by Ben Scott, VP Client Services
I just asked the magic eight ball on my desk if you’re thinking about investing in a predictive model, and it said “Without a doubt”. How does it know? Well yes, magic of course, but probably not a lot of it is needed in this case because just about everyone is looking to get into predictive models right now. Not that IBM has ever been shy about consumer marketing, but we’ve even got Watson commercials running during sporting events now. One of my kids threw a rock into the river the other day and accidentally hit a readmission model. Demand for “analytics” and especially the “predictive” flavor of them is big right now, and we don’t need a scientific study to confirm it for us. Just type “length of stay predictive model” into your search engine of choice.
The value proposition and promise of predictive analytics is clear and easy to understand (probably why Watson is talking to us about predictive maintenance on turbine engines during a break from the Cavs-Warriors game). If we can get machines to do something we know we’d do with our brains if possible (process innumerable variables and weightings to form a very educated guess on what will happen in the future), we can to a degree control the future. Oh, now that’s something we want. But I’d like to point out something about the Watson commercials (and this is where we’ll get to puppies, if you hang with me). Where is Watson directing the output of its predictions? To people. Why? Because while Watson can make a very good guess about the need to change out some widget or other in a turbine engine, it can’t change that widget itself (and until robots are administering healthcare services, this will remain true in the healthcare industry at least). Humans need to do that part. What if Watson made a prediction in those commercials and there weren’t any humans around to hear it? One might fairly ask if the outcome would have been the same as if Watson had made no prediction at all. If Watson predicts that a tree will fall in the woods without anyone around, and then the tree falls, do either make a sound?
For those of you who are still here because you searched for predictive models in healthcare and their relation to puppies, your persistence is about to pay off. Predictive models are less like goldfish, and much more like puppies. The good news is they can be a wonderful part of your organizational family, the maybe less expected news is it’s a lot of work to introduce and integrate them into the family, and if they aren’t integrated into your family, you probably won’t realize their full potential.
For my youngest’s six birthday, he got three guppies. Gray, Banana and Kyle. Banana is my favorite. Actually, I have no idea. Why? Because the ceiling of the guppy is very low, and the amount of investment we make in them is correspondingly low. The whole process of acquiring the guppies and setting up their habitat took about half a day, and cost around $27. From that day to this, we’ve gotten exactly the same level of joy and satisfaction from having them in our family.
Contrast this with the puppy I was lobbied into accepting into the family about a year ago. Lewis. We spent months researching breeds and breeders. The kids had to save their money to chip in, because I don’t even want to type how much he cost up front, not to mention the “puppy care” plan our vet talked us into. We spent hours and hours training him to do the most basic of all family activities, depositing his waste in the correct place, never mind advanced topics like “come” and “stay” and “leave it along for the love of all that’s good!”. In short, Lewis is exponentially more work than Gray, Banana and Kyle put together. But, then again Lewis is a complex creature. My kids don’t go find Banana when they’re feeling sad. Gray doesn’t even try to relate to me. None of the guppies can go for a hike with us on a beautiful day. Lewis is a member of our family, while Gray, Kyle and yes even Banana are really just a part of the scenery.
Predictive models are absolutely powerful, and they do indeed hold a world of promise. They are also complex, and therefore require a requisite amount of planning and work to become valuable members of your organizational family. They need to be built correctly upfront, and they need to be trained (and regularly re-trained) to work in your environment. They need to be integrated into your workflows so that their output can be made useful to the people who will act on it. We get all of this, and we’re very upfront about it. We look at your predictive model implementation as a partnership that includes the data and the data science, but also includes the workflow integration, because what we get really excited about is seeing your model grow to meet its full potential. Let’s work together to be sure that predictive model you’re “without a doubt” about to invest in becomes a valuable member of the family, and doesn’t become just a part of your organizational scenery.
If you are interested in learning about our process for deploying predictive models in healthcare organizations, send us a quick note to email@example.com and we’ll get back to you.
Ben Scott is the VP of Client Services at Proskriptive. He has spent the previous fifteen years working in business intelligence, business analysis and business management. In observing the healthcare industry as a consumer, he concluded there may be no place where there is a greater need for efficient, practical, high-value information than at the point where practitioners meet patients. Ben manages customer engangements to ensure that the Proskriptive vision to use data science and advanced analytics to improve performance measures which enable healthier communities is met.