December 20, 2018 | 3min read
Machine Learning in Healthcare—Interview with Alan Warren
Machine learning and IoT was a big theme at Web Summit 2018. To get some first-hand insights, we met with the CTO of Oscar Health Alan Warren and talked about their product and the role machine learning plays in healthcare and other industries.
Oscar Health is a technology-driven health insurance company that “with more coverage and less hassle” strives to give their members the most accessible healthcare and best experience in the US. Learn more about their awesome product here.
There’s a huge data quality issue in the healthcare industry. Typically 40% of all claims coming into an insurer have some error in either the number of information or the content of information. Some human has to actually go look at it and crack it before it continues processing. We built our models and machinery in such a way that when a claim doesn’t process it runs through our model, which corrects all the information and repeats the process again. We’re also using AI around voice and language recognition, content from phone calls, instant messages, keyword detection, labeling categorization etc.
Alan WarrenCTO,Oscar HealthAbnormal is very nuanced and machine learning models are very good at picking up nuance in large volumes of data.
This is a really interesting question. In the tech world that I come from, you build a product, some app or system, you give it to people and it runs by itself. Then you watch how it works and look at your users’ feedback so you can make a better version of the product. But the machine is pure machine—it runs all by itself.
The insurance world is not like that. We are looking at claims, modifying information and updating state, we have care teams that are looking at member’s information etc… So the insurance world has this human-machine mix, where a machine is running under the hood but with all these people wrapped around it. The product that we actually expose to our end users is a machine learning app that looks like the ones they use every day; that does step-tracking, helps looking up doctors, communicating with them and so forth. That’s the digital part of our product. On the other side it also includes concierge teams, a personalized group of care guides and nurses assigned to each member to help guide their health care journey, plus the whole user-doctor interaction. Our challenge is to make all those elements play together nicely as a coherent product.
Apart from the use of machine learning in healthcare industry, what other areas will be significantly influenced by this technology?
There’s huge work to be done around large volumes of data. When you have a huge amount of data, extracting the needles from a haystack is very hard to do for a human. Machine learning models are very good at looking at large health records and singling out those that look suspicious to find out what’s different about them. Another example is that machine learning systems work well in security, by looking at network traffic and being able to monitor the things that are no longer behaving normally. ‘Abnormal’ is very nuanced and machine learning models are very good at picking up nuance in large volumes of data.
My dream is we’re running the best insurance company in the world and the United States and that people are using our technology to provide health insurance machinery in the world. The dream is that our technology and data-centric approach is used to provide a better and less expensive healthcare experience for everyone, around the globe.
Senior Communication Specialist