I’ve been thinking a lot about AI and machine learning lately, and how they could be applied to healthcare and just about anything else!) Last November, I was fortunate to attend Exponential Medicine (the equivalent of 4 days of TED talks on healthcare), and was exposed to the latest developments in these areas. Some recent client conversations and Kevin Kelley’s latest book The Inevitable (which examines the potential impacts of 12 technological forces on our past, current and future) got me thinking again about how AI and machine learning could be utilized in healthcare.
The problem with data has always been turning it into actionable information. For a long time, the challenge has been manufacturers either didn’t have the analysts that could combine subject matter expertise with analytical skills, or those analysts didn’t have the time to dig through the data to find what was interesting, analyze it, and report on it, unless there was some crisis to diagnose. Service providers typically provide the data and might build beautiful dashboards and custom reports but rarely do they spend the time combing through a manufacturer’s data to really add value. (Lots of our clients are wanting this, by the way.)
What if the machines could do this hunting and pecking for us? As I understand it, machine learning varies along a continuum from supervised learning to unsupervised learning. If you give the computer pictures of cats, and tell it to go find cats in a whole bunch of videos, that’s supervised learning. If you tell it you want to find cats in videos without telling it how, that’s unsupervised learning. The data gurus tell me that’s way harder to do, but we’re getting better at it all the time.
Ever watch a doctor look at a MRI? They spend a couple of minutes flashing through a couple of layers. They’re relying on the amazing power of trained brains to spot patterns that matter. Sorry, they’re not spending hours combing over your scan. They may miss some things. Enlitic’s founder, Jeremy Howard, spoke at Exponential Medicine last year about work they were doing with radiologists, where the computers would comb through every part of a MRI to find what was potentially interesting, and the radiologists would then take those suggestions and assess them. This seems like the perfect augmentation of our capabilities: the machine makes sure nothing is missed, the human makes the final call on what matters.
As the cost of these approaches comes down, think about they could be applied to your functional area to turn data into actionable information. About 15 years ago we were trying to find examples of speculative buying in channel data using smart algorithms. Now computers could be trained to look for those patterns and provide an alert for higher than trend sales and inventory that appeared to be building for particular customers of products with a history of price increases.
Machines could learn to find potential high risk patients for adherence by examining Patient Services Hub and Specialty Pharmacy Provider (SPP) data and alert relevant staff. Potentially they could even suggest or trigger recommended interventions be it a text message or phone call or personalized co-pay. Assess network performance to find bright spots and low spots. Which payers and plans are causing problems? Find the SPPs that are performing better and worse with those plans.
Not new ideas on what to look for, rather a new method for turning the data into something you will actually use. It’s worth spending a few moments thinking about what you do with data now (and what you wish you could do with data) and how a machine could learn to do the hard work of panning through the ever larger river of data, to find the specks of gold for you to work with. Dream big!