Transferring evidence from one population to another may be scientifically plausible and accepted by the European Medicines Agency (EMA) for regulatory purposes. In Germany’s AMNOG early benefit assessment, however, this is not automatically sufficient. The practice of Germany’s Federal Joint Committee (Gemeinsamer Bundesausschuss, G-BA) shows that the decisive question is whether the available evidence can also answer the specific German assessment question in the target population.
An analysis of 52 AMNOG procedures involving evidence transfer shows recurring patterns. Particularly relevant are the comparability of the treatment setting, the suitability of endpoints and the question of which evidence actually supports the assessment of added benefit.
For international teams, evidence transfer in this context means using evidence from one population to inform the German benefit assessment in another population. The number of supportive arguments is not decisive. Baseline treatment effect, comparator setting, endpoints and population need to form a consistent rationale for transfer.
Selumetinib: evidence transfer between regulatory extrapolation and German benefit assessment
The current procedure for selumetinib in children aged 1 to under 3 years with neurofibromatosis type 1 and symptomatic, inoperable plexiform neurofibromas illustrates how differentiated evidence transfer can be assessed under AMNOG.
At the time of marketing authorisation, efficacy results were not yet available for this age group. The EMA extrapolation was primarily based on exposure modelling using data from SPRINT and SPRINKLE. For the German benefit assessment, however, a second SPRINKLE data cut submitted during the commenting procedure also provided results on morbidity and other endpoint categories.
The G-BA considered transferability from the older populations to be limited because of age-related differences. Direct evidence from SPRINKLE was instead decisive for deriving added benefit. Despite the single-arm study design, the change in target lesion volume was considered a patient-relevant endpoint in this particular disease setting. The natural course of the disease, with no spontaneous remissions expected, and the clinical assessment were also relevant. Consistent results from SPRINT in children aged 3 years and older and adolescents provided additional support. The G-BA ultimately concluded that there was a hint of non-quantifiable added benefit.
The case illustrates a central point: evidence transfer under AMNOG is not a binary yes-or-no decision. What matters is which evidence actually supports the added-benefit assessment and which evidence plays only a supportive role.
Why EMA extrapolation and evidence transfer under AMNOG are not the same
Regulatory extrapolation and evidence transfer in the German benefit assessment address different decision questions. The different requirements should therefore not be understood as a contradiction between the EMA and the G-BA.
At the regulatory level, the central question is whether efficacy and safety in an additional population can be sufficiently supported using existing data. Disease and mechanism of action, pharmacokinetic and pharmacodynamic data or exposure models may play an important role.
AMNOG adds a different assessment question: Is the evidence suitable for assessing patient-relevant added benefit versus the relevant German appropriate comparator therapy (zweckmäßige Vergleichstherapie, zVT) in the specific target population?
This requires clarification of four points:
Which treatment effect is intended to be transferred?
Against which comparator was this effect demonstrated?
Are the relevant endpoints suitable for the benefit assessment?
Are the source and target populations sufficiently comparable for the specific assessment question?
A regulatory extrapolation accepted by the EMA can support this rationale. It does not replace the AMNOG-specific assessment.
What previous AMNOG procedures show about evidence transfer
Our analysis identified 52 AMNOG procedures involving evidence transfer. Twelve particularly informative procedures were examined in greater depth using the available documents from the EMA, the Institute for Quality and Efficiency in Health Care (IQWiG) and the G-BA; the current selumetinib procedure was considered additionally.
The cases cover different constellations: accepted and rejected transfers, regulatory extrapolation without corresponding transfer under AMNOG, differences in treatment setting, problematic endpoints, direct data from the target population and different weighting of evidence by IQWiG and the G-BA.
The 52 procedures do not allow a general probability of success to be derived. They differ in indication, evidence base and assessment question, and several belong to the same drug or indication families. The recurring assessment patterns are therefore more informative.
1. EMA extrapolation alone is not sufficient
Several procedures show that regulatory extrapolation accepted by the EMA can be an important supportive argument, but is not a sufficient condition for evidence transfer under AMNOG.
For nintedanib, forced vital capacity (FVC) played an important role in the evidence base. FVC was not used for the benefit assessment, however. The G-BA considered the submitted surrogate validation for mortality unsuitable, and the derivation of patient-relevant morbidity effects was also not established.
For bedaquiline, the G-BA did not follow the proposed evidence transfer, among other reasons because of unclear comparability of the patient populations and disease symptoms as well as further methodological limitations.
Selumetinib also confirms the fundamental distinction: the EMA extrapolated efficacy primarily on the basis of exposure modelling, whereas the G-BA was able to use direct data from the target population that had become available for the benefit assessment.
2. Structural differences can limit evidence transfer
Evidence transfer does not work as an additive checklist in which a sufficient number of positive criteria can compensate for a fundamental difference.
This is particularly clear for the treatment setting. In the pembrolizumab procedure, the G-BA considered a comparable disease presentation and comparable efficacy and safety to meet important minimum requirements in principle. The transfer nevertheless failed because brentuximab vedotin was not used as the zVT in a sufficiently comparable treatment setting in adults and children. The G-BA explicitly concluded that the adult study could therefore not be transferred to the paediatric population.
Similar breaks can arise from endpoints or insufficient comparability between populations.
The procedures examined in depth therefore suggest that residual uncertainty may be more addressable than a fundamental lack of alignment in treatment setting, endpoint or population. This is a cross-case interpretation of the procedures analysed, not a formal G-BA decision rule.
3. Direct evidence from the target population can reduce uncertainty
Direct data from the target population can be particularly valuable when they specifically test the assumptions underlying the transfer.
Dupilumab also shows why the decision to transfer evidence and the final added-benefit conclusion should be considered separately. The G-BA transferred results from the 18-to-under-40-year age stratum of the CHRONOS study to children aged 6 months to 5 years whose disease presentation was sufficiently similar to that of adults. Additional data from PRESCHOOL were also included in the assessment. Nevertheless, the final added benefit was not proven.
Letermovir likewise shows that IQWiG and the G-BA may weight evidence differently in the context of a transfer. In addition to the dossier assessment, the procedure included an IQWiG addendum and clinical comments before the G-BA reached its final decision.
Selumetinib provides another current example. Direct evidence from SPRINKLE formed the basis of the assessment, while results from the adjacent age group supported the observed reduction in tumour volume despite explicitly limited transferability.
Four questions for a robust evidence transfer under AMNOG
The procedures do not provide a universal statistical extrapolation method. Depending on the data constellation, evidence from a source population can enter the assessment in different ways, ranging from transfer of study results to supportive interpretation of direct data from the target population.
Four questions should be addressed early. They should not be understood as rigid sequential steps. In particular, medical comparability between populations provides a foundation against which treatment effect, comparator setting and endpoints also need to be assessed.
1. Which effect is intended to be transferred?
The evidence intended for transfer should first be defined precisely. Is there a relevant positive treatment effect in the source population? Which endpoint supports it? And is it suitable for supporting an added benefit in the target population?
If there is no relevant positive baseline effect, there is no positive effect on which a transfer rationale can be built.
2. Is the comparator setting sufficiently aligned?
For the AMNOG assessment, the zVT, line of therapy, treatment regimen and healthcare context are relevant. Differences between the source and target populations can substantially limit transfer, even where disease and mechanism of action appear comparable.
3. Do the endpoints support the transfer?
Patient relevance, operationalisation and measurability need to be sufficiently aligned across populations. For surrogate endpoints, an additional question is whether adequate validation exists for the claimed patient-relevant effect.
Using the same endpoint label in two studies does not by itself provide a robust basis for evidence transfer.
4. Are the source and target populations sufficiently comparable?
Disease, pathophysiology, mechanism of action, pharmacology and clinical response should be examined for relevant differences. Potential effect modification, for example by age or disease stage, should be explicitly considered.
Direct data from the target population are particularly informative when they actually test these assumptions.
What evidence transfer means for evidence and AMNOG dossier strategy
For practical planning, the focus therefore shifts. The starting point should not be a retrospective justification of medical comparability, but prospective planning of an evidence base capable of addressing the later German AMNOG question.
Define the intended transfer early
During evidence planning, teams should already be clear about which effect from which source population may need to be used for which target population.
Assess the German comparator and treatment context early
A medically convincing rationale can lose its value if the relevant German AMNOG comparison does not match the available evidence.
Plan endpoints across populations
If later evidence transfer is foreseeable, teams should consider early which patient-relevant endpoints can be collected and interpreted across populations.
Use direct target-population data strategically
Direct data are particularly valuable when they test central assumptions underlying the transfer. Selumetinib also shows that the available data set can change during the AMNOG procedure. The second SPRINKLE data cut with additional endpoint data was submitted only during the commenting procedure and subsequently formed the basis of the benefit assessment.
Anticipate potential IQWiG criticism before submission
A distinction should be made between fundamental problems and addressable residual uncertainty. An unsuitable comparator setting or endpoint basis can only be corrected to a limited extent during the commenting procedure. Clinical uncertainties, by contrast, may potentially be further contextualised through additional analyses, direct data, information on the natural course of disease or clinical expertise.
Conclusion: evidence transfer under AMNOG is a chain, not a checklist
Previous G-BA practice does not identify a single criterion that automatically enables evidence transfer. The decisive question is whether baseline effect, comparator setting, endpoints and population form a consistent rationale for transfer.
EMA extrapolation, pharmacological plausibility, direct data from the target population and clinical context can strengthen this rationale. They cannot compensate for a fundamental lack of alignment at key points.
At the same time, the procedures analysed show that residual uncertainty does not necessarily mean that transfer will fail. What matters is which evidence actually supports the benefit assessment, which evidence plays a supportive role and how robust the overall rationale is.
For dossier strategy, the implication is that the requirements of a later evidence transfer should already shape evidence planning rather than being addressed only when the AMNOG dossier is prepared.