“When it comes to health data, Watson hasn’t been much help.”
-STATNEWS, Ross and Swetlitz. Bos Globe 6/18/18
This week all the newspapers (at least in Massachusetts) have been abuzz with the announcement that Atul Gawande, MD has been picked by three moneyed titans of innovation to head their new company to revolutionize health care. Optimism, promise, and hope is in the air! Kind of like when IBM presented Watson, its supercomputer, in 2015 as the tool to provide workable insights into the financial and clinical dilemmas of U.S. hospitals in 2015 via Watson Health.
How is that working out? Watson Health has access to data on tens of millions patients, in part by spending $9 billion to acquire other companies. It’s initial focus was on developing workable products in oncology, designed to help physicians individualize cancer treatments. “With these acquisitions, IBM will be one of the world’s leading health data, analytics, and insights companies, and the only one that can deliver the unique cognitive capabilities of the Watson platform”, said the general manager of Watson Health in 2015.
They (the newly merged companies) struggled with the basic step of learning about the different forms of cancer and the rapidly changing landscape of treatments. Last week Watson Health laid off people partly because, according to some, even Watson had difficulty in digesting all that data. “…They also don’t understand the generation of information, and how it is used, and whether they can do something different with it,” said Robert Burns, professor of health management at U Penn Wharton School. You can almost hear every primary care physician that is struggling to get their new EMR system to give him/her more information and less data cheering loudly in the background, “We couldn’t have said it better!”
The goal of a great deal of innovative technology in health care is “ “zero patient harm”. if Atul can’t do it all with his surgical checklists and Watson can’t do it all with data from tens of millions of patients , what/who can? How about Artificial Intelligence (AI), aka “machine learning”? AI and machine learning is the converting of data into information without the need for human programmers. For instance, if the computer views enough pictures of different dogs, it will learn to correctly identify a cocker spaniel. I think a real test of AI would be to see if it can recognize a Labradoodle, or any other of the many poodle cross breeds. (Don’t you sometimes worry about the moral standards of poodles that seem to be eager to mate with any kind of passing breed?)
The building of knowledge from patterns in data, both visual and language, is labeled “computer vision”. In some medical studies “computer vision” is used to monitor actual bedside events and identify omissions or non-compliance in procedures. It has apparently improved rapidly beyond just identifying dogs or skin rashes because of “deep learning”: a type of machine learning that uses “multilayered neural networks whose hierarchical computational design is partly inspired by biologic neutron’s structure.” (1) Got that? Think Google’s self-driving cars. “Computer vision may soon bring us closer to resolving a seemingly intractable mismatch between the growing complexity of intended clinician behavior and human vulnerability to error.” (2)
So, the effort to cut the Gordian knot of patient safety and cost-effective medicine continues. I suspect that the three titans of innovation have turned to Atul Gawande, a health care innovator who successfully uses clinical insight and re-education to effect change, because they recognize the limitations that are becoming more apparent in big data.
- NEJM April 5, 2018 378:14; 1271-2