Machine Learning for Heart Failure Management

Machine Learning for Heart Failure Management


Machine Learning for Heart Failure Management

Ray Liao of MIT presents Machine Learning for Heart Failure Management at IdeaStream 2022.


Content

6.8 -> Good morning everyone. My name is  Ray, and this is Dr. Steven Hong.  
12.56 -> Dr. Hong is an emergency room physician at Beth  Israel Deaconess Medical Center. Three years ago,  
19.52 -> Dr. Hong reached out to our lab at MIT CSAIL,  and told us how difficult and frustrating it is  
26.16 -> to manage heart failure. 20 percent. 20 percent  of heart failure patients get readmitted to the  
34.16 -> hospital within 30 days of discharge. As a  leading cause of hospitalization in the US,  
42.16 -> heart failure hospitalizations cost the US  healthcare system 12 billion dollars every year.  
49.28 -> Despite the money we spend, patients are dying.  Half of all heart failure patients die within 4  
56.24 -> years of diagnosis. The fundamental challenge  is a variability of this disease. A treatment  
64.64 -> regimen that works for patient John may not work  for patient Jane, and when a patient is out of the  
70.8 -> hospital the therapy has to be constantly adjusted  for the changing and dynamic physiological state.  
79.28 -> Right now, the data is siloed and the intervention  is delayed. What if we can learn from this?  
88.32 -> From millions of patients retrospective  records, we can predict exacerbations and enable  
96.96 -> timely interventions. Empallo empowers caregivers  from the hospital to the home. Specifically, it  
107.2 -> helps caregivers figure out when the intervention  is needed and what interventions are needed.  
114.64 -> Our initial solution integrates data from the  electronic health records and imaging systems,  
121.44 -> identifies patients who are at high risk  of getting readmitted to the hospital,  
127.12 -> and provides actionable insights to providers  to help them adjust and titrate medications.  
136.16 -> In the longer term, we'll integrate data from  wearable and implantable devices. We will provide  
142.64 -> service to medical device manufacturers to help  them identify patients who may benefit the most  
150.64 -> from their product and improve their clinical  trials. Eventually, we're going to enable patients  
158.64 -> to get access to care in a timely fashion. My name  is Ray. I finished PhD in Computer Science at MIT  
166.56 -> last summer. My co-founder Claire has a business  background in healthcare consulting and finance.  
172.56 -> Our team has two machine learning  engineers and a scientific advisory board  
177.68 -> from University of Colorado Hospital, MGH, Mayo  Clinic, MIT, BIDMC, and the VA healthcare system.  
187.68 -> We are seeing strong traction.  Our initial prototype was built on  
192.24 -> 500,000 patient records from Beth Israel  Deaconess Medical Center. We recently signed  
198.4 -> agreements with the VA national healthcare system  and the University of Colorado Hospital system.  
204.8 -> This will give us additional 2 million patient  records to enhance and test our solutions.
214.08 -> We're in an exciting time. In 2008, the adoption  rate of electronic health record systems in the US  
222 -> was only eight percent. In 2010, right before  the Affordable Care Act was signed into law,  
228.96 -> the adoption rate of EHR was 10 percent. In  2015, this rate has grown by nine times.  
238.72 -> The last decade has left us this massive treasure  trove of digital clinical data, and a window  
246.64 -> of opportunity. Our initial customers will  be hospitals because when a heart failure  
253.68 -> patient gets readmitted to the hospital within  30 days that rehospitalization is not reimbursed.  
260.96 -> Furthermore, Medicare/Medicaid financially  penalized hospitals for high readmissions of  
266.88 -> heart failure by deducting the total reimbursement  from CMS by up to three percent. Our long-term  
275.2 -> target customers will be insured because we  save hospitalization costs for their patients  
281.2 -> as will be medical device manufacturers because we  help them identify patients who may benefit from  
288.08 -> their devices, improve their clinical trials, and  enhance their data analytics capabilities. We're  
296.4 -> addressing a big market. Our initial beachhead  market that targets hospitalization cost savings  
302.56 -> will be 3 billion dollars in the United States. As  we grow and expand the home monitoring and other  
309.68 -> adjacent disease applications, we're addressing  a total market size of about 60 billion dollars.  
316.64 -> Very fortunately at MIT, we benefit  from various programs including Sandbox,  
321.76 -> Deshpande, Delta V. We are also fortunate  to get grants from the state government  
327.28 -> as well as NSF. We're at the stage where  additional capital will help us speed up growth.  
334.8 -> We're using 2.4 million dollars to support  our multi-center retrospective study,  
340.8 -> and perspective piloting with our partners, as  well as preparation for FDA pre-submission meeting.  
347.84 -> If you're interested to learn more, please come  talk to us and visit empallo.com. Thank you.

Source: https://www.youtube.com/watch?v=n5Js1Wu9bmM