Default banner
Data Science

'Explainability': The key to machine learning’s future in healthcare

November 1, 2018
A Q&A with Geneia's Chief Analytics and Technology Officer, Fred Rahmanian, about increasing adoption of machine learning in healthcare.

This article originally appeared in MedCity News.

It's not enough to gather data, crunch it and produce data insights. Fred Rahmanian, Chief Analytics and Technology officer of Geneia, explains why his company's committed to keeping doctors and providers in the loop.
Fred Rahmanian, Chief Analytics and Technology Officer

With mountains of patient data on hand and more insights emanating from them through the power of machine learning, how can health care get the most out of bleeding-edge technologies?

For Fred Rahmanian, chief analytics and technology officer at Harrisburg, Penn.-based Geneia, the answer has more to do with people than machines. Rather than just tell physicians that there’s a likelihood of, say, a patient becoming diabetic, the health IT and analytics company has pushed for explainability of its insights.

In an interview with MedCity News, Fred Rahmanian - CATO of Geneia - outlined the current landscape of machine learning as it applies to healthcare, and shares how the company has been able to thread the needle between health plans, providers and employers.

Clinicians have been taught for years to practice evidence based medicine. For this reason, arguably, interpretability of models is probably more important in healthcare than any other domain. Do you agree? Have you seen any promising work in this area?

There's been some important work around interpretability. Organizations are realizing that in order for clinicians to adopt machine learning tools, they need to understand the suggestions. At Geneia, what we do is push for the explainability of insights. We typically don't allow algorithms to go into production unless we can explain the output of our models to clinicians.

Say one can predict the onset of Type 2 diabetes, for example. It's one thing to say that we think there’s a propensity but, typically the next question is “Why?” Most algorithms don't, but we make sure that if we provide a prediction we can also answer those types of questions.

How would you convince health professionals who are somewhat skeptical to the promises of ML?

First, we need to make sure clinicians understand that the role of these technologies is to augment and assist in the decision making. They're there to help. We typically provide insight and try to explain it, but in the end it’s up to the doctor to make the right decision. Also, clinicians need to be educated on how algorithms are developed. Education is key.

What are your thoughts on the overall state of machine learning in healthcare?

Machine learning in healthcare has been lagging behind to some extent. There are reasons for that: we didn't have good data and just recently we've been able to capture good data in electronic form in sufficient quantities that could be used for machine learning.

What about patients? What should they be more aware of?

These days, people hear about artificial intelligence and machine learning and they think Terminator and Blade Runner. They're a little apprehensive about that and so are the providers, but that's due to lack of education. Machine learning in healthcare is not just beneficial to payers and providers. For instance, at Geneia we are working on models that allow providers and payers to identify better ways to engage their patients. This ultimately helps the patients because it encourages them to get engaged in their health in the way that works best for them. At Geneia, we're applying machine learning to care management and population health, which lets us have a more efficient way of stratifying patients, using data to identify what we think will happen to them.

In the last few years we’ve seen great advancements in ML for medical imaging. In your opinion why haven’t we seen the same in other areas of healthcare? Or have we?

Partially this is because clinicians have been more receptive to accepting recommendations from diagnostic imaging tools. The visual nature of recommendations from these tools makes it easier to understand and accept. Secondly, the scarcity of good healthcare data was a contributing factor to the delay of adoption of ML in healthcare.

With recent regulatory requirements that encouraged providers to use EHRs and the popularity of personal devices like Apple watches, Fitbits and even cell phones, we now have access to much larger corpus of data. This has been the driving factor for recent increased interest in use of ML in other areas of healthcare like population health and care management.

Let's talk about affordability of ML platforms in healthcare. Do you foresee costs being reduced over time, thus leading to broader adoption?

The adoption of ML in more areas of healthcare can only lead to better outcomes which will ultimately result in lower cost of care and also higher patient satisfaction. Assisting clinicians via ML will inevitably allow them to provide better care and improve the joy of medicine for the clinicians. This is essentially our value proposition at Geneia.