Mayo Clinic

Adrian Keister

Could you tell me about your role and responsibility at Mayo Clinic, and your journey as a technology industry leader?

I have a diverse professional background. After a theoretical-focused tenure in mathematical physics, I switched to experimental science and engineering. Early in my career, I taught in private schools for three and a half years before joining Mayo Clinic as a data scientist in access management. 

My current job is to optimize provider schedules and enhance patient access to healthcare providers. To achieve this, I have to ensure the delivery of accurate data and metrics for physicians and patients to help Mayo Clinic make informed decisions.

How would you say data science has evolved, and what are some of the trends and challenges that you see?

One of the trends I see lately is the migration from data science into data engineering. Data engineering is more about getting a data pipeline set up so that you can get data from one place to another and have it ready for the data scientist to do data modeling work. A lot of the big success stories are built on data engineering rather than data science. There are valid reasons for that, and while I believe data modeling is crucial to success, it is only as good as the data you have available. Your entire data science team will advance and be more productive if you have good data engineers making sure your data is high quality and high volume. For these same reasons, there is a huge demand for data engineers.

When it comes to the challenges in the data science field, I see a lot of overhype around AI and ML. For instance, the Gartner Hype Cycle is a tongue-in-cheek graph that shows how a technology or application's hype will evolve over time and can be applied to different technologies. It may be humorous, but it has a lot of reality because whenever a new technology is released, there is a lot of excitement and attention, and everyone jumps on the bandwagon. But as soon as they use it, they experience a slew of disappointments because they understand the technology has drawbacks and can't accomplish everything they want. After putting forth a little more effort, they discover that they can salvage the situation somewhat and perhaps improve upon it. At this point, they begin to realize a real return on their investment.

Are there any new projects you have undertaken which are accelerating the technology adoption at your organization?

One of the main projects I have worked on for the last two and a half years at Mayo Clinic is called Appointment Pathways. Here, we trace the connections between appointments. Let's say you come in for general medicine evaluation, and the doctor there says he thinks you have a neuromuscular disorder, and you need to see a neurologist; so we set up that consultation appointment. Pathways models the causal connection between the general internal medicine appointment and the neurology consult, and then reports on all such connections.

“I think that every statistics course should include the New Causal Revolution. Statistics would be a natural place to put it as it gives us tools both to ask and answer better questions”

We use Google Cloud because we are at the medium data level. We need a lot more compute than just a laptop to do this. We use a graph database, and Python is the glue programming language that ties all these pieces together. The end result is a dashboard that our users can look at to visualize the insights. 

What are some of the transformations or disruptions that you expect to take place in the industry in the coming years?

The New Causal Revolution excites me. I think that every statistics course should include the New Causal Revolution. Statistics would be a natural place to put it as it gives us tools both to ask and answer better questions.

There are new graph technologies like TigerGraph. After switching to TigerGraph, it gave us 25 times the speed up in queries. It's a little bit harder to learn, but it paid off in performance. 

I also mentioned data engineering before. I would love to see a lot more people go into data engineering and start to use some of these tools in Google Cloud and other cloud platforms. We use Google Cloud because there is Dataflow, Data Fusion, and a whole raft of data engineering tools you can use, including extract, transform, and load operations. Data scientists don't tend to be heavily trained in data engineering topics, but I think if they were a little bit more aware of the value of these tools, they could see their projects become much more effective.

What is the advice that you would like to impart to budding professionals in the field who are looking to venture along the same lines as yourself?

In data science, statistics is really at the heart of what I do, so I would advise budding professionals to take statistics courses or to self-study. Here, I'm talking about some advanced studies at a Bachelor of Science in Statistics level, including experimental design. It can give you insights into problems you are not aware of otherwise. As an example, if all your data is clustered, you get two-dimensional data in one place: try to fit a line through it, and you will wonder why the coefficient of determination is so low. It's because it's all clustered. Experimental design would dictate that if you had two data clusters at the endpoints, that is the best way to reduce the variation in the coefficient of determination.  

In this regard, statistics, Python/R, and SQL are essential for anyone getting into data science. If you have all these backbone tools (and anything else is just kind of icing on the cake), you can learn everything (domain knowledge) quickly and integrate it into the data science field. It is also essential to focus on the major things, and then any specific tools that a particular job would want you to learn should be reasonably straightforward.