Galen Senior Fellow Brian Blase looks to the future in a new paper describing a medical innovation that can improve health care productivity—largely through more accurate, timely, and affordable diagnoses—and expand access to care by more efficiently deploying specialists’ time and making tests less expensive.
In a new analysis, “Properly Incentivizing Health Care Innovation & A Case Study on the First Autonomous Artificial Intelligence,” Blase describes the benefits of an FDA-approved innovation which uses autonomous artificial intelligence (AI) to generate a medical decision without human oversight. But Medicare’s outdated payment methodologies may prevent seniors from receiving its benefits.
In this paper, Blase examines why productivity in health care lags the rest of the economy, focusing on the problems with Medicare’s payment methodology. Medicare primarily bases reimbursement on complex formulas that place value on inputs like physicians’ time, failing to account for and reimburse automations that can produce faster, more accurate diagnoses.
To demonstrate the problem, Blase uses a case study of an autonomous AI that can diagnose diabetic retinopathy, a condition which can lead to severe eye damage and vision loss, including blindness. Autonomous AI delivers a much cheaper and more accurate diagnosis than a clinician’s examination. By improving the diagnosis and reducing its cost, the technological breakthrough can increase access to important treatments that will prevent vision loss and blindness.
But this promise will not be fulfilled unless the Centers for Medicare and Medicaid Services appropriately reimburses autonomous AI, something it failed to do in its 2021 proposed Medicare physician fee schedule rule, in large part because of antiquated and perverse rules for how it sets Medicare payments.
Washington must figure out how to properly compensate technological breakthroughs that radically improve diagnoses and treatment to move our health sector into the 21st century. The first step is to fully understand the problem, and Blase’s analysis and the case study demonstrate the problem and need for reform.
The terms AI and autonomous AI get thrown around a lot these days. Lots of hype and under very controlled conditions, it can indeed appear to be impressive. However, much of the supposed AI in commercial use these days is little more than systems utilizing specialized programming geared towards a very specific and narrow field.
For most medical uses, a team of programmers loads thousands or tens of thousands of case studies into a system, along with the diagnosis that was done by the case physicians and specialists. The more complex the medical condition, the more case studies and diagnosis have to be entered into the system to achieve an accuracy rate in-line with being equal the medical specialists who normally do the work. After all, the objective of all these AI systems is to equal or exceed the quality of diagnosis performed by the human doctors.
One the system has been properly programmed, the system then can associate certain medical readings and imaging data to correlate the best choice for a diagnosis. This isn’t true AI of course, as the system doesn’t think. It more accurately compares the programmatic data and imaging analysis based on the comparison of the data programmed into the system against the patient’s data. The more case data programmed into the system, the more accurate the probable diagnosis will be. Various hospitals have been testing so-called AI systems for at least 6 or 7 years now and the results are mixed at best for the cost differential involved. Results were acceptable for some diseases, but inadequate for others being a second issue.
So my guess is that CMMS is unwilling to authorize reimbursement of AI services until the accuracy and cost of these system is more comparable to what a human physician doing this work would be. At some point in the relatively near future, humanity will develop a true AI. A system that will not only be equal to the human physician, but far exceeding both the speed and capabilities of the human physician. Thankfully, as Elon Musk and others would say, we’re not there yet.