(Cross-posted from Klick Health blog.)
This past week, IBM’s Jeopardy-winning AI Watson furthered its medical career with new jobs that could significantly disrupt healthcare and permanently shift pharmaceutical marketing practices.
That might seem dramatic given Watson’s humble start as a game show contestant. But ever since it (I keep struggling to not say “he”) trounced the world’s best Jeopardy players, I’ve followed its rapid progress from Alex Trebek’s stage through medical education and now into gainful employment. With each step, I’ve become increasingly convinced that Watson and systems like it will drive or at least facilitate fundamental changes in healthcare—and healthcare marketing. So how might we respond?
From Jeopardy to oncology
For those who don’t know, it was reported this week that Watson will officially begin work helping doctors diagnose and treat patients, while helping insurers evaluate treatment coverage. This follows many months of Watson getting educated in medicine. For oncology, Watson was trained by the world’s best oncologists and has effectively consumed, as I understand it, every piece of useful data on the topic, including the latest research—with which it stays continuously up to date. It then uses the same approach it used on Jeopardy to apply that knowledge to diagnosis and treatment.
Physicians give Watson a case, just as Trebek gave it an answer, and Watson gives them diagnostic and treatment recommendations with varying probabilities of correctness. (Example: A 95% chance that, based on the information provided, a person has prostate cancer.) Physicians then choose to agree or disagree with Watson. But, as one oncologist noted during a pilot, it’s hard to disagree with a system that knows exponentially more than you, is trained by the world’s best physicians, and is completely up to date with the latest research.
And Watson isn’t the only system showing promise at improving diagnosis and treatment while reducing costs. Also this past week, Indiana University researchers reported on predictive modelling techniques that improve patient outcomes by 40% and reduce treatment costs by 50%.
Towards a new discipline: “Artificial Intelligence Optimization”
While the creators of these systems take pains to say they’re not displacing doctors, just augmenting them, I consider that political correctness rather than fact. With soaring costs, doctor shortages and challenges for doctors to keep pace with the increasing volume of research and data, it seems inevitable that a shift to artificially intelligent doctors will occur—especially since their “brain” can be distributed as software or made available via the cloud. (At the most controversial end of the spectrum, venture capitalist Vinod Khosla says machines will do 80% of what doctors currently do.)
Such a shift is already happening with web- and smartphone-based consumer tools. For example, Symcat provides patients with free big data-driven diagnoses, Medify provides patients with disease and treatment guidance based on evaluation of published research, and Treato, which I’ve written about previously, analyzes social media conversation to determine what real patients think of treatments.
If the shift towards machine-driven diagnostics and treatment, and away from physician-driven, continues, it could have a significant impact on pharmaceutical marketing. How do you influence an algorithm?
From my perspective, the common thread through these services is their reliance on data–structured and unstructured. Hence the focus of pharmaceutical marketing might need to shift increasingly towardsmaximizing the generation and digital distribution of data that improves appropriate diagnosis and treatment of patients who can benefit from an intervention. For example, investment might be warranted in:
- Additional clinical trials that result in published data Watson and other systems can use for diagnostic, treatment and insurance recommendations
- Encouraging patients to talk openly (in a compliant way, of course) in public forums about their positive experiences with a treatment, so social media aggregators pick up and analyze the posts
- Data monitoring to identify and address prospective negative data that could adversely impact treatment usage
Of course, these are just some early thoughts. I imagine that as this trend hastens, we’ll develop dedicated disciplines to optimize data for artificial intelligence and analytics engines the way we currently optimize content for search engines. The good news is that, overall, this should improve the application of data to diagnosing and treating patients, while helping address ballooning healthcare costs. The challenge is that the path is uncertain; Watson is an amazing diagnostician, but it can’t predict the future.
At least not yet.