EKGs generate vast quantities of data about patients’ hemodynamic status. This IBM client wanted to use EKG data to predict hemodynamic instability before it appeared in their traditional vital signs.
The client is developing a solution that predictively models streams of EKG waveform data, assigns a risk score to each patient, and notifies physicians in real time via their mobile devices.
Predictswhen a patient’s health may decline, hours before instability occurs
Enablesearly intervention to help prevent further decline before it happens
Deliverssavings by reducing the need to admit patients to the intensive care unit
Business challenge story
Seeking predictive insight
The human cost of critical illness is beyond measurement, and the economic burden is also significant. Critical care costs the U.S. economy over USD 260 billion per year, and makes up over 40 percent of total hospital costs. A significant proportion of these costs relate to patients with serious cardiovascular conditions: according to annual statistics, the U.S. sees 450,000 cases of sudden cardiac death, 795,000 cases of stroke, and 1,000,000 cases of sepsis per year, many of which involve admission to an intensive care unit (ICU).
This IBM client’s mission is to conduct multidisciplinary translational research to improve care for victims of critical illness and injury. Finding new ways to help patients with declining health—specifically hemodynamic instability—is one of the key priorities. Hemodynamic instability is when patients suddenly have difficulty maintaining adequate blood pressure. This can be due to the heart losing its ability to pump blood, blood vessels dilating due to inflammation, or internal bleeding.
One of the organization’s physicians had the idea to use real-time data and predictive analytics to help doctors identify patients who are going to become hemodynamically unstable, and take pre-emptive action to prevent the patient’s condition from deteriorating.
As a pilot project, the organization decided to focus on the data generated by electrocardiograms (EKGs), which monitor the heart’s electrical activity. These devices take readings hundreds of times per second, and produce streams of waveform data 24 hours per day.
The project’s goal was to find a way of collecting and storing this EKG data, and using predictive modeling techniques to help identify risk factors and predict when a patient is likely to suffer a rapid decline in health. This new insight would potentially help physicians predict deterioration in a patient’s condition several hours before it happens, and then intervene to improve the patient’s likely outcomes.
For example, if necessary, the physician could move the patient to a higher echelon of care. If they are in a general bed and their prospects are not good, they could be transferred to a specialist ward, where their condition can be monitored and stabilized more easily. By preempting problems before they occur, there is less risk that a patient will ultimately have to be admitted to the intensive care unit (ICU), where the cost of care is extremely high and the outcomes for patients are often poor.
Taking advantage of analytics
To get its pilot project up and running, the organization began developing predictive models based on existing EKG datasets. By mining this data for correlations between unusual EKG activity and subsequent hemodynamic events, the team was able to build a model that detects hemodynamic instability (abnormal or unstable blood pressure), which is a strong predictor of severe problems with the cardiovascular system—such as the patient going into shock.
To make it possible to operationalize these models in a clinical context and deliver a real-time monitoring capability, the organization needed to put the right technology infrastructure in place. The project team chose IBM® Streams®, IBM BigInsights®, and an IBM PureData® for Analytics appliance. IBM Streams passes the data through predictive models to produce risk scores for each patient. BigInsights and PureData are used to land and store the retrospective data, to help train and validate the analytic models.
Finally, the results are passed to AirStrip ONE, a mobile interoperability platform from IBM Business Partner Airstrip, which displays the information in a series of intuitive dashboards that clinicians can view in real time on their smartphones and tablets.
Today, the organization uses a homegrown software solution and patient information from its EPIC electronic health records system to annotate the retrospective waveforms. The annotated waveforms are used to train the machine learning models.
The project team selected IBM because of its ability to provide a comprehensive solution, including a real-time stream processing engine, a highly scalable Hadoop platform for unstructured big data storage, and a high-performance data warehouse appliance for analyzing structured datasets. Having a single vendor for all three components of the solution was seen as a significant advantage, and the team has been impressed with its performance—particularly in terms of accelerating query speeds.
With the solution in place, the organization hopes to transform the treatment and management of acutely ill patients.
Based on the retrospective data that the team has now analyzed, the results are extremely promising. The solution’s new predictive models are able to make accurate predictions about the health of individual patients several hours in advance of a critical event.
Essentially, these models are helping the organization to create a new generation of vital signs that can be monitored and provide actionable data to help avert complications for critical care patients. Traditional vital signs—such as heart rate, blood pressure, respiratory rate and temperature—change too late to be useful in predicting dangerous cardiovascular conditions; but the new metrics derived from the solution will give physicians the advance warning they need to take useful action.
The project team is now working with physicians to train the models on data from their own patients. This should help to make the models even more accurate, and build confidence that the solution will deliver value in a real-world clinical context.
The solution also has the potential to influence some of the key metrics that contribute to the high cost of critical care. For example, it could help to reduce patients’ length of stay (LOS) in the ICU, the overall LOS in the hospital, the number of intra-hospital transfers, the number of readmissions, and the cost per visit. There is also a hope that it will contribute to long-term patient health post-discharge.
The project team plans to roll the solution out to cover ICU and emergency departments and general beds. Eventually, it even hopes to provide a mobile monitoring service to patients at home, through the use of wearable monitors—so they only need to come into hospital if the solution indicates that their condition is likely to deteriorate. This would make it possible for more patients to spend more time at their own homes, and less in a hospital bed—making care less invasive, less expensive, and more efficient.
Ultimately, the organization would like to open up the solution to other healthcare organizations, and possibly even provide it as a service. The hope is to change the face of healthcare not only for the organization’s own patients, but for the greater population across the U.S. and the wider world.
A leading U.S. critical care research center
This IBM client is a U.S.-based medical research organization, and one of the world’s leading specialists in critical care research. Its mission is to improve healthcare standards and outcomes for patients who suffer from critical illnesses and injuries—encompassing everything from brain and spinal cord injuries to trauma, sepsis and cardiac problems such as heart attacks and stroke
- HC: Intelligent Platforms
- Private Cloud Withdrawn
- PureData System for Analytics (powered by Netezza technology)
Take the next step
IBM Analytics offers one of the world's deepest and broadest analytics platform, domain and industry solutions that deliver new value to businesses, governments and individuals. For more information about how IBM Analytics helps to transform industries and professions with data, visit ibm.com/analytics. Follow us on Twitter at @IBMAnalytics, on our blog at ibmbigdatahub.com and join the conversation #IBMAnalytics.
Based in San Antonio, Texas, AirStrip provides a complete, vendor- and data source-agnostic enterprise-wide clinical mobility solution, which enables clinicians to improve the health of individuals and populations. For more information on AirStrip, please visit: www.airstrip.com