Sometimes it’s not about how much you do or have of something … but rather about the quality of the things you do. Or in other words, it’s about answering the right question, not having answers to all the wrong questions.
Effort alone can only take you so far. You can practice a golf swing or shooting a basketball a million times, but if your form is bad, then all you are really doing is reinforcing your errors. Improvement, oddly enough, is not necessarily about pure quantity, but about correcting the little things that add up to the end result we want. It can be difficult to pause, and contemplate what that end result should be, to visualize how the things we are doing might lead up to it … but it is often the failure to do so that leads to our ineffectiveness in the end.
Data science is no different, particularly in healthcare.
In this era of Big Data and Data Science and Analytics, it is easy to get caught up in the buzzwords. They have certainly made in-roads into the healthcare industry over the last decade as well. Seems like everyone and their grandma runs an analytics company these days, or is selling some “AI tool”. The proliferation of accessible machine learning platforms – from software like IBM’s SPSS Modeler and Knime to programming libraries like Apache Spark and Caffe and Scikit – has opened up new technologies to more people and organizations. The accessibility of new types of data has also exploded – genetics data, clinical info extracted via natural language processing (NLP), social determinants from population health tools, vast repositories of claims data, etc.
The real question though is whether we are collecting the right data. In healthcare, one fundamental driving concern lays at the core of all we do – did the patient get better or worse, and at what cost? And even more importantly, what actions did we take to cause those.
Any data or technologies need to drive toward that concern. Ironically, I would posit that a lot of it does not. It is easy enough these days to get some data, and build, for example, a deep learning application to provide a risk score of people’s chance of developing diabetes. But if those predictions are not tied to actions, and that knowledge cannot be used to make better health choices, then it misses the mark.
At its heart, all data science and machine learning and artificial intelligence are an exercise in behavioral re-engineering. About off-loading things we used to have to do in our head, into tools in our environment, so that we perform better. So that we maximize human potential. It is not simply about changing what we know, but rather changing what we do.
By Casey Bennett, PhD
Chief Scientific Officer
Note: Casey Bennett will moderate the panel on Meaningful Data at the 2017 Cohen Veterans Care Summit in DC. Bennett will be joined by Dr. Robert Grossman, University of Chicago, and Tom Herzog, Netsmart.