To pay or not to pay for AI applications in radiology?

In an article recently published in npj digital, the authors pose a question: “to pay or not to pay for artificial intelligence applications in radiology?” They go on to propose a framework designed to help establish who should pay for what when it comes to these applications, concluding that it comes down to the nature of the benefits and the evidence in this space.

The authors highlight the importance of motivating and incentivising adoption, and as such, the question of who will pay is “critical”. They note that healthcare systems need to find the right balance when determining which AI applications to pay for separately, which they acknowledge can be a challenge.

After examining the current reimbursement pathways for AI algorithms in radiology and reviewing the evidence basis for current reimbursement, the authors have developed a framework to help establish payment criteria.

The authors describe the X axis of this framework as as “the benefit the application brings” – whether it improves provider efficiency or convenience, improves diagnostic performance or provides new diagnostic information not otherwise possible from the imaging study. The Y axis is “whether the AI application has evidence supporting the improvement of clinical and/or economic outcomes”.

By answering yes or no to statements related to these two criteria, stakeholders are able to see whether they should pay for the application(s) in question.

As a rule of thumb, the authors “propose that radiology AI applications that improve diagnostic performance of the imaging study or provide new diagnostic information that did not hitherto exist, must be reimbursed separately provided they have evidence supporting that this improved diagnostic performance leads to improved outcomes from a societal standpoint.”

In addition, they suggest that payers must consider separate payment for radiology AI applications that improve diagnostic performance “substantially”, or provide “transformational new diagnostic information even without supporting clinical evidence” if the anticipated benefits are large and “reasonably certain”.

However, “we need to recognise that developing such evidence for diagnostics may be challenging sometimes. Radiology AI applications do not have the strong intellectual property benefits that pharmaceuticals enjoy.” In addition, the authors acknowledge that developing long-term evidence on clinical outcomes may be particularly challenging if this requires patients to be followed through diagnosis, treatment, and subsequent clinical outcomes.

Citation: Lobig, F., Subramanian, D., Blankenburg, M. et al. To pay or not to pay for artificial intelligence applications in radiology. npj Digit. Med. 6, 117 (2023).