Healthcare Needs Better Data Science: 6 Questions with Dr. Michael Johnson
In a recent NEJM Catalyst article, the authors discussed the topic of data analytics and the need for a more data-driven mindset in healthcare. According to the article, the healthcare industry contributes to 30% of global data. One industry harboring such an immense portion of the world’s information presents significant opportunities for better data science.
We reached out to Data Scientist Dr. Mike Johnson from St. Charles Healthcare System in Bend, Oregon to get his thoughts on the article.
Dr. Johnson actually came to SCHS from a non-traditional background. He was a very successful Director of Institutional Research at Dickinson College where they utilize predictive modeling for enrollment optimization as well as a variety of other projects on campus. His work there allowed him to dive into Rapid Insight’s analytics and data prep tools that helped support his data-driven mindset. We often hear Mike say, “The proof will be in the data.” Bringing this analytic mindset to SCHS has allowed them to utilize data to support quality patient care and outcomes.
You left a career in analytics and data at Dickinson College where you were the Director of Institutional Research. How does the work you did in one career apply to doing analytics work in your new role?
I was a little bit apprehensive at first, but optimistic. As it turns out, it was a much easier transition than I had ever expected. This is often a topic of conversation I have with people who are considering going from higher education to healthcare – it really does translate well. The toolbox is the same. The only difference is the characteristics of your data field are different and the dynamics in the work place might be different. The way you go about getting decisions made might be a little bit different. The process, however, is so incredibly similar. The impact of the decision is perhaps even more profound in healthcare in some cases.
What was it like adjusting to the new environment?
You don’t have the liberty of taking a lot of time to get acclimated. You need to expect to hit the ground running. They need the results. That really wasn’t an issue. I am in an environment that was very helpful in that regard. I hit the ground running and felt like a kid in a candy shop. One of the things that were commonplace in higher education was the mature use of data analytics. I don’t think the healthcare industry was quite as mature with regard to their use of data analytics. A lot of the time you’ll be sitting in a meeting and say “why don’t we use this approach?” And they’ll go, “I never thought of that. You can do that?” So you find yourself in the driver seat frequently, which I thought was really gratifying.
Who is doing the data science work at your organization? It is a mixture of data scientists and others or is it just pure data scientists?
It’s always a collaborative effort. I may push the buttons and do the work in regards to the actual method, algorithm or process that we’re using to turn the data into information to help make a decision. I absolutely cannot do that without the subject matter experts, looking over my shoulder and being an integral part of that process. There are so many specialty fields across the hospital. I’m not an expert in public policy, hospital finance, or the medical field and patient care. I am the data scientist, but I do not do anything on my own. I also rely heavily on the IT folks to help me get the data, ensure that it’s the dataset that I’m looking for, and they help me understand the characteristics of it.
Dr. Mike Johnson
St. Charles Health System
The article references a build vs. buy dilemma. Can you explain how St. Charles decided to build models in-house instead of just outsourcing this work?
A lot of the times for a one-off project, we just need to get a consultant or we need to get a third party involved for something way outside our area of expertise. I think that each project or task like that needs to be evaluated. There’s pros and cons with that decision. If you go to an outside source, sure you’re buying their years of experience, but you lose that in-house expertise of knowing the intricacies about your data or your process. Frequently, somebody that’s looking from the outside in, doesn’t understand those nuances. The biggest thing I’m finding with doing things in–house is, once you walk down that road, it’s so much easier to walk down that road a second time. If you create a predictive model to help identify patients who are most likely to extend their length of stay, it’s not that far of a stretch for you to create a predictive model for those patients who are most likely to have an unscheduled readmit within the next 30 days. Once you have done it once, you quickly become an expert. That’s probably the biggest benefit of doing it yourself. If you crunch the numbers, it’s also way cheaper to make the investment in-house.
From your perspective, has it been a good decision? What have the returns been? What advantages do you get doing in-house models?
A huge benefit is that across campus, whether it’s a hospital campus or any kind of business, once people realize that this is an option that can be done accurately and quickly, word spreads like wildfire. All of a sudden, you will get calls from somebody, who saw that you answered someone else’s question, asking you for help. They would have never thought to ask if you had had an outside consultant come in to fix your problem. Usually an outside source is a one time deal. Now that you do have that in-house expertise, people want to leverage it, and that’s what’s really cool.
They talk in the article about the challenge of reliance on instinct rather than data results for decision making. Obviously this isn’t the case for St. Charles. Can you give some insight into how your organization made the decision to become data-driven instead of instinct driven?
That was an interesting part of the article because I didn’t completely agree with it. You can’t discount years of experience and the information that comes from the subject matter experts. I think the best answer comes from that collaboration between the data scientists and those with the instinct. One has to trust the other and know the strengths and limitations of both. You can’t just disregard intuition because of what the data is telling you. That’s not going to get you very far. Sometimes you have a gut feeling and the data is telling you something slightly different, so we have a discussion. We talk about anomalies, and sometimes they’re what we’re concerned with. There’s no cookie-cutter answer. I think you need to keep the right level of subject matter experts’ intuition and the data scientists’ analytical approach. The in-house approach, with its flexibility, even makes it easier to explain why intuition might be a little bit off.
Dr. Mike Johnson is a powerhouse when it comes to better data science in healthcare. Checkout more on his story and read the St. Charles Health System case study.