We are delighted to bring you this interview with Sanjeeva F., the VP of Research at OptumLabs, part of UnitedHealth Group. He leads the Center for Applied Data Science (CADS), an OptumLabs innovation team focused on creating software product concepts for Optum and UnitedHealthcare that leverage the latest breakthroughs in Artificial Intelligence (AI). His team takes advantage of the amazing breakthroughs in Deep Learning, Graph Analytics and Neural Machine Translation that are transforming AI in health care today. In fact, they recently trained a new deep learning model to support more efficient and compliant medical chart reviews for the Optum360 business.
UHG: Did you start in technology, or did you immediately go into Data Science?
SF: I studied Computer Science as an undergrad, and my major focused on programming, computer architectures and databases. As the data science field emerged, it took shape by drawing on statisticians and computer scientists to study large sets of data – computationally larger than anything we had seen before.
UHG: Was the transition from IT to Data Science seamless? What, if any, was the biggest learning curve?
SF: Catching up on Statistics, Algorithms, Algebra and Calculus have been the biggest challenges for me. I had not really thought about Calculus since my AP test in high school and had forgotten most of the Linear Algebra I learned in college. My son is a senior in high school and it has been fun catching up on Calculus with him.
UHG: What impact do you think Artificial Intelligence and automation will have on Data Science in the next 10-15 years?
SF: Artificial Intelligence is rapidly transforming data science. When data science was first coined as a term, the focus was on extracting summaries and basic insights from large data sets. Using these insights, we could make better decisions on how software programs should run by better understanding which features people liked, what ads to show them and more. Today, the datasets are so big, and the data science techniques are so much more sophisticated that we can train software programs to learn directly from the data. Rather than explicitly tell the software programs what to do, the models infer how to proceed, by ‘learning’ from the data. This offers an exciting new way to approach software development. It brings both opportunities and challenges.
UHG: Many people have set their career sights on Data Science, but it may not be right for everyone. What skills and/or technical aptitude do you believe would make someone more successful in the this world?
SF: I think having a real interest in testing and experimentation is key – great data scientists allow the data to guide their discovery. It can be a trap to apply your own point of view and then try to transform the data to support your opinion.
UHG: What in your opinion separates our technology and analytics work from other global organizations?
SF: We have a great mission – to help people live healthier lives and to help make the health system work better for everyone. It’s exciting to be mission driven and recognize that the breakthroughs in AI we are pursuing can make an impact in driving down cost, expanding access and improving quality in our health care system.
UHG: What kind of programming languages do we use at Optum?
SF: Many of the data scientists I work with are programming in Python, especially those training machine learning models. We also have a strong community around R, and a number of analytics professionals who use SAS and Stata.
UHG: What happens at our company when an idea fails?
SF: When an idea doesn’t work out, we put it “on the shelf” and move to the next idea. Our software product concept process is rooted in experimentation. Like a bio or chem lab, we state our hypothesis or expected result and test that through a series of experiments. We focus on the riskiest portion of the idea – what is most likely to make it fail − and test that first so we focus our efforts on the most promising ideas.
UHG: What’s the most important thing someone should know before joining our company?
SF: For data scientists, it’s important to remember we are stewards of often sensitive and personal health care data. In other industries, there may not be the regulatory requirements to tightly manage personal health information and personally identifiable information. While our data is de-identified, we are stewards of this information, and should always be asking ourselves what is the safest way to work with the data to preserve the confidentiality of the information we have been entrusted with.