured data, where the rules engine assigns the VTE label to each study.

“Technology only does one thing,” Martin says. “It only automates and simplifies things that a human could do — but maybe not as quickly or as accurately. It’s a tool. We were able to apply that tool, that technology, to automate a process that wasn’t previously automated, while increasing its accuracy. Then we can get valuable human time focused on the right cases.”

The payoff

CHOP’s VTE analytics effort has paid dividends, Martin says. The NLP tool identifies patients with VTE with a high degree of sensitivity and specificity, and it has uncovered VTE sufferers who were overlooked by CHOP’s existing VTE screening process.

The NLP engine is now an important component of CHOP’s VTE prevention improvement efforts, according to Martin, who adds that his team is exploring other ways to use NLP applications and hopes to develop methodologies that can be adopted at other healthcare institutions.

1 2 3 4 5 6 Page 2
Page 2 of 6
7 secrets of successful remote IT teams