Real-world evidence from Germany: representativeness analysis and mortality endpoint validation in electronic health record-derived oncology cohorts.

Background

Cancer is a major public health challenge in Germany, representing one of the leading causes of morbidity and mortality nationwide [ 1 ]. As advances in oncology accelerate, the need for high-quality, real-world data (RWD) to inform clinical practice, regulatory decisions, and health technology assessment has never been greater [ 2 – 4 ]. Despite this need, robust oncology RWD in Germany has been limited, in part due to strict data privacy regulations and slow evolution from paper-based to electronic health record (EHR) systems [ 5 , 6 ]. National cancer registries provide essential incidence and survival information but often lack the clinical detail and longitudinal follow-up necessary to evaluate real-world treatment patterns, biomarker testing, and patient outcomes across care settings [ 1 , 7 ]. This limitation is particularly acute given rapidly evolving standards of care, where timely and granular data are critical for understanding new interventions’ impact on diverse patient populations.

The adoption of EHRs in German healthcare has progressed more gradually than in other developed healthcare systems. While mandatory EHR implementation timelines exist in countries such as the United States, German healthcare was largely paper-based through the 2000s and early 2010s. The German e-Health Act (2015) and subsequent regulatory initiatives have accelerated EHR adoption, particularly among larger cancer centers and hospital networks [ 8 ]. However, as of 2024, adoption remains heterogeneous across practice settings. The Flatiron Health Germany Research Database was established to capture longitudinal EHR data precisely during this transition period, providing a unique opportunity to characterize contemporary German oncology practice as the health system modernizes its information infrastructure. The database encompasses both established hospital-based cancer centers with mature EHR systems and community oncology practices transitioning to electronic documentation.

EHR-derived datasets offer a promising approach to address these gaps by capturing rich, longitudinal clinical information from routine oncology care delivery [ 9 , 10 ]. However, without understanding how well these datasets reflect patients’ demographic and clinical characteristics across Germany, research findings may not be applicable to national decision-making. Establishing external validity requires assessing key variables, such as disease stage, histologic subtype, and molecular profile, contextualized with national benchmarks to determine how well these data sources mirror their intended populations. A persistent challenge is identifying suitable external benchmarks, as national cancer registries and epidemiologic studies often lack the recency, clinical depth, or outcome data required for direct comparison with contemporary EHR-derived cohorts [ 7 , 11 ]. Nevertheless, leveraging the best available external data remains essential to situate EHR-derived cohorts within Germany’s broader cancer care landscape.

This paper reports on novel real-world German breast, lung, and colorectal cancer cohorts as examples of curated, disease-specific cohorts now available. Flatiron Health’s EHR-derived datasets encompass growing cancer types, with increasing sample sizes and participating sites. By comprehensively characterizing these cohorts, evaluating representativeness relative to national benchmarks, and validating a composite real-world mortality endpoint, we establish the fitness-for-purpose of the Flatiron Health Germany EHR-derived data and support its use as a reliable real-world evidence (RWE) source for oncology research and decision-making in Germany.

Methods

Data source

This study utilized the Germany Flatiron Health Research Database, a longitudinal, deidentified database derived from EHRs collected across a network of oncology practices in Germany. Participating sites use diverse, boutique oncology EHR systems, each requiring custom data integration with Flatiron Health to extract and harmonize data into a standardized research format. When sites join the network, historical EHR data is extracted retrospectively, enabling cohorts with longitudinal follow-up. The database achieved research-readiness in December 2023 and is maintained with prospective data refreshes at minimum every 90 days. The number of contributing sites has expanded since launch and continues to grow.

German oncology care involves coordination across multiple care settings, with patients often receiving initial diagnosis and treatment planning at a hospital-based cancer center and ongoing treatment at various community practices or clinics. A distinctive feature of German medical documentation is that when patients receive care from external providers, those providers send detailed clinical letters and treatment summaries to the coordinating institution, which are incorporated into the patient’s EHR. This structured communication pathway means that the Germany Flatiron Health Research Database captures treatment information not only from direct encounters at participating sites, but also from out-of-network care documented through incoming provider correspondence stored within the EHR system. Comprehensive unstructured data is an eligibility criterion for a site entering the Flatiron network in Germany. A preliminary feasibility assessment is completed for each site prior to integration to ensure that unstructured data and information flow between care settings is provided.

The provenance and governance have been described previously [ 12 ]. The dataset includes structured data (demographics, diagnosis codes, laboratory results, treatment records) and unstructured data extracted from clinical notes, pathology reports, and radiology reports through technology-enabled extraction and manual review. The biomarkers table includes test type, test date, result and test details. It captures mutations, rearrangements, gene amplifications and protein expression tests. To ensure mortality data completeness, EHR death information was supplemented with additional sources including national cancer registries (available for some sites) and the German postal service. Data processing follows standardized protocols with rigorous quality control procedures per local regulations (Fig. 1 ).

Prospective longitudinal follow-up continues with prespecified variable definitions ensuring consistency across analyses. Data are refreshed at least every 90 days, enabling timely capture of clinical events, treatments, and outcomes. The network continuously expands with new sites and patients (including hospitals with DKG-certified centres), increasing database breadth, depth, and representativeness for oncology research applications.

Study populations

The analytic sample comprised three cancer cohorts: breast cancer, non-small cell lung cancer (NSCLC), and colorectal cancer (CRC). Adult patients (≥ 18 years) were included based on cancer-specific diagnostic criteria with relevant diagnoses documented between January 1, 2016, and December 31, 2024. Breast cancer patients required expert abstraction confirmation of early-stage or metastatic disease. NSCLC patients required confirmed diagnosis with histologic subtypes and biomarker data (epidermal growth factor receptor [EGFR], programmed death-ligand 1 [PD-L1]) captured where available. CRC patients required confirmed adenocarcinoma of colon/rectum through expert abstraction. No treatment history restrictions were applied, allowing comprehensive population characterization.

Staging information was extracted from structured data fields and clinical documentation. When both clinical and pathological staging were available for an individual patient, pathological staging was prioritized as the more definitive assessment, consistent with standard oncologic practice. Discrepancies between clinical and pathological assessments were not separately adjudicated; rather, the pathological assessment served as the reference standard. Stage was recorded at diagnosis as documented in the EHR. Missing or ‘unknown’ stage was preserved as a distinct category for transparency and to allow readers to assess potential selection bias.

For mortality validation, we identified 1,280 patients across 7 cancer types (breast, NSCLC, CRC, small-cell lung cancer, multiple myeloma, prostate cancer, and diffuse large B-cell lymphoma) from one partner site with corresponding German Cancer Registry data, enabling direct comparison of vital status and death dates between the real-world composite mortality variable and a registry reference standard.

Data curation and dataset development

The Flatiron Health Germany database employs a disease-by-disease data curation strategy to develop research-ready datasets. This structured approach involves: (1) identification of source EHR data across all patients with a given malignancy in the participating provider network; (2) systematic data cleaning, variable harmonization, and clinical abstraction performed by a team of specialized oncology nurse abstractors based in Germany; (3) application of standardized policies and procedures and clinically relevant variable definitions tailored to each cancer type; and (4) ongoing prospective maintenance with data refreshes at minimum every 90 days.

At the time of manuscript preparation, breast cancer, NSCLC, and colorectal cancer represented the first three disease cohorts to complete this rigorous curation pipeline and achieve research-readiness for publication. These three cancers were selected for the current representativeness validation analysis because they had achieved: (i) sufficient sample size ( n > 700 per cohort), (ii) longitudinal follow-up with complete key clinical variables (diagnosis, stage, histology, biomarker status), and (iii) availability of external German national benchmarks for comparison. Subsequent to manuscript submission, this curation methodology has been extended to additional malignancies including prostate cancer, multiple myeloma, and diffuse large B-cell lymphoma, with further cancer types in development. The representativeness framework and validation approach presented here can be applied prospectively to each additional disease cohort as it achieves research-readiness, ensuring consistent quality standards across the Germany Flatiron Health Research Database.

Statistical analysis

Descriptive statistics summarized demographic and clinical characteristics including age at diagnosis, sex, stage (pathological prioritized over clinical), histologic subtype, tumor grade, and biomarker status. Frequencies and proportions were calculated for categorical variables; medians and interquartile ranges (IQR) for continuous variables. Denominators for percentages excluded ‘Unknown/not documented’ counts. Kaplan-Meier methods estimated real world overall survival (rwOS).

Representativeness framework

We followed Gerke et al.’s framework for assessing oncology EHR-derived RWD representativeness [ 13 ]. Key variable distributions in the Germany data were compared to published national statistics from the German Centre for Cancer Registry Data (ZfKD) and other publicly available sources. Subgroup analyses evaluated alignment with national patterns across age groups, disease stages, molecular subtypes, and treatment regimens.

We assessed mortality construct validity by stratifying rwOS by key prognostic factors (age, biomarker status) and evaluated clinical plausibility against published literature. Analysis examined data capture completeness for variables relevant to clinical decision-making and health technology assessment. All analyses used standard R statistical software within a trusted research environment, a secure, cloud-based platform enabling data access in compliance with local privacy regulations. Results were interpreted considering dataset coverage and potential selection bias.

Validation of composite real-world mortality endpoint

To validate composite mortality variable quality, we calculated validity metrics for a comparison of the Germany composite mortality variable to a German Cancer Registry reference standard. We assessed (1) sensitivity and specificity for vital status classification (alive/deceased) and (2) among deceased patients, the date agreement (exact and ± 30-days). Sensitivity was calculated as the percentage of deaths in the registry that were correctly identified as such by the comparator. Specificity, as the percentage of patients alive (without a death date) in the registry that were identified as non-deceased by the comparator. Date agreement analyses were restricted to include only patients who had a death date in the comparator. The 30-day date agreement was calculated as the percentage of death dates in the comparator that matched a record in registry data within a ± 30-day window.

We selected a ± 30-day window for temporal agreement analysis based on the methodology established in a prior comprehensive validation study of composite mortality in the US Flatiron Health Research Database ( n = 160,436 patients) [ 14 ]. This 30-day window accommodates: (1) potential delays in death ascertainment and documentation across data sources; (2) differences in timing conventions between EHR recording (date of death notification) and registry recording (date of vital status confirmation); and (3) practical healthcare system workflows where death may be documented in systems outside the primary EHR environment within a short time window. The 30-day standard has become established practice in oncology real-world evidence research for assessing temporal accuracy of mortality endpoints.

Results

Breast cancer

The German breast cancer cohort included 1,305 patients with a median age at diagnosis of 58 years (IQR: 48–70), with 99% being female. Age distribution was: 29% aged < 50 years, 35% aged 50–64 years, 19% aged 65–74 years, and 17% aged ≥ 75 years. Most patients presented with early-stage disease: 35% stage I and 40% at stage II. Stage III and IV disease comprised 15% and 10%, respectively, with 7% having unknown stage. Invasive ductal carcinoma was the predominant histology (80%), followed by invasive lobular carcinoma (14%). Less common histologies (inflammatory, medullary, metaplastic) each represented < 1% of cases. Tumor grading at initial diagnosis showed 64% grade 2, 29% grade 3, and 7% grade 1 tumors. Molecular subtype distribution among patients with documented biomarker status was: 67% HR+/HER2- disease, 19% had triple-negative breast cancer (HR-/HER2-), 9% HR+/HER2 + disease, 5% HR-/HER2 + disease. Biomarker data were missing for 24% of patients (Table 1 ).

Regarding treatment, 70% received systemic therapy, and 86% of metastatic patients received first-line (1 L) therapy. Among 1 L-treated patients, CDK4/6 inhibitor-based regimens were most common (35%), followed by endocrine-based therapy (24%) and chemotherapy (17%). HER2-targeted therapies were used in 12% of patients, with < 2% receiving investigational agents as part of clinical studies. Median rwOS from 1 L initiation was 26.0 months (95% CI: 22.1–28.6). Survival varied by age: patients 35–49 years had the longest median rwOS (29.8 months; 95% CI: 20.2–51.8), while those aged 18–34 years had a notably shorter survival (14.6 months; 95% CI: 10.4-not estimable) (Table 4 ).

Lung cancer

The German NSCLC cohort comprised 866 patients with a median age of 69 years (IQR: 61–76). Age groups were 31% aged 50–64 years, 35% aged 65–74 years, 30% aged ≥ 75 years, and 4% were < 50 years at diagnosis. Gender distribution was 56% male and 44% female. Smoking history was documented in 89% of patients, with 11% reporting no smoking history, and for 21% having unknown status (Table 2 ). Advanced disease was common, with 49% presenting as stage IV. Early stages comprised 12% stage I, 12% stage II, and 28% stage III, with 5% having unknown stage. Histologically, 73% had non-squamous cell carcinoma, 24% squamous cell carcinoma, and 3% not otherwise specified. Multiple primary tumors occurred in 4% of patients. Biomarker testing revealed EGFR status in 45% of patients (16% positive among tested) and PD-L1 status in 54% (66% positive among tested) (Table 2 ). Systemic therapy was administered to 70% of the cohort, with 72% of metastatic patients receiving 1 L systemic therapy.

PD-1/PD-L1-based immunotherapy without chemotherapy was the most frequent 1 L treatment (48%), followed by platinum-based chemotherapy combinations (18%), and EGFR tyrosine kinase inhibitors (8%) (Table 2 ). Other regimens included anti-VEGF-based therapies, PD-1/PD-L1 plus CTLA-4 combination therapy, single-agent chemotherapy, and other targeted or investigational agents. Median OS from 1 L initiation was 13.4 months (95% CI: 10.9–16.3). Age-stratified survival showed: 50–64 years, 17.5 months (95% CI: 14.0–22.8); 65–74 years, 14.2 months (95% CI: 9.5–20.8), and ≥ 75 years, 9.5 months (95% CI: 6.7–13.0). PD-L1 positive patients had a median rwOS of 14.9 months versus 13.0 months for PD-L1 negative patients (Table 4 ).

Colorectal cancer

The German CRC cohort included 774 patients with a median age of 67 years (IQR: 58–76). Age groups were: 32% aged 50–64 years, 31% aged 65–74 years, 28% aged ≥ 75 years, and 10% aged < 50 years, capturing both early- and late-onset disease. Gender distribution was 57% male and 43% female. Stage distribution was: 9% stage I, 22% stage II, 38% stage III, and 31% stage IV, with 5% unknown. Primary tumor location was: 40% rectum, 32% right colon, 27% left colon, and < 2% other/unknown locations (e.g., overarching primary tumors). Treatment patterns showed 90% underwent primary surgery and 28% received radiotherapy, with the majority of those receiving external beam radiation for rectal cancer. Biomarker testing rates in the total cohort were: 38% for MSI/MMR (12% MSI-H/dMMR among tested), 17% for BRAF (16% positive among tested), and 21% for KRAS / NRAS (51% positive among tested). Testing rates increased in the metastatic subset: 53% MSI/MMR, 46% BRAF , and 54% KRAS / NRAS .

Systemic therapy was administered to 68% of all patients and 75% of metastatic patients. First-line regimens were: chemotherapy-only (38%), chemotherapy with anti-VEGF (37%), and anti-EGFR based therapies (18%) (Table 3 ). Median rwOS from 1 L initiation was 21.5 months (95% CI: 18.6–23.7). Age-stratified survival showed: 50–64 years, 30.3 months (95% CI: 21.6–35.4); 65–74 years, 21.3 months (95% CI: 17.7–26.3); ≥75 years, 15.6 months (95% CI: 13.1–20.4) (Table 4 ).

Mortality data validation

Among 1,280 patients spanning 7 cancer types (breast, NSCLC, CRC, small-cell lung cancer, multiple myeloma, prostate cancer, and diffuse large B-cell lymphoma), the validation analysis comparing the Germany composite mortality variable to the German Cancer Registry demonstrated high sensitivity of 87.7%, specificity of 91.7%, ± 30-day temporal agreement of 98.8%, and exact date match of 84.8%. Out of 579 dates of death documented in the Cancer Registry, 71 were not identified in the EHR composite mortality variable. Conversely, out of 566 patients with a date of death in the EHR, 58 were not yet recorded in the Cancer Registry. Missing death of date information was not randomly distributed; patients with longer follow-up duration and more recent healthcare encounters had more complete mortality ascertainment. This pattern is consistent with an assumption of missingness at random conditional on follow-up duration and healthcare utilization, though some loss to follow-up among patients transitioning to palliative care settings outside the EHR network may have occurred. Directional analysis revealed that deaths missing from the Cancer Registry were predominantly very recent, reflecting the expected lag in registry data processing. The EHR documented deaths contemporaneous with clinical care, while the registry required time for provider reports to be submitted and integrated.

Discussion

This study provides comprehensive characterization of Flatiron Health Germany’s EHR-derived oncology data across three major cancer types, demonstrating strong alignment with epidemiological patterns. Using Gerke et al.’s structured framework for assessing RWD representativeness [ 13 ], we systematically benchmarked key cohort characteristics against national statistics, cancer registries, and published studies. This approach enables robust evaluation of external validity and generalizability for research and policy applications in German oncology.

Breast cancer representativeness

Our breast cancer cohort closely mirrors German registry data across key characteristics. The median age at diagnosis (58 years) aligns with national reports (57–62 years) [ 1 , 15 ] and histological patterns match national data: invasive ductal carcinoma (80% vs. 70–80% nationally) and invasive lobular carcinoma (14% vs. 10–15%) [ 1 , 15 , 16 ]. Molecular subtype distributions demonstrate excellent concordance, with HR+/HER2- disease predominating (67% vs. 70% nationally) and HER2-positive disease occurring in 14% of patients, consistent with national estimates of 10–15% [ 17 – 19 ]. However, our national cohort showed higher early-stage prevalence (75% stage I-II vs. 51% nationally) [ 20 , 21 ], reflecting oncology practice setting and selection towards patients requiring systemic therapy. Treatment patterns reflect contemporary German practice, with CDK4/6 inhibitor-based regimens most common (35%, followed by endocrine therapy 24%). This predominance aligns with current German treatment guidelines and clinical adoption patterns, validating the dataset’s capture of real-world therapeutic decision-making.

Lung cancer representativeness

The NSCLC cohort demonstrates consistent alignment with national patterns. Median age (69 years) matches German statistics (69–70 years) and advanced disease prevalence (77% stage III-IV closely mirrors national data (77–79%) [ 1 ]. Smoking history patterns also align well, with 89% reporting smoking history compared to 91% in national statistics [ 22 ]. Biomarker testing rates and mutational frequencies validate the dataset’s clinical relevance. EGFR positivity (16%) matches the CRISP registry findings (15.1% in non-squamous carcinoma) [ 23 ], while PD-L1 positivity (66%) aligns with published German data (62%) [ 22 ]. Treatment patterns, with PD-1/PD-L1 immunotherapy predominating (48%), reflect contemporary German practice and guideline adherence.

Colorectal cancer representativeness

The CRC cohort shows strong demographic concordance with national data. Gender distribution (57% male) closely matches registry patterns (54.5% male) [ 1 , 24 ], and median age (67 years) approximates national medians (72.2 years) [ 1 , 24 ]. Importantly, early-onset disease (10% under 50 years) aligns with national estimates (5–10%) [ 1 , 24 ], capturing this emerging clinical concern. Disease stage distribution (38% stage III, 31% stage IV) parallels national data (27.5% and 25%, respectively) [ 1 ], with minor differences potentially reflecting oncology practice settings. The anatomical distribution in our study (40% rectal cancers) aligns with German registry data indicating approximately one-third of colorectal cancers arise in the rectum [ 24 ]. The high surgical intervention rate (90%) aligns with German standard-of-care [ 25 – 27 ], while biomarker testing patterns and molecular characteristics (12% MSI-H/dMMR) match expected population frequencies (5–15%; 23.7%) [ 28 , 29 ]. The treatment patterns observed, with chemotherapy-only regimens being most common (38%) in 1 L, align with guideline-recommended standard treatment approaches and population-based estimates [ 25 , 30 ]. The median rwOS of 21.5 months from 1 L observed in our cohort, with the expected age-related differences in survival outcomes, is comparable to contemporary survival data in Germany, where age has been consistently identified as a significant prognostic factor [ 1 , 26 , 30 , 31 ]. These similarities further validate the representative nature of the Flatiron Health Germany colorectal cancer cohort for ongoing and future research applications.

Mortality endpoint validation

A critical strength of this study is the validation of our composite mortality endpoint against German Cancer Registry data. Demonstrating sensitivity of 87.7%, specificity of 91.7%, and ± 30-day temporal agreement of 98.8%, validates the reliability of this German mortality data by benchmarking against the best patient-level reference standard available. The occurrence of missing deaths in the Germany Flatiron Health Research Database was of similar magnitude to the occurrence of missing deaths in the gold standard German Cancer Registry. These findings align closely with a 2021 validation study of the United States Flatiron Health oncology database ( N = 160,436) composite mortality variable, which demonstrated sensitivity of 83.9%−91.5%, specificity of 93.5%−99.7%, and ± 30-day temporal agreement of 96.3–97.9% [ 14 ]. The ± 30-day temporal agreement of 98.8% in our study is comparable to the US validation study (96.3%−97.9%), suggesting that this approach provides reliable temporal accuracy for time-to-event analyses across different healthcare systems and regulatory environments.

The consistency between Germany and US validation results demonstrates the robustness of Flatiron Health’s composite mortality methodology across different healthcare systems and regulatory environments. The EHR and registry sources have complementary operational characteristics: the registry provides validated historical completeness, while the EHR enables real-time mortality documentation. This distinction is valuable for prospectively maintained research databases requiring current vital status information between registry refreshes. The high temporal accuracy is particularly important for time-to-event analyses, as even small discrepancies in death dates can significantly impact survival estimates and comparative effectiveness research. Our composite approach, integrating EHR data with external mortality sources, represents a validated and robust methodology for mortality ascertainment in Germany. The comparable performance across German and US healthcare contexts supports the generalizability of this approach for international oncology research and regulatory submissions.

Clinical and research implications

Collectively, these findings establish the Flatiron Health Germany dataset as a rigorously characterized, validated source of real-world evidence that demonstrates the potential to support oncology research, health technology assessment, and regulatory decision-making in the German healthcare context. The demonstrated representativeness across multiple cancer types, combined with robust mortality endpoint validation, positions this dataset as a valuable resource that aims to advance cancer care through evidence-based research and policy development, while acknowledging inherent limitations in any real-world data source derived from clinical practice.

Limitations and considerations

Several limitations warrant consideration. First, the oncology practice-based sampling may underrepresent patients managed exclusively in surgical or primary care settings, particularly those with early-stage disease not requiring systemic therapy. Hospitals with DKG-certified centres will soon be included in the dataset. Second, missing biomarker data (approximately 19–24% across cohorts) may introduce bias in subtype-specific analyses. Third, the dataset lacks comprehensive comorbidity, socioeconomic, and patient-reported outcome data, limiting contextualization of treatment decisions and outcomes.

A potential limitation is that mortality ascertainment completeness may differ by patient subgroup; however, the high sensitivity and specificity achieved in our validation analysis, and the comparability to US validation results obtained across diverse care settings, suggest that this approach provides reliable vital status classification for the populations and time periods studied.Additionally, variable follow-up durations and potential loss to follow-up may influence survival estimates, particularly among patients with infrequent healthcare encounters. The absence of a common data model across compared EHR data sources may introduce measurement error or misclassification, though our systematic approach attempts to mitigate concerns regarding selection bias and comparability.

Unfortunately, socioeconomic data are rarely systematically documented in oncology EHRs, as these are not required for clinical care delivery or billing purposes. Variables such as income, education level, or occupation are typically not collected during routine oncology visits. While some proxies (e.g., insurance type, ZIP code) may be available in US-based EHR systems, the Germany Flatiron Health Research Database does not systematically capture insurance status or residential location data that could serve as socioeconomic proxies, partly due to stricter European data privacy regulations under GDPR. Unlike administrative claims databases where comorbidities are systematically coded for reimbursement, EHR-derived data capture comorbidities only when documented by clinicians during encounters. Comorbidities unrelated to cancer care may not be recorded during routine oncology visits if they do not directly impact treatment decisions. Comorbidities unrelated to cancer management may therefore be under-captured.

Conclusions

This comprehensive validation study demonstrates that Flatiron Health Germany EHR-derived cohorts are highly representative of national cancer populations across three major tumor types. Our findings reveal demographically and clinically diverse patient populations with age distributions, disease staging, histologic subtypes, and molecular profiles that closely align with German national cancer registry data. The observed treatment patterns reflect contemporary standard-of-care and routine clinical practice in Germany, providing crucial context for real-world evidence interpretation.

Critically, our composite mortality variable validation establishes the dataset’s reliability for survival analyses. This validation performance aligns with international benchmarks, as demonstrated by comparable results to US Flatiron Health validation studies, confirming the robustness of our methodology across different healthcare systems. The consistency between overall survival patterns and expected clinical outcomes across age groups and molecular subtypes reinforces the dataset’s clinical credibility.

Collectively, these findings establish the Flatiron Health Germany dataset as a validated, reliable source of real-world evidence for oncology research, health technology assessment, and regulatory decision-making in the German healthcare context. The demonstrated representativeness across multiple cancer types, combined with robust mortality endpoint validation, positions this dataset as a valuable resource for advancing cancer care through evidence-based research and policy development.

Acknowledgements

This study was sponsored by Flatiron Health, Inc., which is an independent member of the Roche Group. The authors would like to thank the patients whose deidentified data made this research possible, as well as the clinicians and staff at participating oncology practices across Germany who contributed to the Flatiron Health Research Database. We are grateful to the Flatiron Health data curation and research operations teams for their support in data abstraction and quality assurance. Editorial support was provided by Darren Johnson, in accordance with Good Publication Practice (GPP) guidelines, programming support was provided by Wiliam Xue. The authors also acknowledge the contributions of colleagues at Flatiron Health Germany and Flatiron Health, Inc. for their scientific input and operational support throughout the study.

Abbreviations

B-Raf proto-oncogene (a gene that can cause cancer when mutated)

Colorectal cancer

Cytotoxic T-lymphocyte-associated protein 4 (an immune checkpoint protein)

German Cancer Society-certified

Epidermal growth factor receptor (a protein involved in cell growth)

Electronic health record

Hormone receptor/human epidermal growth factor receptor 2 (breast cancer biomarkers)

Interquartile range (a statistical measure)

Microsatellite instability/mismatch repair (genetic markers for cancer)

Neuroblastoma RAS viral oncogene homolog/Kirsten rat sarcoma viral oncogene

Non-small cell lung cancer

Programmed cell death protein 1/programmed death-ligand 1 (immune checkpoint proteins)

Real-world data

Real-world evidence

real-world overall survival

Vascular endothelial growth factor (a protein that stimulates blood vessel formation)

German Centre for Cancer Registry Data

Authors’ contributions

Conceptualization: BA, QZ, NB, SBW, EH; Methodology: BA, QZ, NB, SBW, EH; Validation: BA, QZ, NB, SBW, CB, AS, SS; Investigation: BA, QZ, NB, SBW, CB, AS, SS; Writing – Original Draft: BA, SBW, CB, AS; Writing – Review & Editing: BA, QZ, NB, SBW, CB, EH, AS, SS; Visualization: BA, SBW; Supervision: BA.

Funding

This study was sponsored by Flatiron Health, Inc., which is an independent member of the Roche Group.

Authors’ Contributions : Conceptualization: BA, QZ, NB, SBW, EH; Methodology: BA, QZ, NB, SBW, EH; Validation: BA, QZ, NB, SBW, CB, AS, SS; Investigation: BA, QZ, NB, SBW, CB, AS, SS; Writing – Original Draft: BA, SBW, CB, AS; Writing – Review & Editing: BA, QZ, NB, SBW, CB, EH, AS, SS; Visualization: BA, SBW; Supervision: BA.

Data availability

The data that support the findings of this study are available from Flatiron Health Germany but restrictions apply to the availability of these data, which were used under license for the current study, and so are not publicly available. Data are however available from the authors upon reasonable request and with permission by contacting PublicationsDataAccess@flatiron.com.

Declarations

Ethics approval and consent to participate

The (non-interventional) study was conducted using anonymized data only. The researchers working on the analysis do not have access to any identifiable data. The datasets were assembled and anonymized on the basis of patient consent and in line with relevant exceptions under applicable national and European regulations (Art. 9 para. 2 (j) GDPR, Art. 27 German Federal Data Protection Act and Art. 6 German Health Data Use Act).

Therefore no ethics approval is required according to Art. 15 of the German Professional Code for Physicians and Article 23 of the Declaration of Helsinki. Consent and ethics approval requirement under clinical trial regulation in conjunction with Art. 40, 41 German Medicines Act (Arzneimittelgesetz), Art. 62 Medical Device Regulation in conjunction with Art. 31, 31a Medical-Devices Implementation Act (Medizinprodukterecht- Durchführungsgesetz) and Art. 34 para. 2 No. 2 Radiation Protection Act (Strahlenschutz- Gesetz) are also not applicable.

Consent for publication

Not applicable.

Competing interests

All authors report current employment with Flatiron Health, Inc., Flatiron Health UK, or Flatiron Health Germany—independent members of the Roche group. BA, SBW, EH, SS, QZ, LK, RS, AS, and CB report stock ownership in Roche.

Footnotes

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

References

1. Ronckers C, Spix C, Trübenbach C, Katalinic A, Christ M, Cicero A, et al. Krebs in Deutschland für 2019/2020. 2023. https://edoc.rki.de/handle/176904/11438 .

2. Lerro CC, Bradley MC, Forshee RA, Rivera DR. The bar is high: evaluating fit-for-use oncology real-world data for regulatory decision making. JCO Clin Cancer Inform. 2024;8:e2300261. 10.1200/cci.23.00261.

3. Arlett P, Umuhire D, Verpillat P, Foggi P, Liminga UW, Sepodes B, et al. Clinical Evidence 2030. Clin Pharmacol Ther. 2025;117:884–6. 10.1002/cpt.3596.

4. Hernandez RK, Critchlow CW, Dreyer N, Lash TL, Reynolds RF, Sørensen HT, et al. Advancing principled pharmacoepidemiologic research to support regulatory and healthcare decision making: the era of real-world evidence. Clin Pharmacol Ther. 2025;117:927–37. 10.1002/cpt.3563.

5. Moser K, Massag J, Frese T, Mikolajczyk R, Christoph J, Pushpa J, et al. German primary care data collection projects: a scoping review. BMJ Open. 2024;14:e074566. 10.1136/bmjopen-2023-074566.

6. Molnár-Gábor F. Germany: a fair balance between scientific freedom and data subjects’ rights? Hum Genet. 2018;137:619–26. 10.1007/s00439-018-1912-1.

7. Katalinic A, Halber M, Meyer M, Pflüger M, Eberle A, Nennecke A, et al. Population-based clinical cancer registration in Germany. Cancers. 2023;15:3934. 10.3390/cancers15153934.

8. Pohlmann S, Kunz A, Ose D, Winkler EC, Brandner A, Poss-Doering R, et al. Digitalizing health services by implementing a personal electronic health record in Germany: qualitative analysis of fundamental prerequisites from the perspective of selected experts. J Med Internet Res. 2020;22:e15102. 10.2196/15102.

9. Fleurence RL, Kent S, Adamson B, Tcheng J, Balicer R, Ross JS, et al. Assessing real-world data from electronic health records for health technology assessment: the SUITABILITY checklist: a good practices report of an ISPOR task force. Value Health. 2024;27:692–701. 10.1016/j.jval.2024.01.019.

10. Callahan A, Shah NH, Chen JH. Research and reporting considerations for observational studies using electronic health record data. Ann Intern Med. 2020;172(11_Supplement):S79-84. 10.7326/m19-0873.

11. Giusti F, Martos C, Trama A, Bettio M, Sanvisens A, Audisio R, et al. Cancer treatment data available in European cancer registries: where are we and where are we going? Front Oncol. 2023;13:1109978. 10.3389/fonc.2023.1109978.

12. Adamson B, Horne E, Xu C, Samani A, Buhl C, Mpofu P, et al. Characterisation of oncology EHR-derived real-world data in the UK, Germany, and Japan. ESMO Real World Data and Digital Oncology. 2025;7:100113. 10.1016/j.esmorw.2025.100113.

13. Gerke T, Ackerman B, Long L, Baxi S, Miksad R, Adamson B, et al. Representativeness of real world data: A framework for assessing oncology EHR-derived data. SSRN Electron J. 2024. 10.2139/ssrn.4727892.

14. Zhang Q, Gossai A, Monroe S, Nussbaum NC, Parrinello CM. Validation analysis of a composite real-world mortality endpoint for patients with cancer in the united States. Health Serv Res. 2021;56:1281–7. 10.1111/1475-6773.13669.

15. Gremke N, Wagner U, Kalder M, Kostev K. Changes in the incidence of early-onset breast cancer in Germany between 2010 and 2022. Breast Cancer Res Treat. 2023;202:167–72. 10.1007/s10549-023-07048-1.

16. Hübner J, Katalinic A, Waldmann A, Kraywinkel K. Long-term incidence and mortality trends for breast cancer in Germany. Geburtshilfe Frauenheilkd. 2020;80:611–8. 10.1055/a-1160-5569.

17. Rüschoff J, Lebeau A, Kreipe H, Sinn P, Gerharz CD, Koch W, et al. Assessing HER2 testing quality in breast cancer: variables that influence HER2 positivity rate from a large, multicenter, observational study in Germany. Mod Pathol. 2017;30:217–26. 10.1038/modpathol.2016.164.

18. Dannehl D, von Au A, Engler T, Volmer LL, Gutsfeld R, Englisch JF, et al. Cancers. 2024;16:1490. Implementation and Evaluation of a Breast Cancer Disease Model Using Real-World Claims Data in Germany from 2010 to 2020. 10.3390/cancers16081490.

19. Rüschoff J, Lebeau A, Sinn P, Schildhaus H-U, Decker T, Ammann J, et al. Statistical modelling of HER2-positivity in breast cancer: final analyses from two large, multicentre, non-interventional studies in Germany. Breast. 2020;49:246–53. 10.1016/j.breast.2019.12.005.

20. Katalinic A, Eisemann N, Kraywinkel K, Noftz MR, Hübner J. Breast cancer incidence and mortality before and after implementation of the German mammography screening program. Int J Cancer. 2020;147:709–18. 10.1002/ijc.32767.

21. Jansen L, Holleczek B, Kraywinkel K, Weberpals J, Schröder CC, Eberle A, et al. Divergent patterns and trends in breast cancer incidence, mortality and survival among older women in Germany and the united States. Cancers. 2020;12:2419. 10.3390/cancers12092419.

22. Stuschke M, Eberhardt W, Passlick B, Groeschel A, Christopoulos P, Reck M, et al. EP05.01-030 crisp: first real-world evidence of NSCLC stage I, II and III in Germany - AIO-TRK-0315. J Thorac Oncol. 2022;17:S280. 10.1016/j.jtho.2022.07.477.

23. Griesinger F, Eberhardt W, Nusch A, Reiser M, Zahn M-O, Maintz C, et al. Biomarker testing in non-small cell lung cancer in routine care: analysis of the first 3,717 patients in the German prospective, observational, nation-wide CRISP Registry (AIO-TRK-0315). Lung Cancer. 2021;152:174–84. 10.1016/j.lungcan.2020.10.012.

24. Waldmann A, Borchers P, Katalinic A. Temporal trends in age- and stage-specific incidence of colorectal adenocarcinomas in Germany. BMC Cancer. 2023;23:1180. 10.1186/s12885-023-11660-1.

25. der WMF eVA, DK eV, Krebshilfe D. Kolorektales Karzinom. Leitlinieprogramm onkologie. 2019. https://www.leitlinienprogramm-onkologie.de/leitlinien/kolorektales-karzinom/ . Accessed 22 June 2025.

26. Rudolph C, Engholm G, Pritzkuleit R, Storm HH, Katalinic A. Colorectal cancer survival in German–Danish border regions—a registry-based cohort study. Cancers. 2023;15:4474. 10.3390/cancers15184474.

27. Hofheinz RD, Arnold D, Borner M, Eisterer W, Faber G, Folprecht G, et al. Kolonkarzinom. Onkopedia. https://www.onkopedia.com/de/onkopedia/guidelines/kolonkarzinom/@@guideline/html/index.html . Accessed 22 June 2025.

28. Mulet-Margalef N, Linares J, Badia-Ramentol J, Jimeno M, Monte CS, Mozo JLM, et al. Challenges and therapeutic opportunities in the dMMR/MSI-H colorectal cancer landscape. Cancers. 2023;15:1022. 10.3390/cancers15041022.

29. Mei W-J, Mi M, Qian J, Xiao N, Yuan Y, Ding P-R. Clinicopathological characteristics of high microsatellite instability/mismatch repair-deficient colorectal cancer: a narrative review. Front Immunol. 2022;13:1019582. 10.3389/fimmu.2022.1019582.

30. Marschner N, Seufferlein T, Potthoff K, Zahn M-O, Uhlig J, Dörfel S, et al. Changes in survival of > 3800 patients with metastatic colorectal cancer in germany: results from a 16-year prospective longitudinal real-world data analysis. ESMO Real World Data Digit Oncol. 2024;4:100040. 10.1016/j.esmorw.2024.100040.

31. Kostev K, Krieg S, Krieg A, Luedde T, Loosen SH, Roderburg C. In-hospital mortality and associated factors among colorectal cancer patients in Germany. Cancers. 2024;16:1219. 10.3390/cancers16061219.

Associated Data

Data Availability Statement

The data that support the findings of this study are available from Flatiron Health Germany but restrictions apply to the availability of these data, which were used under license for the current study, and so are not publicly available. Data are however available from the authors upon reasonable request and with permission by contacting PublicationsDataAccess@flatiron.com.