Like so many healthcare organizations, Beth Israel Deaconess Medical Center has massive amounts of data from its clinical care, education and research efforts, and the volume keeps growing. Putting all that data to good use, especially to improve care and patient outcomes, is a top priority for the Boston-based, 650-bed hospital, which treats hundreds of thousands of inpatients and outpatients each year.

This fall, the medical center, which is a teaching hospital of Harvard Medical School, will begin pushing live feeds of data into a custom application that caregivers can use to analyze risk levels in the intensive care unit (ICU) at any given time. Ultimately, the application will help Beth Israel Deaconess predict which patients are at risk of developing dangerous complications like infections, blood clots, or bleeds.

“We’ve been using big data for a while now. We’ve used clinical data to help predict, for example, hospital length of stay, and we have used demographics and other data to make strategic decisions such as hospital expansions,” says Kenneth Sands, chief quality officer and a senior VP at Beth Israel Deaconess. “But much of the work around quality and patient safety has not been part of the big data exercise.”

The project, known as Risky States, includes IT and care teams at Beth Israel Deaconess, scientists from the Massachusetts Institute of Technology (MIT) and human-factors experts at Aptima, which provides data collection, measurement, analytical, modeling and decision support systems. Work on Risky States, and its specific intensity index application, began more than two years ago, and required months of data extraction, translation and loading to prepare the data and to determine the appropriate data elements to include. It also involved the creation of a model to synthesize and correlate the data and the development of a user interface to render the data into actionable information that nurses and doctors could use to improve patient care.

Healthcare organizations are increasingly looking to leverage the large volume of data they are amassing. In its “Healthcare IT Vision 2015” report, Accenture identifies the “intelligent enterprise”—which makes use of big data—as one of five key trends in the healthcare industry. The report notes that 41 percent of health executives say the volume of data their organization manages has grown by more than 50 percent in the last year.

Risky States is just one component of Beth Israel Deaconess’ overall initiative to eliminate preventable harm in critical care, an effort that is funded by a $5.3 million grant from the Gordon and Betty Moore Foundation. It includes Risky States and the intensity index application as well as two other applications: MyICU, a patient-centered portal for ICU caregivers and families of patients in the ICU, and Content-Sensitive Checklist, an application that guides caregivers through appropriate and routine patient care and that takes into account context and an individual patient’s current state in the ICU. Aptima has been working with Beth Israel Deaconess on all three, according to Kevin Sullivan, VP of operations at Aptima.

The Data

Risky States relies on mostly structured data from ICU clinical information systems that continuously gather vital-sign data, such as blood pressure, electrocardiogram (ECG) and pulse oximetry physiologic metrics, says Pat Folcarelli, RN, PhD, who serves as senior director of patient safety at Beth Israel Deaconess. In addition, data comes from a custom-built hospital information system that holds patients’ electronic medical records, labs and patient flow that provides data about the movement of patients to different departments such as radiology. Also, there is data from HR systems that are used for staffing, scheduling and payroll to provide information about which nurse was working when, how much training and experience the nurses have and whether a nurse on the floor regularly works in a specific ICU or is working there as a substitute.

Accumulating the data hasn’t been a problem. For many years, Beth Israel Deaconess has been using electronic applications including EHRs, networked medical devices and other systems that create a lot of data, and the hospital has about 3 petabytes of stored data, which continues to grow at a rate of 25 percent a year. The bigger challenge for Risky States, and other big data projects the hospital embarks on, has been normalizing the data and prepping it for analytics. The way clinical care is documented can vary greatly; for example hypertension, high blood pressure and elevated blood pressure are three different terms that describe the same condition.

“There has been a lot of data cleanup that needed to be done, and in the process, we’ve learned a lot about structured data, and quality of data,” says Folcarelli. She says it took at least a year to normalize the data and determine the data points that would work well in the model. Statisticians and analysts worked with clinicians and nurses during this process.

The hospital’s IT team uses scripts that extract data from the transactional systems—the HIS, the clinical ICU systems, the HR systems—on a regular basis. The extracts are sent to the hospital’s clinical data warehouse, which is built with Microsoft SQL Server technologies. The extraction, transformation and loading (ETL) process is managed using SQL Server Integration Services (SSIS), a platform for building enterprise-level data integration and data transformations solutions. SSIS can be used to update data warehouses, clean and mine data, manage SQL Server objects and data and extract and transform data from a variety of sources -- such as XML data files, flat files, and relational data sources -- and then load the data into one or more destinations, according to Microsoft.

Beth Israel Deaconess has been using Microsoft SQL Server technologies for several years, and not too long ago it implemented data warehouses based on Microsoft SQL Server 2014 which features built-in data compression technologies to improve application performance, a decision that has cut query times and improved access to big data with a hybrid cloud solution and better business intelligence (BI) tools, such as Microsoft Azure HDInsight and Microsoft Power BI for Office 365. HDInsight is an Apache Hadoop implementation that runs in the cloud and Power BI provides a set of online analytics and reporting tools.

The Model

Beth Israel Deaconess worked with data scientists at MIT to create decision trees and develop the statistical data models for Risky States using retrospective data it had compiled on all ICU patients from 2012 to 2014. “The two years of data covered 1,800 patients and every 12 hour shift worked during that time. A 12-hour shift is defined as one unit, so for a patient who stays seven days there, that’s 14 units, and there are data points associated with that. So there are hundreds and hundreds of data points per patient,” explains Sands.

Risky States also has benefitted from the expertise of Harvard researchers who developed custom software tools and wrote open-source code to mine large clinical data sets. “Their work has helped our efforts, even if only indirectly,” Sands says. “Some of the skill sets that allowed development of these software tools allowed us to put together the data set that we’ve used for the Risky States analysis.”

The models will have to continually be updated, and going forward, that will be handled by Aptima. “All we’ve done so far is to show these are things that influence risk at a point in time,” Sands adds.

Creating the models is a critical function of the project that helps the hospital better understand which particular sets of events—determined from analyzing a variety of data sets—are more likely to occur with particular harms, according to Aptima’s Sullivan. “Perhaps there’s an uncommon condition with certain patients that drives risk, or maybe having more junior-level staff than senior-level staff during a shift. Any combinations could cause risk to go up or down,” he says.

With the models in place, Aptima designed a big data analytics application for Beth Israel Deaconess. The principal analytic technique used is a statistical approach known as recursive partitioning. “Basically, the way it works is by finding the factors that are most effective in splitting the data set up in ways that identify risk of harm,” says Sands. The IT team uses Microsoft’s SSIS to manage the process of sending data from the data warehouse to the analytics application built by Aptima.

The Caregiver Application

The analytics application Aptima has built features a user interface to present the risk levels and the factors that contribute to them so clinicians and nursing staff can do something about the risk at that time. Clinicians and the nursing staff in all seven of Beth Israel Deaconess’ ICUs – medical, coronary, surgical, etc. – will have access to the application, which is web-based, to run risk reports on PCs and even handheld devices. The application is hosted at Beth Israel Deaconess, is built on a Java back-end (as are the MyICU and the Content-Sensitive Checklist applications), and works with the hospital’s Microsoft SQL Server data warehouses.

Staff can query the application, and the application will calculate a risk score for the ICU or even a particular patient by comparing data from the patients’ records and HR data with the data in the models. The scores are displayed on a visual dashboard to the ICU staff.

Until now, the application has been using only dummy data against the models, but Beth Israel Deaconess is about to go live with new data feeds pulled from the data warehouse into the analytical application (again, using SSIS) in real time, every 15 minutes. The live feeds will enable staff to make more proactive decisions, such as delaying a patient’s surgery—decisions that could reduce risks.

“Over time, the hope is to reduce the prevalence of risks by making changes in care. Or, for example, by providing more training to junior staff, better managing staff workloads, or blending junior-level staff with more senior-level staff during certain times,” Sullivan says.

Beth Israel Deaconess also expects the application will help it tap into care techniques that no one’s thought about before by performing what-if analysis. “We only know what we know,” says Folcarelli. “We’d like to build in capability that would let clinicians signal the app that perhaps the risk level and intensity is different than what the system is saying. Maybe risk is higher. Is the system missing something that we need to add in?” she adds.

The key is for big data, and the applications that leverage all that data, to augment the knowledge and expertise clinicians and nurses already have in patient care, not supplant it.

“This is the beginning,” says Sands, “and it is all about patient care.”

Register or login for access to this item and much more

All Health Data Management content is archived after seven days.

Community members receive:
  • All recent and archived articles
  • Conference offers and updates
  • A full menu of enewsletter options
  • Web seminars, white papers, ebooks

Don't have an account? Register for Free Unlimited Access