Quick Buzz Feed

Reimbursement Puzzle: How Should Doctors Be Paid for Using AI Tools?

Gary Lloyd | Sep 24,26 | 01:34 EST

Technology

As AI tools play an ever larger role in healthcare, one question remains: How will doctors get paid for using them?

‘How the Heck Do We Price That?’ The integration of AI platforms into medicine has sparked a critical debate about how private health insurers and government payers, such as Medicare, will compensate clinicians for using these advanced tools. A key challenge arises from AI's unique cost structure: high upfront development expenses contrasting with near-zero marginal cost per subsequent deployment. Lee Fleisher, former chief medical officer at the Centers for Medicare & Medicaid Services (CMS), expressed caution, indicating that CMS has not yet committed to offering additional payments for AI use, despite a federal proposal suggesting a 60%-80% reimbursement rate compared to human services. Ateev Mehrotra of Brown University notes that traditional Medicare reimbursement models are based on clinician time and effort, which is fundamentally at odds with AI's efficient, repeatable nature. Currently, while over a thousand AI-enabled medical devices have received FDA authorization, only a few AI-based services, such as diabetic retinopathy screening and cardiac CT fractional flow reserve (FFR-CT) calculations, have specific Current Procedural Terminology (CPT) codes for reimbursement. Costs for efficiency and administrative AI tools, like ambient scribes, are typically absorbed by the healthcare system as overhead. Narges Sharif Razavian, a health technology expert at NYU Langone Health, highlights the significant lag (7-10 years) required for FDA-authorized technologies to achieve routine reimbursement, by which time the underlying AI may be outdated. She advocates for direct reimbursement of AI services to support developers' extensive upfront investments and help providers manage implementation costs, asserting that AI-augmented care should be considered standard patient entitlement. A generous per-use reimbursement, exemplified by the ~$900 Medicare payment for FFR-CT, can boost adoption but also raises concerns about potential overpayment as AI computing costs continue to decline. Should AI Tools Be Treated as a Practice Expense? The discussion extends to whether AI tools should be categorized as a practice expense rather than assigned individual CPT codes, akin to electronic health records or rent. Mehrotra argues this approach would allow physicians to recoup initial investments through improved efficiency. Over time, Medicare could gradually reduce underlying payments to maintain cost neutrality, reserving separate payments only for high-value and underutilized AI services. Ravi B. Parikh, a medical oncologist, proposes a three-tiered payment system for AI software: classifying efficiency tools (like ambient scribes) as overhead, integrating decision-support AI into existing service payments with potential new technology add-ons, and reserving separate, lower per-use reimbursement for autonomous AI platforms that could replace clinician services. The crucial distinction lies in whether AI replaces or supplements human billing codes. Simply adding AI payments to existing billing could drastically increase healthcare spending without necessarily improving patient care, especially given current concerns about healthcare affordability. Fleisher emphasizes that FDA authorization, while ensuring safety and efficacy, does not prove that an AI service is 'reasonable and necessary for the diagnosis of an illness or injury' as mandated by Medicare. He stresses the need for CMS to demand evidence of meaningful improvements in patient outcomes, not merely the detection of more abnormalities. The CMS Innovation Center's ACCESS Model is suggested as a mechanism to rigorously test AI services lacking an established Medicare benefit category, requiring simultaneous cost neutrality or savings and improved patient outcomes. Fleisher warns that adding AI services to the budget-neutral physician fee schedule could ultimately reduce physician payments. While a 60%-80% reimbursement proposal is being considered, experts suggest that a single, universal reimbursement strategy is unlikely. Instead, the payment rate will likely be complex, dependent on the AI’s specific function, its collaborative or substitutive role with clinicians, the remaining clinical work required, and demonstrated patient benefits.

‘How the Heck Do We Price That?’

The integration of AI platforms into medicine has sparked a critical debate about how private health insurers and government payers, such as Medicare, will compensate clinicians for using these advanced tools. A key challenge arises from AI's unique cost structure: high upfront development expenses contrasting with near-zero marginal cost per subsequent deployment. Lee Fleisher, former chief medical officer at the Centers for Medicare & Medicaid Services (CMS), expressed caution, indicating that CMS has not yet committed to offering additional payments for AI use, despite a federal proposal suggesting a 60%-80% reimbursement rate compared to human services. Ateev Mehrotra of Brown University notes that traditional Medicare reimbursement models are based on clinician time and effort, which is fundamentally at odds with AI's efficient, repeatable nature. Currently, while over a thousand AI-enabled medical devices have received FDA authorization, only a few AI-based services, such as diabetic retinopathy screening and cardiac CT fractional flow reserve (FFR-CT) calculations, have specific Current Procedural Terminology (CPT) codes for reimbursement. Costs for efficiency and administrative AI tools, like ambient scribes, are typically absorbed by the healthcare system as overhead. Narges Sharif Razavian, a health technology expert at NYU Langone Health, highlights the significant lag (7-10 years) required for FDA-authorized technologies to achieve routine reimbursement, by which time the underlying AI may be outdated. She advocates for direct reimbursement of AI services to support developers' extensive upfront investments and help providers manage implementation costs, asserting that AI-augmented care should be considered standard patient entitlement. A generous per-use reimbursement, exemplified by the ~$900 Medicare payment for FFR-CT, can boost adoption but also raises concerns about potential overpayment as AI computing costs continue to decline.

Should AI Tools Be Treated as a Practice Expense?

The discussion extends to whether AI tools should be categorized as a practice expense rather than assigned individual CPT codes, akin to electronic health records or rent. Mehrotra argues this approach would allow physicians to recoup initial investments through improved efficiency. Over time, Medicare could gradually reduce underlying payments to maintain cost neutrality, reserving separate payments only for high-value and underutilized AI services. Ravi B. Parikh, a medical oncologist, proposes a three-tiered payment system for AI software: classifying efficiency tools (like ambient scribes) as overhead, integrating decision-support AI into existing service payments with potential new technology add-ons, and reserving separate, lower per-use reimbursement for autonomous AI platforms that could replace clinician services. The crucial distinction lies in whether AI replaces or supplements human billing codes. Simply adding AI payments to existing billing could drastically increase healthcare spending without necessarily improving patient care, especially given current concerns about healthcare affordability. Fleisher emphasizes that FDA authorization, while ensuring safety and efficacy, does not prove that an AI service is 'reasonable and necessary for the diagnosis of an illness or injury' as mandated by Medicare. He stresses the need for CMS to demand evidence of meaningful improvements in patient outcomes, not merely the detection of more abnormalities. The CMS Innovation Center's ACCESS Model is suggested as a mechanism to rigorously test AI services lacking an established Medicare benefit category, requiring simultaneous cost neutrality or savings and improved patient outcomes. Fleisher warns that adding AI services to the budget-neutral physician fee schedule could ultimately reduce physician payments. While a 60%-80% reimbursement proposal is being considered, experts suggest that a single, universal reimbursement strategy is unlikely. Instead, the payment rate will likely be complex, dependent on the AI’s specific function, its collaborative or substitutive role with clinicians, the remaining clinical work required, and demonstrated patient benefits.