Real-World Evidence in the AMNOG Dossier: Which Data Sources Can Robustly Define the Target Population?

Real-world evidence can support the German AMNOG dossier by informing epidemiological estimates, the statutory health insurance (GKV) target population, and the German healthcare context. Relevant sources include national registries, patient registries, epidemiological studies, and German claims data. The German Health Research Data Centre (Forschungsdatenzentrum Gesundheit, FDZ) adds a cross-insurer source of GKV routine data.

The appropriate RWE data source depends on the specific AMNOG question. A larger dataset is not automatically a better dataset. The key question is whether the relevant population, disease characteristics, and treatment criteria can be captured well enough to derive the target population transparently and robustly.

RWE supports epidemiological estimates in Module 3 of the AMNOG dossier

Real-world evidence can support the description of the disease, epidemiology, and GKV target population in Module 3 of the AMNOG dossier. Sections 3.2.3 on prevalence and incidence and 3.2.4 on the number of patients in the target population are particularly relevant.

Prevalence and incidence need to reflect the disease or disease stage covered by the approved indication. Age- and sex-specific differences and, where relevant, differences between other patient groups also need to be considered.

The derivation of the GKV target population needs to be transparent. This includes the individual calculation steps, the sources used, and a range reflecting uncertainty. Patients who are currently untreated may also need to be considered when determining the target population.

The epidemiological analysis therefore addresses a specific question: How many patients covered by German statutory health insurance actually meet the criteria of the indication under assessment?

Four RWE data sources can characterize the AMNOG target population in different ways

Four major source categories can support epidemiological analyses for AMNOG: national registries, patient registries, epidemiological studies, and GKV claims data.

They differ particularly in population coverage and the level of clinical detail available.

A single data source does not need to provide every parameter required for an epidemiological analysis. Particularly for complex target populations, several sources may support different steps in the derivation.

National registries can directly capture incidence and case numbers in Germany

National registries are particularly relevant when a disease is systematically recorded in Germany. Examples include the German Haemophilia Registry and mandatory reporting data collected under the German Infection Protection Act.

A major advantage is the direct representation of the German healthcare context. Depending on the registry, age and other characteristics may also be available.

The limitation is the depth of available information. A registry may reliably capture the number of cases without recording body weight or other clinical characteristics required to define the specific indication.

Additional data sources may therefore be needed to derive the AMNOG target population.

Bedaquiline shows how German registry data can support a stepwise target population derivation

In the AMNOG procedure for bedaquiline, the target population of adolescents with multidrug-resistant pulmonary tuberculosis was derived step by step using German reporting data and additional information.

The starting point was data from the Robert Koch Institute on multidrug-resistant tuberculosis. The relevant GKV target population was subsequently narrowed through four steps:

1. extrapolation of reported MDR-TB cases,

2. restriction to patients aged 12 to under 18 years with a body weight of at least 30 kg,

3. restriction to patients with pulmonary MDR-TB,

4. application of the proportion covered by statutory health insurance.

The initial estimate of 146 to 209 patients with MDR-TB was reduced to 14 to 21 patients in the relevant age group. After accounting for pulmonary manifestation, 10 to 15 patients remained. Applying the GKV proportion resulted in a final target population of 9 to 13 patients.

The bedaquiline example illustrates how strongly individual selection steps can influence the final patient estimate. For small populations, uncertainty should therefore be transparently represented at each relevant step of the calculation.

Patient registries can add clinical characteristics to the AMNOG target population

Patient registries can capture clinical information that may be missing from administrative routine data or national reporting systems. This can include disease stage, body weight, or other patient characteristics.

The additional clinical detail comes with a potential limitation: patient registries do not necessarily capture every patient with the disease. Epidemiological analyses therefore need to assess how representative the registry population is of the German target population.

The central trade-off is clear: A clinically detailed population is not automatically a population-level representative population.

Epidemiological studies can provide missing parameters for the AMNOG target population

Epidemiological studies can provide prevalence, incidence, and clinical characteristics when these data are not directly available from registries or routine data.

For AMNOG, transferability is particularly important. German studies can directly reflect the national healthcare context. International studies can also provide relevant parameters but require an assessment of whether their populations and healthcare settings are transferable to Germany.

Epidemiological studies can also provide proportions for specific subgroups. They can therefore be combined with registry or routine data.

Olaparib shows how multiple sources can define a complex target population

The epidemiological derivation in the AMNOG procedure for olaparib combined data from the German Centre for Cancer Registry Data (Zentrum für Krebsregisterdaten, ZfKD), regional cancer registries, and epidemiological studies.

Starting with patients with ovarian, fallopian tube, or primary peritoneal cancer, the population was narrowed using several clinical criteria:

1. relevant tumor entity,

2. epithelial tumor type,

3. high-grade carcinoma,

4. advanced FIGO stage III or IV,

5. BRCA1/2 mutation,

6. platinum-based first-line chemotherapy without bevacizumab,

7. response to first-line chemotherapy,

8. GKV coverage.

The first step produced an incidence estimate of 7,695 to 8,218 patients. After applying the subsequent criteria, 580 to 701 patients remained in the target population. Applying a GKV coverage proportion of 89.6% resulted in a final estimate of 520 to 628 GKV-insured patients.

The olaparib derivation illustrates a typical epidemiological cascade in AMNOG: one source defines the starting population, while additional sources provide the proportions required to identify clinically defined subgroups.

GKV claims data capture diagnoses and treatment patterns in the German healthcare system

GKV claims data capture diagnoses, treatments, and prescriptions from routine care. They are therefore particularly useful when the objective is to examine actual healthcare utilization and populations covered by German statutory health insurance.

Routine data analyses can define populations using ICD-10-GM diagnoses, prescription data, and other reimbursement criteria.

Their limitation is often the availability of clinical information that is not required for reimbursement. Laboratory values, body weight, specific disease stages, or clinical endpoints may be unavailable or may only be approximated indirectly.

The key methodological question is therefore: Can the criteria of the approved indication actually be operationalized using the available routine data?

Dapagliflozin shows how claims data and epidemiological studies can complement each other

The epidemiological derivation for dapagliflozin in chronic heart failure used both epidemiological studies and GKV claims data.

An analysis based on the WIG2 database defined the population using ICD-10-GM diagnoses and prescription data, among other criteria. Patients with stage 4 or stage 5 chronic kidney disease were excluded using the corresponding ICD-10 codes.

Ejection fraction could not be derived directly from the claims data. Epidemiological studies were therefore used to estimate the proportions of patients with reduced and preserved ejection fraction. In the derivation, 68% of patients with heart failure were assigned to the reduced ejection fraction group and 32% to the preserved ejection fraction group.

The dapagliflozin derivation illustrates why routine data and epidemiological studies can serve different purposes: claims data identify the healthcare population, while additional studies provide clinical characteristics that cannot be directly observed in reimbursement data.

Uncertainty needs to be transparent when deriving the AMNOG target population

The GKV target population is often the result of several data sources, selection steps, and assumptions. A robust derivation therefore needs to make not only the final estimate but also its uncertainty transparent.

Uncertainty can arise from:

• incomplete registry data,

• different case definitions,

• transfer of international studies to Germany,

• missing clinical characteristics in routine data,

• assumptions about subgroup proportions,

• different observation periods.

Ranges can make different plausible assumptions visible. At the same time, the analysis should distinguish clearly between parameters based on directly observed data and those derived from additional assumptions or external sources.

The quality of an epidemiological derivation should therefore not be judged by whether it produces a single highly precise patient number. The relevant question is how transparently the data, assumptions, and uncertainties lead to the final GKV target population.

The German Health Research Data Centre expands RWE analyses with cross-insurer GKV data

The German Health Research Data Centre (Forschungsdatenzentrum Gesundheit, FDZ) provides a cross-insurer research infrastructure for GKV healthcare and claims data. The data include diagnoses, treatments, and prescriptions and are made available in pseudonymized form for research and analytical purposes.

Unlike conventional routine-data projects based on individual health insurers, the FDZ provides access to a broader GKV data base.

This difference can be relevant for AMNOG questions when the representativeness of individual insurer populations is a concern or when larger case numbers are required for small subgroups.

The FDZ does not generally replace conventional claims data. It expands the available options with a data source that has a different population, data structure, and access model.

FDZ data can be particularly relevant for small AMNOG target populations

Cross-insurer FDZ data can offer advantages for small or difficult-to-identify populations. Larger case numbers may broaden the evidence base compared with analyses based on individual health insurers.

Three situations are particularly relevant:

1. Small or rare populations: A larger GKV base may provide additional cases.

2. High requirements for generalizability: The cross-insurer database can reduce dependence on the specific population structure of an individual health insurer.

3. Heterogeneity between health insurers: Differences between insurer populations may be relevant to epidemiological estimates.

The FDZ is less flexible for short-term or highly exploratory questions. Its standardized access process requires data, cohorts, and analyses to be largely predefined.

An FDZ project for an AMNOG dossier requires approximately 10 to 14 months of lead time

An FDZ project involves project conception, study protocol development, application and review, data access, analysis, preparation of results, and integration into the dossier. Approximately 10 to 14 months should be planned for the overall process.

The individual phases can be approximated as follows:

1. project conception and service-provider selection: approximately 1 to 2 months,

2. kick-off and study protocol: approximately 2 months,

3. FDZ application and review process: approximately 3 months,

4. data access and analyses: approximately 1 to 3 months,

5. preparation of results and reporting: approximately 1 to 2 months,

6. dossier integration and finalization: approximately 2 months.

This timeline has direct implications for AMNOG planning. An FDZ analysis cannot realistically be initiated only shortly before Module 3 is finalized.

The FDZ application limits subsequent changes to the analysis

Access to FDZ data follows a structured application process. The application specifies, among other elements, the research objective, research question, methodological approach, reporting years, tables, variables, cohort definition, and planned outputs.

One feature is particularly important for subsequent analyses: Only data and analyses included in the application are accessible later.

This increases the importance of the feasibility and protocol phases. Populations, variables, and analytical steps need to be defined much more precisely before data access than in more flexible routine-data projects.

The FDZ is therefore particularly suitable for clearly defined research questions. Highly exploratory analyses or short-notice changes are more difficult to implement.

The FDZ data structure changes across reporting periods

FDZ data are not available in an identical structure across the entire observation period. The data models differ between reporting periods:

• 2009 to 2015: data model 1,

• 2016 to 2018: data model 2,

• 2019 to 2023: data model 3,

• from 2024: data model 4.

Variables may therefore not be consistently available across all years. Definitions and the level of detail can also change.

For longitudinal analyses, it is necessary to assess in advance whether the required characteristics can be operationalized consistently across the full observation period.

FDZ and conventional claims data differ in population, flexibility, and project timelines

The difference between FDZ data and conventional claims data is not limited to population size. The two approaches have different operational and methodological characteristics.

The choice should therefore not follow a simple “bigger is better” principle. FDZ and conventional claims data are suited to different questions and project conditions.

RWE in the AMNOG dossier should combine data sources deliberately

Complex AMNOG target populations often cannot be derived completely from a single data source. Registries, epidemiological studies, and routine data can provide different information on the same population.

A typical derivation may follow a pathway such as:

Registry population → clinical subgroup → restrictions of the approved indication → GKV proportion → GKV target population

A routine-data analysis may follow a different pathway:

GKV population → diagnosis definition → treatment criteria → exclusion criteria → clinical information from an external source → GKV target population

The AMNOG examples of bedaquiline, olaparib, and dapagliflozin represent different versions of these approaches. What they have in common is that the quality of the target population depends not on one individual source but on the transparent integration of the available evidence.

The AMNOG question should be defined before selecting the RWE data source

Selection of an RWE data source should begin with a precise definition of the target population. Only then can it be determined which data source actually captures the required characteristics.

Four steps are particularly relevant:

1. Define the target population. Which criteria of the approved indication need to be identifiable in the data?

2. Assess the suitability of available data sources. Which source adequately captures the population, healthcare setting, and relevant clinical characteristics?

3. Identify evidence gaps. Which parameters need to be supplemented using additional registries or studies?

4. Plan uncertainty and timelines. Which assumptions affect the patient estimate, and how long will the data analysis require?

For FDZ projects, this planning needs to begin particularly early. A project timeline of approximately 10 to 14 months and limited scope for subsequent changes require the research question to be defined well in advance.

The central question for RWE in AMNOG is therefore not: Which data are available?

The relevant question is: Which data are required to derive the German GKV target population robustly?