AI can potentially expedite diagnosis, treatment of patients with FH
Click Here to Manage Email Alerts
ATLANTA — Research into the use of artificial intelligence for precision screening of several CV conditions has shown the technology’s potential to benefit underdiagnosed patient populations.
The FIND FH machine learning model, presented at the FH Foundation’s FH Global Summit in October and simultaneously published in The Lancet Digital Health, was able to analyze large health care encounter databases to identify patients with familial hypercholesterolemia (FH), as Healio previously reported on-site at the meeting.
Healio spoke with Katherine Wilemon, founder and CEO of the FH Foundation, and Kelly D. Myers, chief technology officer for the FH Foundation, to learn more about the machine learning model, the study that assessed its capabilities and how it may affect the diagnosis and treatment of patients with FH.
Question: What do the findings add to the knowledge base?
Myers: What this study shows is we’ve been able to develop a cutting-edge tool that leverages machine learning and big data to address the No. 1 problem with familial hypercholesterolemia: lack of diagnosis. The model and implementation process make precision screening a reality by reducing the number of individuals who need to be screened dramatically. If you are hypothetically hoping to identify eight individuals with FH, you would need to screen about 2,000 individuals in a population-based screening approach. With this precision screening approach, you would need to screen 10 individuals to find those same eight individuals who likely have FH.
Wilemon: Just to put it into context, what we see is that FH is theoretically highly recognizable, but it is not being diagnosed consistently within health systems. Ninety percent of this population is underdiagnosed. What this study shows is the value of machine learning in making a diagnosis of a condition more feasible and then hopefully improving outcomes as a result of it.
Q: How easily can a model like this be implemented?
Wilemon: The way that the FH Foundation has designed the FIND FH initiative is that we work in partnership with a health system to make screening more efficient. We get the de-identified data from the system that we then run the algorithm on, returning results to the health system and the clinician of an individual that warrants further screening. We’re working in partnership with these health systems to pull these individuals into care in a very efficient manner. We think instead of adding additional burden to health systems and clinicians, it actually takes a burden off of them.
Myers: In working with the health systems, they share data, we apply the model to that data and then we work collaboratively on a process to get those folks screened and into care. One of the important aspects of this process is that a large number of individuals who are yet to be diagnosed for FH are with primary care physicians, so this process gives the health system a way for specialists, lipidologists, endocrinologists and cardiologists to collaborate with primary care physicians to expedite the process of who should be screened and who should be brought into care.
Q: What further research is needed in this area?
Wilemon: We’re now on the fourth version of the FIND FH model, so we have been continually improving and refining the model as it’s exposed to more data. Now the area of research that needs to be moved forward is how to effectively implement this model. There’s a lot of interest currently in implementation science and research, and what it can teach us about how we effectively change the course of care for a certain population, how we change human behavior, physician behavior, nurse practitioner behavior, administrators and patient behavior. That’s what the FH Foundation will focus on: how do we effectively apply technologies like this to make care more efficient and more effective?
Myers: I wholeheartedly agree that optimizing the implementation process is our highest priority right now. We’re using tools of implementation science to do that. From a modeling perspective, one of the things we’re turning our attention to is identifying a broader population that is at risk for a CV event. We’re looking to build and refine a model that would look at anyone at risk of a future CV event, not just those individuals with FH.
Wilemon: Physicians only have so much time with a patient, and especially in the United States with a health care system is so fragmented. One of the interesting things that the application of machine learning can do in an algorithm model like FIND FH is to look across an individual’s entire medical history and assess at what age did this person first have, for example, in this case, high cholesterol marked in their record? It may not have even been noted. We know lots of children get lipid panels done, but pediatricians don’t consider that an area where they need to focus. We’re using a huge amount of data to see how does their medical history indicate that this individual is at high risk or may have a probable condition? What that means for other conditions is very exciting because physicians will never have enough time to review this much data on their own and see those patterns both in an individual or across a population.
Where do we go from here with regard to this model?
Wilemon: This is a great tool, but a tool is only as powerful is how it’s wielded. How do we apply this in an effective manner where all of the partners — physicians being the key partner — welcome it as a means of practicing medicine even more efficiently and effectively.
Myers: A lot of times when studies like this are done, the publication is kind of the capstone event. For us, the publication is the end of the beginning because the academic exercise of doing this was rewarding, but the reason we did it was to have credibility so that all stakeholders in the health system could say they could act on the results of this type of model. In addition, machine learning and more general terms like AI are very famous right now, and there are a lot of applications out there that use machine learning to figure out what the optimal number of times to show you an ad is before you’ll click on it is. This model is about saving lives. There’s over a million individuals that have been identified by the FIND FH model that need early diagnosis or as early diagnosis as possible. This is a tool and a process to help optimize that and bring these folks into care.
Wilemon: That’s the reason that the FH Foundation embarked on this initiative almost 6 years ago. We currently have the therapeutic means to help individuals with FH. We currently can diagnose them with a variety of diagnostic criteria and increasingly genetic testing, and yet it is not happening. We know that 90% of people with familial hypercholesterolemia aren’t diagnosed. There was just a great article published in The New England Journal of Medicine (Luirink IK, et al. N Engl J Med. 2019;doi:10.1056/NEJMoa1816454.) that demonstrated that we could target those individuals who are at the highest risk, treat them early before cardiovascular disease sets in and protect them and prevent them from what would have otherwise been their destiny. If you can’t tell, we’re very excited about it and pleased to have this publication behind us so that we can go forth in terms of applying it. – by Darlene Dobkowski
Reference:
Myers KD, et al. Lancet Digit Health. 2019;doi:10.1016/S2589-7500(19)30150-5.
For more information:
Katherine Wilemon, can be reached at kw@thefhfoundation.org; Twitter: @KAWilemon.
Kelly D. Myers, can be reached at km@thefhfoundation.org; Twitter: @KellyDMyers.
Disclosures: The study was funded by the FH Foundation and supported by Amgen, Regeneron and Sanofi. Wilemon is an employee of the FH Foundation. Myers is a paid consultant for the FH Foundation.