The Struggle to Highlight the Importance of Healthcare Analytics

To Jean Huddleston, MD, healthcare needs more people who know both how to build hospital analytics systems and also how to apply them.

An associate professor of medicine and one of the physician faculty members at the Mayo Clinic’s Center for Innovation, Huddleston also has an industrial engineering degree. Deeply invested in the use of analytics to optimize patient treatment and improve hospital outcomes, she sat down with Healthcare Analytics News™ at the Center for Innovation’s annual Transform meeting last week.

She described how analytics can help hospitals recognize patterns and problems. But she also spoke to the challenges that new technologies face when vying for widespread adoption in a healthcare system where they may not be understood.

Can you tell me what the primary challenges you face, and that you’re trying to address, are?

I believe that innovation and healthcare systems engineering are two sides of the same quarter: You can't spend just half of a quarter; you actually have to spend the other side when you put it in a machine. I believe you can't just design something. You actually have to have the analysis to go along with it to make sure it works.

From a redesign perspective, that's one of the places where we're falling down the most. We've got brilliant people with brilliant ideas and amazing innovation, but we're not really good at analyzing how it's impacting the practice and then actually talking about the things that we're doing in healthcare to make it better. Many times people only hear the bad stuff, but there's some really cool stuff going on in medicine at Mayo Clinic and around the world that we don't hear as much about—because I don't think we're good at putting the whole story together and sharing it.

What things that we can control in this sphere need to change?

There's a lot of different pieces and facets to the problems in healthcare right now. The days of having one discipline be able to solve that are gone. The low-hanging fruit is gone. Because of the multi-faceted nature of the problems, we need people from multiple disciplines who actually can work together. And within that, there needs to be a couple of translators who understand multiple different fields and who can actually help people talk to each other across the different languages that we use in our individual disciplines.

If we can bring those teams together and include providers in those teams, then we'll be able to make a difference.

One of the biggest issues I see with healthcare analytics right now is that you've got some brilliant, incredibly smart engineers and data scientists who are doing some groundbreaking work. However, it has not met up with the definition of a real problem in healthcare. Just because we have a huge set of data doesn't mean we need to actually do all the analysis on it. It needs to actually solve a problem, or whatever we do is just going to increase the stack of articles on our desk and isn't actually going to make a difference in a single life.

What would you see as the peak expression of analytics in healthcare, the best use case that actually accomplishes a clear goal?

I'm really biased. Our number one problem at Mayo around taking care of patients in the hospital was the failure of providers to recognize and rescue patients who were acutely deteriorating. We set out to solve that, and we used a full systems-engineering approach, including some data science, and did some gradient-boosted machine modeling. We were able to recognize patients who were getting sick, and put into practice a standardized response.

With that, we're able to get patients their antibiotics and their IV fluids at least an hour sooner.

The two parts of that problem are the recognition and the rescue. It does absolutely no good to just recognize something, to have a score. Whatever use case you're using that just has a score, that just throws up a red flag. It's not necessarily going to help if you don't have a standardized response to that particular thing.

We need to come together as data scientists, engineers, and analysts, together with the physicians, the nurses, the pharmacists and the administrators, to develop a full model: both sides of the coin, for both recognizing a particular problem but then doing something about it to actually impact the lives.

You’re in a very forward-thinking institution, and one that is willing to try a lot of new initiatives. The work that you do, do you ever bring that out to smaller, more traditional hospital systems?

Back in the day, when we first started doing research and publishing things, you could publish that you did, X, Y, or Z, and people could replicate that article in their hospital and actually do it. Small hospitals can't do that now. They can't read the data science paper and then automatically—voila—put it into their healthcare system. The approach that Mayo and many other institutions are taking is to actually commercialize it.

Unfortunately, from a nonprofit perspective, that's one of the only ways to get things out there, because it takes a full team to implement. It takes equipment and technology in these things to actually move healthcare forward now for some of these latest discoveries that we have.

We’re going to be publishing everything so that some of the lessons we learned around the implementation component that people can still do without having all the added technology. I want to make sure that we get that out, too, into as many hospitals as possible, and we'll do that through publication.

Outside of prohibitive expensive, is there a measure of resistance in corners of the health system, or particularly on the provider end, to adopting some of these methods?

There is a fair amount of resistance to adopting some of these methods. The whole black box concept is something that physicians that I work with have a really hard time getting their arms around.

We demand data to prove to us that we need to change our practice, yet we want to completely understand mathematically how that data was derived. Data science has moved way past that. There's very few people who can understand all of the math that went into those models, and as a result it is really hard for people to accept.

We're going to have to figure out how to have those conversations. And that's why I mentioned earlier having a translator: somebody who knows both worlds and who can actually have a conversation describe things in a way that it'll be trusted, and yet still get enough buy-in that we can try it. I sincerely believe that once we try some of these things, if we can get a willing few to try, the demonstration of the impact will be so significant that the uptake will move faster. It's just going to be a slow start.