Healthcare IT is hard. I bring this up as the drumbeat of negative press about IBM Watson has been getting louder recently, especially around healthcare. And now comes this comprehensive report from STAT News – a well-respected source in the health sector – on IBM’s efforts with Watson for Oncology. The report outlines the struggles clinicians have had getting advice from Watson beyond what they – or a set of doctors at Memorial Sloan Kettering who are training the software – already know. This comes a few months after MD Anderson cancer center in Texas abandoned its Watson project.
Perhaps offering if not a cure for cancer, then a better way of treating it was a bold step for any technology vendor, especially if things then don’t go to plan, which they don’t seem to be as far as Watson is concerned. Its missteps have led to some wider negativity around AI and machine learning, which I think are misplaced. I heard some similar sentiments – though many conflicting and positive ones too – at last night at London.AI’s excellent #LONDONAI10 event, which was squarely focused on the intersection of healthcare and AI/machine learning. But there are plenty of other companies attempting to use technology to detect cancer early, notable among them Grail, which has raised a staggering $1.1bn in funding from six investors in two rounds.
I know how hard healthcare IT using machine learning can be because I had a go at it when I ran product marketing at Recommind. We got to the stage of a proof-of-concept (PoC) with very promising early results using our machine learning technology, but then issues elsewhere in the company meant our funding dried up and we had to give up on the healthcare market. But I learned a lot getting to that stage. So as I say, this is a tough nut to crack, and not just for technical reasons, though they do play a major part. Many other issues come into play.
Healthcare is a fragmented world of individual physicians, physician’s practices, hospitals and ever-growing hospital systems, insurance companies and government agencies, which makes data interoperability a major challenge. Large investments in expensive MRI machines, CT scanners and various types of patient monitoring devices have brought tremendous advances, but healthcare is one of the few (only?) industries where the application of technology has driven costs for customers (i.e. patients) up, rather than down. The data that comes from such machines is well understood, although often takes humans to interpret it, though we already know machine learning can help here.
Plus, now we have people who consider themselves healthy using wearable technology to collect data and measure their own well-being, thus expanding the reach and role of technology in healthcare and the volume and variety of health data. Just last week the Food and Drug Administration approved the first mobile app to help treat substance use disorders, from Pear Therapeutics, pointing the way ahead for a new relationship between healthcare and technology.
There is a pent-up desire to access healthcare data to improve services, track widespread health trends, develop drugs more quickly and improve insurance claim processes. There is also an enthusiasm among many on healthcare that I’ve come across to use AI/machine learning to help make this happen. In fact, it gained the highest percentage score when we asked healthcare professionals to identify the three most disruptive technologies in healthcare in a recent survey 451 Research did, with more than half (54%) choosing it.
I would just urge those that think all this AI stuff is just a fad and can’t possibly effectively diagnose and perhaps treat illnesses to not get concept fatigue before realizing what’s possible.