DOGRA CMAA

After reviewing the CMAA proposal alongside the Dogra et al. analysis, I’d say:

Where CMAA Helps (in a narrow sense)

• Recognition of algorithmic services without physician work. CMAA is designed for cases where an algorithm produces a clinically meaningful output (risk score, index, probability) but no physician interpretation is embedded. That’s a real coding gap today, especially for “autonomous” AI outputs.

• Descriptor structure. CMAA descriptors force applicants to specify discipline, inputs, result type, and Appendix S category. That helps standardize how algorithmic services are presented, and at least acknowledges the augmentative vs. autonomous taxonomy.

• FDA tie-in. Requiring FDA authorization and linking codes to specific proprietary algorithms (via Appendix X registry) ensures these aren’t speculative codes.

Where CMAA Needs Work (relative to Dogra’s insights)

• Granularity of value capture. Dogra stresses that generalist AI (radiology, genomics, digital pathology) creates value across multiple tasks, modalities, and autonomy levels — and evolves under FDA’s PCCP framework. 

• CMAA, by contrast, is still static and siloed: one code = one algorithm = one output. That doesn’t solve the problem of dynamic AI that shifts roles and autonomy over time.

• Valuation mechanics. Dogra highlights the fundamental problem that CMS’s practice expense model can’t handle SaaS costs (licenses, cloud, maintenance). 

• Like other AMA codes, CMAA codes don’t tackle valuation at all; they just create a home for descriptors. That leaves the RVU and PE assignment problem entirely untouched, and an issue for CMS and other payors.

• Coverage and payment relativity. Dogra underscores that generalist AI needs composite RVU strategies, cross-setting consistency, and outcomes-linked payment options. 

• CMAA is silent as it’s a coding container, not a reimbursement framework.

• Adaptability and PCCPs. Dogra emphasizes the need for codes that remain valid as algorithms adapt under FDA’s predetermined change control plans. 

• CMAA, however, could require new codes for material changes (inputs added, autonomy level shifts, etc.). 

  That reintroduces the “coding whack-a-mole” problem Dogra is trying to get us past.

• Risk-based models. Dogra asks how SaaS/AI integrates into APMs and chronic disease management. CMAA codes don’t touch this; they are still FFS-centric artifacts.

My Take

CMAA is more a taxonomy exercise than a solution to the real reimbursement issues. It might help with cataloging FDA-cleared algorithms that don’t fit Cat I or III.  But others will  need to the ball on valuation, relativity, coverage, or adaptability.  

---