Medicare has denied $16M worth of radiology artificial intelligence claims
Medicare denied approximately $16 million worth of radiology artificial intelligence claims during a five-year period, according to new research published Wednesday.
AI use across diagnostic imaging and other medical specialties continues to grow, with regulatory approval of over 1,200 such devices. Payment has helped motivate this push, with the Centers for Medicare & Medicaid Services reimbursing physicians for such imaging-focused AI software since 2018.
To better understand this landscape, researchers recently conducted what they believe is the first nationwide, longitudinal analysis of AI adoption by radiologists in the Medicare program. Radiologists provided and submitted claims for a total of 83,392 AI services between 2018 and 2023, the analysis found.
Of those, about 47% (or 39,535) were accepted and reimbursed by Medicare for a total of $8 million. Conversely, nearly 53% (or 53,857) of services were submitted and denied, representing $16.4 million in rejected payments, experts wrote Sept. 10 in the Journal of the American College of Radiology.
“While submitted and paid claims represent a salient way to measure utilization, particularly under fee-for-service incentives, it is conceivable that AI services were used outside of claims,” corresponding author Joshua M. Liao, MD, MSc—an internal medicine specialist with the University of Texas Southwestern Medical Center, Dallas, and member of the Medicare Payment Advisory Commission (MedPAC)—and colleagues concluded. “Nonetheless, the rapid increase in, but only partial reimbursement for, AI services among radiologists in Medicare underscores structural and behavioral facilitators and barriers to diffusion of AI innovation. These must be addressed in future research, policy, and practice redesign to ensure these technologies are smoothly integrated into practice.”
For the study, Liao et al. used Current Procedural Terminology (CPT) billing codes to identify 23 different AI services approved by Medicare for reimbursement through the physician fee schedule. Researchers restricted eligible services to global and professional component claims made by radiologists. Adoption appeared to increase over time, jumping from 1,507 services in 2018 up to 53,586 by 2023. Accepted services increased from 424 (28.1%) at the beginning of the study period, reflecting $30,000 in payments, up to 17,014 (or 31.8%) in 2023, reflecting $4.6 million.
Denied services also increased during the study period, the analysis found. In 2018, a total of 1,083 services (71.9% of all submitted), reflecting $60,000 in potential payments, were denied. By 2023, there were 36,572 services rejected (52.6%), reflecting $10.2 million in potential payments. During the study period, fractional flow reserve-derived from CT had the greatest utilization with 35,845 services (90.7% of all accepted), reflecting $7.8 million in reimbursement. Remaining utilization was comprised of the LumineticsCore (2,997 services), Low Ejection Fraction AI-ECG (563), Cleerly Labs (84) LiverMultiScan (29), Quantitative Magnetic Resonance Cholangiopancreatography (13), and Coverscan (4).
LiverMultiScan, a noninvasive MRI tool that accurately assesses signs of liver disease, had the highest denial rate at 98.2%, reflecting $900,000 in potential payments. Low Ejection Fraction AI-ECG followed closely, also at 98.2% and representing $2.7 million in potential payments. In contrast, investigators found lower denial rates for fractional flow reserve derived from CT (18.5% or $5.5 million in potential payments) and LumieticsCore (5.8% or $20,000).
During the study period, radiologists most often adopted AI services in on-campus outpatient hospitals (24,254 accepted services or 61.3%). Physician offices were next closest (12,721 or 32.2%), while a smaller minority were adopted in off-campus outpatient hospitals (1,365), inpatient hospitals (901), independent clinics (254) and emergency departments (30).
Experts believe their findings highlight inherent challenges with paying for AI via dedicated CPT billing codes.
“While feasible for a small number of codes, this approach may be problematic at scale,” the authors noted. “The proliferation of AI services may trigger the need for large numbers of codes, while rapid technological advances may require dynamic coding in which codes become obsolete and require replacement by new codes. Code-based reimbursement can also inadvertently trigger administrative burden (e.g., documentation to adhere to coverage determinations and avoid denials)—concerns corroborated by the significant service denial rates observed in our study.”
As an alternative, the authors suggested potentially grouping similar AI services into “code families” to permit evolution and “dynamic” coding over time. Radiologists also could consider creating new code and reimbursement systems geared specifically toward AI. Experts see great potential to boost utilization of radiology AI, particularly in value-based care models.
“As incentives for providing care coordination codes may be limited in general use but much stronger in value-based models that financially reward clinicians for cost efficiency and quality outcomes,” the authors wrote. “Similarly, incentives for AI adoption and disincentives toward overuse may prove stronger within outcome-oriented models rather than avenues such as fee schedules.”
Marty Stempniak has covered healthcare since 2012, with his byline appearing in the American Hospital Association's member magazine, Modern Healthcare and McKnight's. Prior to that, he wrote about village government and local business for his hometown newspaper in Oak Park, Illinois. He won a Peter Lisagor and Gold EXCEL awards in 2017 for his coverage of the opioid epidemic.