In Germany, AI lung reading is a cost the radiologist carries and the payer never sees. With no separate billing, vendors sell to practices and hospitals on workflow, and IQWiG has signalled that diagnostic accuracy alone will not carry a reimbursement case. Its stated bar is prospectively planned studies large enough to follow patients from diagnosis through therapy.

• AI vendors: screening is the prize, and the evidence there is contested. Annual low-dose CT for heavy smokers aged 50 to 75 creates the reading volume, and the only two includable accuracy studies in that setting disagree. IQWiG excluded other studies for not confirming whether flagged nodules were actually cancer, so a verified reference standard is the minimum entry requirement.
• Evidence planners: design for the diagnostic-therapeutic chain now. IQWiG frames the question as whether patient-relevant benefit can be shown in RCTs, and says wider use of AI elsewhere in medicine is no justification for routine use here. The topic window closed in July 2023 and the verdict arrived in October 2026, so studies started now are what the next German review will see.
• Hospital buyers: the liability and the risk stay in-house. IQWiG notes that legal responsibility for every finding remains with the physician and flags de-skilling as radiologists come to rely on the software. Without a billing route or outcome data, the purchase case rests on throughput and workflow, which is where vendors in Germany will have to compete.