Data is a good thing, particularly in a clinical setting. Having a wealth of accurate data about a patient can significantly enhance clinical decision making and improve patient outcomes. The problem is too much data can quickly result in information overload. There is only so much data a physician can analyze and interpret with the naked eye. Sooner or later, technology needs to be deployed to support this process.
This is the situation Dr. Partho Sengupta, director of cardiac ultrasound research and associate professor of medicine in Cardiology at Mount Sinai Hospital, recently found himself in. Using ultrasound technology, Dr. Sengupta collects vast amounts of data about the overall heart function in his international research studies. He uses this data to develop new clinical algorithms for diagnosing patient conditions and recommending treatment. However, manually analyzing this ultrasound data takes an incredible amount of time, and even then, he was only able to include a small fraction of the data in his overall evaluation. As a result, Dr. Sengupta has to spend an extraordinary amount of time on a given case for optimal diagnosis or require patients to undergo further testing in order to accurately arrive at a diagnosis. In response, Dr. Sengupta turned to analytics technology in an effort to improve his diagnostic processes and improve patient care. He guides us through his journey in this Q&A.
In a recent pilot study, cardiac specialists have started leveraging data analytics technology to differentiate two commonly misdiagnosed cardiac conditions and improve diagnosis accuracy.
Data is a good thing, particularly in a clinical setting. Having a wealth of accurate data about a patient can significantly enhance clinical decision making and improve patient outcomes. The problem is too much data can quickly result in information overload. There is only so much data a physician can analyze and interpret with the naked eye. Sooner or later, technology needs to be deployed to support this process.
This is the situation Dr. Partho Sengupta, director of cardiac ultrasound research and associate professor of medicine in Cardiology at Mount Sinai Hospital, recently found himself in. Using ultrasound technology, Dr. Sengupta collects vast amounts of data about the overall heart function in his international research studies. He uses this data to develop new clinical algorithms for diagnosing patient conditions and recommending treatment. However, manually analyzing this ultrasound data takes an incredible amount of time, and even then, he was only able to include a small fraction of the data in his overall evaluation. As a result, Dr. Sengupta has to spend an extraordinary amount of time on a given case for optimal diagnosis or require patients to undergo further testing in order to accurately arrive at a diagnosis. In response, Dr. Sengupta turned to analytics technology in an effort to improve his diagnostic processes and improve patient care. He guides us through his journey in this Q&A.
Q: What was the initial problem you were trying to address with analytics technology?
A: In cardiology, there are two heart conditions that are very difficult to distinguish from one another. They present similarly, but each requires a very different treatment approach. One of these conditions is called constrictive pericarditis. This is a disease where the fibrous sac around the heart (called the pericardium) becomes thickened. Patients with this condition end up with heart failure because the heart can’t distend and fill well. The other condition is a muscle disease where it’s not the sac, but the wall of the heart itself that is abnormal. In this scenario, the heart muscle is unable to relax, which also leads to heart failure.
Patients with pericarditis typically require surgery to remove the thickened surface of the pericardium. Patients with a muscle disease, on the other hand, require heart failure management treatment or cardiac transplantation. As you can see, correctly diagnosing these conditions is of the utmost importance because the treatment options are so different. The problem is the symptoms and rhythmic and image data associated with each of these conditions are very similar, making an accurate diagnosis painstakingly difficult. I was hopeful that leveraging analytics technology could aid me in this process.
Q: How did you make these diagnoses prior to leveraging analytics technology?
A: I would start by taking an ultrasound of the patient’s heart. The ultrasound provides more than just a moving image. It also provides a ton of data about the heart including movement, speed of movement, mechanics, rate of strain, cavity shape change, rate of volume change inside the cavity, etc. This data essentially allows you to track heart muscle motion over time. Traditionally, I would take this data and plot it on a graph using X and Y coordinates on a two-dimensional plane. I would look at the wave or curve this data produced and try to identify the average attributes or patterns within specific points in time. This is admittedly a very reductionist approach, because a ton of potentially valuable data is filtered out in the process. The ultrasound generates an incredibly rich data set — for example there may be over 10,000 distinct attributes per heartbeat, per patient that need to be evaluated. However, this is impossible for a physician to do manually. Looking at averages may not be the best approach, but it is necessary for manual assessment. There’s simply no way I can analyze this amount of information myself. Therein lies one of the biggest challenges associated with Big Data. How do you handle all of the data and make sense of it? I was able to accurately differentiate between pericarditis and muscle disease in 76 percent of cases using this approach, but it still wasn’t nearly good enough. Furthermore, I often required patients to submit to additional tests, such as an MRI or invasive catheter-based tests where the pressure inside the heart chambers is directly measured using fine tubes or wires. If we could increase the yield of the ultrasound studies, this could potentially lead to fewer tests necessary to arrive at a final diagnosis. That’s when I started to evaluate different analytics solutions.
Q: What analytics technology did you leverage and why?
A: I started researching how analytics technology was being applied to complex data sets in other industries. Through this research, I became aware of some work Saffron Technology was doing in the field of defense with high-profile organizations such as Boeing. The company’s Natural Intelligence Platform seemed well-suited to handle the complex ultrasound data sets I was generating at Mount Sinai. The platform is not only built to unify data sources, but also learn and identify patterns in real time, continuously adapting when context changes. This capability was essential in identifying changes in heart biometric patterns indicative to pericarditis versus muscle disease. I reached out to them, and they agreed to work with us on a pilot.
Q: How did you approach the pilot?
A: We approached the pilot in a very controlled fashion. I’m an academic at heart. I trust data, but I also want to ensure the results are reproducible. Therefore, we targeted a very specific group of patients to start. We started analyzing the ultrasound data sets for these patients in the form of CSV (comma separated variable) files using the Saffron engine. The Natural Intelligence Platform ingested this information and started identifying key patterns and generating heat maps. The software algorithm quickly began to pick up on clinical attributes that it took me 10 to 15 years of research to identify for disease differentiation. It was pretty amazing.
Q: What preliminary results has this data analytics pilot produced?
A: In this controlled application, we found that the data analytics platform achieved 88 to 90 percent accuracy in differentiating between pericarditis and muscle disease. This is nearly 14 percent more accurate than I was able to do manually. However, the improved diagnosis accuracy isn’t what’s most compelling to me, because it may ultimately vary based on different patient populations. In my opinion, the most impressive aspect of this pilot is the potential time savings it provides.
For example, our cardiologists currently spend 70 to 80 percent of their time trying to understand what a patient is suffering from. Much of this time is spent manually evaluating ultrasound data. Using data analytics technology can dramatically reduce the time it takes for physicians to accurately discover and identify an issue. Faster diagnosis reduces patient risk and allows us to dedicate more time to developing personalized treatment options and ensuring patient compliance. Faster diagnosis also reduces the number of consultations and office visits a patient must endure, which drives down healthcare costs and improves the patient experience. Even if the diagnosis accuracy of the data analytics platform was equal to our manual efforts, the time savings is a huge benefit.
Q: How do you plan to expand your use of data analytics technology, considering the results of the pilot?
A: The first step will be to increase our sample size in the area of constrictive pericarditis and muscle disease (constriction versus restriction). The next step will be to apply the technology to diagnose patients with other cardiac conditions. After that, we hope to apply the technology to all types of chronic noncommunicable diseases — not just heart disease, but cancer, stroke, diabetes, hypertension, etc. Our hope is that the results of the pilot can be replicated in these other scenarios.