Are machines taking over your diagnosis and treatment? The short answer is, yes.
There is unprecedented innovation happening in healthcare, where artificial intelligence is taking over many aspects of diagnosis and care. One would think that something like healthcare is so varied and so complex, that it needs a human’s judgment to make sound decisions. In reality, this variation and complexity is the precise reason machines are doing so well.
Take this example: Innovation is similarly accelerating in imaging techniques, medicine, medical devices and other aspects of healthcare. To keep up, big hospitals and small physicians practices alike rely on prescribed Clinical Pathways to decide on course of treatment. There are many platinum standard organizations that publish these pathways. For example, US Oncology and National Comprehensive Cancer Network are two of the most reputed ones in United States just for cancer. Such pathways are usually updated every quarter.
If you put payer’s in the mix, whether insurance companies or government agencies, the complexity explodes, because payer’s may prescribe to their own set of pathways. As a result each payer-provider partnership negotiates a custom set of rules. Of course, these vary by every state to account for different regulatory policies. In other words, a simple thing like getting a procedure pre-approved by an insurance company becomes a nightmarish task for all parties involved.
In such situations, human judgment is limited in its efficacy at best. Machines are taking over this process already. One of my friends, a medical doctor, is fond of commenting, “Can AI do some basic jobs in medicine today? Yes. Can AI replace a doctor. Not by a long shot.” This is a great articulation of the state of art, however the definition of ‘basic’ keeps evolving. Consider these questions:
- Can a machine look at all the pertinent data, properly structured, and run some rules to see if the case fits the prescribed procedure and either extreme — those that are outright rejects, and those that are outright matches?
- Can a machine convert all the contracts, pathways and guidelines into a structured representation, perhaps an ontology, to then do step 1 much better, essentially closing the gap between outright rejects and outright matches?
- Can a machine look at all the historical cases and start to infer these rules as well as patterns in human judgment calls to further adjudicate every case in step 2 with varying degrees of confidence?
- Can a machine convert all electronic health data to a structured format to further improve confidences in step 3? This includes text, images.
- Can a machine start to look at comprehensive personal and family health history to further improve the confidences in step 3?
The answer to all these questions is yes today or is going to be very soon. There always will be situations that machines can’t resolve, but their numbers are going to go down over time. The questions about regulation, ethics, bias and fail-safes also remain. I always prescribe to my clients that for critical applications like healthcare, human supervision is a must.
Still, machines are already taking over. And that is a good thing.