A pragmatic framework for evaluating the financial impact of radiology AI
Experts are offering new insight into radiology AI implementation, suggesting these decisions do not always offer the return on investment vendors promise.
Integrating AI into clinical workflows is typically accompanied by a hefty bill, making the decision one that needs to be approached strategically, they write in JACR. Choosing to buy an AI product should depend on several factors, including caseload, patient demographics and staffing levels.
“As more commercial AI solutions become available, radiology departments must assess their clinical and financial value to inform adoption decisions. Although there are reimbursement codes for some algorithms, the majority of AI applications in radiology are not reimbursable,” Miriam A. Bredella MD, MBA, vice chair of strategy with the department of radiology at NYU Langone Health, and colleagues noted. “As a result, return on investment is often driven by potential efficiency gains, enabling radiologists to interpret more studies in the same amount of time."
Researchers sought to gain a more detailed understanding of how AI affects organizations’ efficiency, contribution margin, and return on invested capital. To do this, they developed a financial calculator that can estimate the difference between baseline and AI-integrated workflows. The calculator incorporated annual case volume, professional reimbursement per study based on Current Procedural Terminology (CPT) codes, radiologist compensation and productivity, interpretation times, AI acquisition and implementation costs, and the potential for downstream revenue.
Researchers focused on three use cases for AI: (1) an intracranial hemorrhage triage algorithm for noncontrast head CT, (2) a pulmonary embolism triage algorithm for CT pulmonary angiography in the emergency department, and (3) a breast cancer detection algorithm for screening mammography. Radiologist compensation, reimbursement and AI acquisition/implementation costs were modeled using national benchmarks.
Their findings emphasize the importance of gathering data prior to implementing AI, as it may not be worth the investment for some organizations, according to their analysis.
Here are their takeaways for each use case:
For the intracranial hemorrhage triage algorithm, interpretation times dropped by around 1.15 minutes per case. This allowed for more interpretations per hour, resulting in a contribution margin improvement from −6.0% to −2.4%. Financially, the algorithm remained favorable with the average radiologist salary set at $440,000, but this outlook decreased alongside lower salaries under $269,000 and investments into AI that exceeded $327,000. The team determined investments in similar algorithms would likely be financially favorable depending on caseload.
The outlook for the pulmonary embolism triage algorithm was less favorable. While it reduced interpretation times by around 0.83 minutes per case, it also produced a negative contribution margin of −10.1% and a negative return on invested capital of −76.6%. The team estimated that the algorithm would either need to produce a reduction in read times of around 3.55 minutes per case or lower the amount invested from $200,000 to approximately $47,000 to produce adequate efficiency or financial gains.
Finally, the impact of the breast cancer detection algorithm was multifactorial. Its success was dependent on its downstream value, rather than interpretation revenue and read times. The algorithm was perceived as financially favorable in the long-term due to its increased cancer detection capabilities. In this study, the algorithm detected an additional 35 breast cancers, which resulted in approximately $937,590 in annual downstream contribution margin. The group estimated that around $388,000 in downstream contribution margin would be required to offset the costs of implementing the AI tool.
“Our findings demonstrate that efficiency gains alone may be insufficient to justify investment in some triage algorithms, whereas diagnostic algorithms may generate substantial financial value through downstream detection of actionable disease. Incorporating algorithm-specific value pathways, sensitivity analyses, and local financial inputs can support more transparent and informed AI purchasing decisions,” the authors concluded.
Read more here.
In addition to her background in journalism, Hannah also has patient-facing experience in clinical settings, having spent more than 12 years working as a registered rad tech. She began covering the medical imaging industry for Innovate Healthcare in 2021.