High Concordance of Different Assays in the Determination of Homologous Recombination Deficiency–Associated Genomic Instability in Ovarian Cancer

Abstract

PURPOSE

Poly(ADP-ribose) polymerase inhibitors (PARPi) have shown promising clinical results in the treatment of ovarian cancer. Analysis of biomarker subgroups consistently revealed higher benefits for patients with homologous recombination deficiency (HRD). The test that is most often used for the detection of HRD in clinical studies is the Myriad myChoice assay. However, other assays can also be used to assess biomarkers, which are indicative of HRD, genomic instability (GI), and BRCA1/2 mutation status. Many of these assays have high potential to be broadly applied in clinical routine diagnostics in a time-effective decentralized manner. Here, we compare the performance of a multitude of alternative assays in comparison with Myriad myChoice in high-grade serous ovarian cancer (HGSOC).

METHODS

DNA from HGSOC samples was extracted from formalin-fixed paraffin-embedded tissue blocks of cases previously run with the Myriad myChoice assay, and GI was measured by multiple molecular assays (CytoSNP, AmoyDx, Illumina TSO500 HRD, OncoScan, NOGGO GISv1, QIAseq HRD Panel and whole genome sequencing), applying different bioinformatics algorithms.

RESULTS

Application of different assays to assess GI, including Myriad myChoice, revealed high concordance of the generated scores ranging from very substantial to nearly perfect fit, depending on the assay and bioinformatics pipelines applied. Interlaboratory comparison of assays also showed high concordance of GI scores.

CONCLUSION

Assays for GI assessment not only show a high concordance with each other but also in correlation with Myriad myChoice. Thus, almost all of the assays included here can be used effectively to assess HRD-associated GI in the clinical setting. This is important as PARPi treatment on the basis of these tests is compliant with European Medicines Agency approvals, which are methodologically not test-bound.

Validating diverse HRD tests for ovarian cancer treatment, beyond Myriad myChoice, expands diagnostic options in Europe.

INTRODUCTION

In the past years, poly(ADP-ribose) polymerase inhibitors (PARPi) have become a standard maintenance treatment option as first-line treatment for patients with recurrent platinum-sensitive ovarian cancer. Because these drugs have been perceived to work specifically in a synthetic lethal fashion in tumors affected by homologous recombination deficiency (HRD), several biomarker test strategies to identify this phenotype have been developed.^1-3

CONTEXT

• Key Objective

• In the therapy of high-grade ovarian cancer, the presence of homologous recombination deficiency (HRD) positivity or genomic instability (GI) more and more becomes a criteria for the treatment with poly(ADP-ribose) polymerase inhibitors. Myriad myChoice, the most commonly used diagnostic test, is only centralized available, which is not standard of care in Europe. Therefore, alternative tests are urgently needed.

• Knowledge Generated

• We performed an analytical bridging study with the goal to technically validate and harmonize different independent and decentral HRD test solutions—to our knowledge, the first one comparing a multitude of assays. Comparison of different technologies/methods for HRD testing shows high concordance to the Myriad myChoice assay.

• Relevance

• Besides Myriad myChoice, a multitude of assays can be safely used to assess HRD associated GI in the clinical setting. The alternative assays tested can be implemented in routine clinical practice, allowing for widespread adoption and accessibility.

The potentially easiest way to identify tumors with HRD has been to screen for alterations of homologous recombination repair (HRR) genes, first and foremost BRCA1/2 mutations. Indeed, early data suggested that the presence of BRCA1/2 mutations could be used as a selector for therapy response to PARPi in recurrent ovarian cancer.^4-6 Thus, initially, treatment in such a clinical scenario was limited to patients harboring a BRCA1/2 alteration.⁷

Although most of the early studies in ovarian cancer focused on BRCA1/2 mutations as selection criteria, it has been known for quite some time that HRD is not only induced by the deleterious BRCA1/2 mutations but also by mutation of other HRR genes and a variety of nonmutational processes, including copy number losses and epigenetic silencing. Thus, alternative biomarker strategies were developed focusing on HRD. Efforts have been focused on a variety of different factors such as specific HRD mutational signatures or expression profiles⁸ but also analyzed specific genomic structural alterations termed genomic scar, which includes loss of heterozygosity, telomeric allelic imbalance, and large-scale state transitions.⁹ The genomic scar criteria are the basis for the most broadly used commercial HRD assay applied in the majority of ongoing clinical studies, Myriad myChoice. Importantly, in Europe, PARPi treatment is not assay-dependent in contrast to the United States where Myriad myChoice is approved by the US Food and Drug Administration (FDA).

Recently, several phase III clinical trials with olaparib, niraparib, and veliparib in the first-line setting of patients with advanced ovarian cancer showed impressive results with a significant improvement of progression-free survival^1,3,10,11 and for olaparib even in overall survival.¹² Most of these studies included not only standalone BRCA1/2 testing but also genomic instability (GI) testing via Myriad myChoice. Biomarker subgroup analysis consistently showed benefit not only for patients with a BRCA1/2 mutation but also for patients with HRD-positive, BRCA1/2wt tumors, which exhibited evidence of HRD, as determined by high Myriad myChoice scores.

Thus, several companies have already or will likely try to obtain a biomarker-tailored European Medicines Agency (EMA) label for first-line ovarian cancer treatment with PARPi. Olaparib is already approved for maintenance therapy in HRD-positive advanced ovarian cancer in combination with bevacizumab.³ If EMA grants a drug label on the basis of HRD testing, this will result in a greatly increased demand for HRD testing that may not be met in most European countries, as commercial providers such as Myriad are not well established. Furthermore, the decentralized development and use of molecular pathological approaches is standard of care in European routine diagnostic laboratories. Thus, to assure optimal patient diagnostics and treatment in Europe—alternative and valid test strategies are urgently needed to identify patients who might benefit from PARPi therapy. Furthermore, the development and implementation of HRD testing strategies by academic centers will contribute to optimize the read-out of HRD status from clinical samples potentially beyond the performance of currently available commercial tests.

To address this, we performed an analytical bridging study with the goal to technically validate and harmonize different independent and decentral HRD test solutions.

The obtained HRD results should be aligned with the Myriad myChoice scores/categories to ensure correct future patient stratification when using HRD assays other than Myriad myChoice for HRD assessment.

Such test strategies can potentially be transferred to other entities/clinical scenarios because—once developed and validated—all preanalytical data on test performance and quality assurance schemes for these assays will be publicly available. The immediate clinical evaluation of these assays will circumvent real-world drug implementation problems that will likely occur in Europe if PARPi are approved in a scenario where the only validated HRD test solution is offered in a central laboratory.

METHODS

Study Setup

In this German national study, seven different academic institutions participated to validate the use of different assays for the determination of HRD status in comparison with the Myriad myChoice assay. For this purpose, seven different assays (single nucleotide polymorphism [SNP] array or NGS-based) were used at the participating institutions. To perform an interlaboratory comparison, each assay was evaluated independently at two different sites; therefore, DNA for 14 assays had to be provided, with varying input amounts of 40-200 ng per assay. For an overview of the study setup, see the Data Supplement (Fig S1).

The assays used were either array-based (CytoSNP 850K, Illumina, San Diego, CA and OncoScan CNV array, Thermo Fisher Scientific, Waltham, MA), panel-based (AmoyDx HRD Focus Panel, AmoyDx, Xiamen, China; NOGGO GISv1 [custom-designed], Agilent, Santa Clara, CA; TSO500 HRD, Illumina Inc, San Diego, CA; QIAseq HRD Panel, Qiagen, Hilden, Germany), and whole-genome sequencing (WGS; TrueQuant DNA Library Preparation Kit, GenXpro, Frankfurt, Germany). TSO500 HRD (2) was carried out in equal parts by sites 1 and 3.

The definition of GI status and HRD score is shown in Table 1. For detailed description of the assays and bioinformatics used, see the Data Supplement.

TABLE 1.

Overview of the Applied GI Cutoffs Per Assay As Determined by the Manufacturer or Literature

Assay	HRD-Score Cut-off	Detection of
Myriad myChoice	≥42	LSTs + TAI + LOH + BRCA
OncoScan	≥42	LSTs + TAI + LOH
nLST + OncoScan	≥15	nLST
Infinium CytoSNP 850K assay	≥42	LSTs + TAI + LOH
AmoyDx	≥50	Proprietary algorithm (version 3) + BRCA
NOGGO GISv1 (Agilent)	≥83	Custom developed workflow
TSO500 HRD	≥42	LSTs + TAI + LOH + BRCA+
QIAseq HRD	≥56	Proprietary algorithm + BRCA
(Lowpass) WGS	≥18 (borderline: 15-18)	LGAs

Abbreviations: GI, genomic instability; HRD, homologous recombination deficiency; LGA, large genomic alteration; LST, large scale state transition; nLST, normalized large scale state transition; TAI, telomeric allelic imbalance; LOH, Loss of heterozygosity; WGS, whole-genome sequencing.

Tissue Evaluation and Workup

Tumor tissue samples were obtained from the Tumor Bank Ovarian Cancer and the Institute of Pathology at Charité Medical University, Berlin, Germany. All patients provided informed consent from 2001 to 2012 to participate in the biobank tumor repository. The protocol was approved by the Charité Institutional Review Board (EK207/2003).

Only patients with high-grade serous ovarian cancer, with available paraffin-embedded tissue samples, and previously tested with myChoice from Myriad have been selected for this study.

All patients underwent surgery at first diagnosis and on subsequent relapse. The treatment decision toward tumor debulking was the physician's choice and not influenced by participation in this study.

Germline testing was not available in these patients; therefore, the mutations described here might be either of somatic or germline origin. The clinical data of the patients were presented in our previous study.¹⁴

Furthermore, 25 samples from the OPINION study were kindly provided by AstraZeneca and included in this study.¹⁵ Owing to the limited amount of available DNA, four assays were selected (AmoyDx, CytoSNP 850K, NOGGO GISv1, and WGS) and applied only once. Ten of these samples were analyzed as replicates with the same DNA, but different aliquots, applying only WGS.

Statistics

Receiver operating characteristic (ROC) curves were calculated to assess the ability of each assay to replicate the HRD status provided by Myriad and to confirm the predefined cutoffs for HRD status, and the area under the curve and its 95% CI were calculated. Cohen's Kappa was calculated for specific cutoffs. To assess the agreement of the continuous Myriad scores and the respective assays, Pearson correlation coefficient, the concordance correlation coefficient, scatterplots, and Bland-Altman plots were used. All statistical analysis was performed using the software R version 4.1.2¹⁶ and particularly the two packages pROC¹⁷ and blandr.¹⁸

RESULTS

Performance of the Assays: Determination of HRD Status

In total, 117 samples (including 10 replicates) were available for this study. Data were generated using seven different workflows on seven different sites. Each assay was analyzed twice on two different sites with the same samples. For detailed study setup, see the Data Supplement (Fig S1).

For 10 samples of the total 117, the Myriad myChoice assay failed to generate a HRD score. As most of the investigational assays were able to determine a HRD score for these samples, they were not excluded from the further evaluation. Of the analyzed samples, for 59 samples, a full concordance (HRD positive/negative; 100%) with Myriad myChoice could be achieved, and for three of those samples without known Myriad myChoice score, all evaluated assays showed a 100% concordance; therefore, these were also accounted as true positive or negative, respectively. In the remaining 58 samples (including seven samples without Myriad myChoice score) for 25 samples the determination of HRD positive or negative was in concordance between the tests except for one of the assays. Finally, for 33 samples, the differences in comparison with Myriad myChoice varied between two and all tested assays. In case of five of these samples (SP31, SP62, SP89, SP93, and SP101), there was no concordance (SP62, SP89) or only one of the investigational assays showed concordance with Myriad myChoice (SP31, SP93, SP101; Data Supplement, Figs S2 and S3). Therefore, it might be assumed that in these samples, the Myriad myChoice HRD score might be false positive or negative, respectively.

In total, we showed a median concordance of 94% (range, 0%-100%) between Myriad myChoice and the alternative assays, or if Myriad myChoice was not available, in between the tested assays. A closer look to the samples with no concordance revealed that most of these tissues showed a high degradation rate (shown by quality checks of the corresponding assays) and need for repetition of the assays because of quality issues (Data Supplement, Table S1).

The amount of available tissue and, therefore, extracted DNA was limited, and not all samples could be tested with all planned assays. However, in at least one of the participating sites for each assay, the complete planned number of samples/DNAs could be analyzed, except for the QIAseq HRD assay. Because of the availability, this assay was included at a later time point of the study, and only a limited amount of DNA was left for all of the samples. Therefore, differences in analyzed sample number per assay have nothing to do with the performance or reliability of the assay itself but almost exclusively with the limited availability of the DNA. Data Supplement (Table S2) shows which samples were processed at each of the participating sites with the different assays and depicts the failure rate of each assay.

We further evaluated the results of WGS analysis of 10 replicates in our study. Here, the replicate samples also revealed a high concordance between Myriad myChoice and with each other (Data Supplement, Fig S3). Three samples showed no or low levels of concordance to Myriad myChoice but nearly the same HRD score in both replicates analyzed by WGS. At least for sample pair SP89/SP110, a retesting of the Myriad myChoice assay would be recommended since it showed high concordant results in all other tested assays. The reason for discordance of sample pair SP83/SP108 might be the borderline scores in the WGS replicates which are close to the cutoff of 15 LGAs. The authors of the shallowHRD tool¹⁹ recommended counting scores between 15 and 19 LGAs as borderline scores.

Interassay Comparisons, Including Myriad myChoice

Performance of the Assays: Correlation of GI Score

To further investigate assay performance characteristics, we also compared raw GI values between the assays—including Myriad myChoice—revealing a high overall correlation with Pearson correlation ranging between 0.6 and 0.9 for almost all comparisons (Fig 1) and as for the comparison with the Myriad myChoice assay. Only very few assays performed at single sites showed a lower value. The results of this comparison are depicted in Figures 1A (heatmap) and 1B (scatter plot). The two assays with a slightly lower performance at one partner site are either still under development—QIAseq HRD Panel—or not originally developed for HRD testing—CytoSNP. Nevertheless, with an r value of 0.81 (QIAseq) and 0.84 (CytoSNP) at the second partner site compared with Myriad myChoice, these assays show a strong potential to correctly determine HRD.

FIG 1.

(A) Correlation between Myriad MyChoice and seven alternative assays. GI scores were determined at two different sites by AmoyDx GSS v1,1,1 (181 analyses), CytoSNP (167 analyses), OncoScan (151 analyses, two different algorithms), Illumina TSO500 HRD (144 analyses), Qiagen QIAseq HRD panel (117 analyses), WGS assay (147 analyses), Agilent NOGGO GISv1 (187 analyses). Correlation between the assays and Myriad MyChoice is depicted in a correlation heatmap. Numbers in brackets indicate performing partner sites 1 or 2. (B) Correlation scatter plots. GI scores were determined, and pairwise correlation is depicted in correlation scatter plots. The Pearson R-value, the concordance correlation coefficient (CCC) and number of the tested samples (n) are shown. GI, genomic instability; HRD, homologous recombination deficiency; WGS, whole-genome sequencing.

Overlap of Positivity/Negativity Groups Between the Tested Assays and Myriad myChoice Data

Kappa values for groups generated by the standard cutoffs for GI (defined by the vendor or by previous works) were high compared to Myriad myChoice with kappa being between 0.6 and 0.9 except in two cases (Table 2). Kappa values were even higher when the optimized Youden's cutoffs were used. After this procedure, most kappa values were in the range between 0.7 and almost 1. Furthermore, in Figure 2 the ROC values also demonstrate high sensitivity and specificity reflected also by the numerical values shown in the Data Supplement (Table S3). This can further be observed in the Bland-Altman comparisons shown in Figure 3.

TABLE 2.

Cohen's Kappa Values According to Predefined Cutoffs and Youden's Cutoffs

Assay	Kappa with Prespecfied cutoff_lower	Kappa with Prespecfied cutoff_estimate	Kappa with Prespecfied cutoff_upper	Kappa with Youden's cutoff_lower	Kappa with Youden's cutoff_estimate	Kappa with Youden's cutoff_upper
WGS (1)	0.698	0.856	1.000	0.649	0.817	0.986
NOGGO GISv1 (1)	0.722	0.850	0.977	0.722	0.850	0.977
TSO500 HRD (2)	0.690	0.823	0.956	0.810	0.909	1.000
NOGGO GISv1 (2)	0.688	0.808	0.927	0.717	0.830	0.943
WGS (2)	0.651	0.784	0.917	0.659	0.789	0.920
TSO500 HRD (1)	0.632	0.781	0.930	0.664	0.809	0.953
AmoyDx GSS (2)	0.593	0.753	0.914	0.636	0.786	0.936
nLST + OncoScan (2)	0.571	0.732	0.894	0.613	0.764	0.915
CytoSNP (1)	0.566	0.715	0.864	0.571	0.718	0.866
nLST + OncoScan (1)	0.513	0.695	0.878	0.590	0.758	0.927
AmoyDx GSS (1)	0.515	0.668	0.821	0.548	0.694	0.840
OncoScan (1)	0.450	0.654	0.857	0.617	0.782	0.947
OncoScan (2)	0.453	0.639	0.824	0.481	0.656	0.831
QIAseq HRD panel (2)	0.396	0.620	0.844	0.456	0.667	0.878
CytoSNP (2)	0.226	0.426	0.625	0.294	0.474	0.654
QIAseq HRD panel (1)	0.029	0.311	0.593	0.125	0.319	0.512

NOTE. Comparison of GI group, as calculated with the predefined cutoffs but also when working with Youden's cutoff. All assays were compared with Myriad myChoice. Lower means lower limit of 95% CI, upper means upper limit of 95% CI, and estimate means the actual estimate that we have reported previously.

Abbreviations: GI, genomic instability; HRD, homologous recombination deficiency; WGS, whole genome sequencing.

FIG 2.

AUC graphs depicting sensitivity (Se) and specificity (Sp) for all assay measurements compared with the Myriad myChoice standard. Cutoffs for calling a case positive given by the assay developer or for the purely academical assays as determined in previous experiments are given as black crosses. Optimized cutoffs as calculated with Youden's method are given in red. AUC, area under the curve.

FIG 3.

Bland-Altman-plots. The Bland-Altman plots show the agreement of each assay with Myriad myChoice. The difference between the assay and Myriad myChoice for each individual measurement can be read from the y-axis, and on the x-axis the score of the corresponding assay is depicted. The solid black line indicates no difference, the middle dashed line shows the mean difference, and the other two dashed lines correspond to mean difference ±1.96 standard deviations. Only assays with the same cutoff as Myriad myChoice (42) are shown.

Interlaboratory Comparisons

All participating molecular diagnostic laboratories included in the project were highly qualified and accredited or under process of accreditation (DIN EN ISO/IEC 17020) and at least one of the sites has experience with the assays assigned to them. To perform interlaboratory comparisons, all assays were run twice, and each of these duplicates was run in different laboratories (Data Supplement, Fig S1). For this analysis, n-numbers of the assays compared correlate with numbers of samples with valid GI scores for the relevant assays. Myriad scores were neglected. Because of the limited amount of available DNA from the samples of the OPINION study, these samples were analyzed only once applying the AmoyDx assay, the CytoSNP 850K, the NOGGO GISv1 assay, and WGS. Furthermore, the 10 replicates with samples from the OPINION study were tested only applying WGS.

Continuous Raw Value Comparison

As shown for interassay comparison, correlations between GI values generated with the same assays in different laboratories were high (Fig 4) with Pearson correlation being above 0.9 in five of eight tested assays/algorithms. The slight variation of some assays was comparable with the range of variation seen between assays, and again, similar to the comparison with myChoice, the same two assays (QIAseq and CytoSNP) showed slightly higher variations. HRD status of the assays can be seen in the Data Supplement (Fig S2).

FIG 4.

Intraassay correlations. To define robustness of the assays, they were run at two different sites on the same samples. Correlation coefficients (Pearson-R) and concordance correlation coefficient (CCC) were determined.

DISCUSSION

Most frequently, the Myriad myChoice assay has been applied in clinical studies to measure GI. This assay is currently only centrally available, and therefore, alternative assays that can be used locally in standard diagnostic laboratories are urgently needed. All of the alternative assays can be carried out decentralized in a standard diagnostic laboratory; this also applies for the analysis. This is of great importance as decentralized testing is the standard methodology in most diagnostic laboratories across Europe and would decrease time for therapy decision making, avoid costs, and reduce turnaround times of outsourcing and ultimately lead to benefit for patients. Here, we evaluated the concordance of HRD statuses generated by several different molecular assays including Myriad myChoice. Before the clinical use of alternative assays, such validation is essential to ensure the accurate and consistent identification of patients with ovarian cancer who may benefit from PARPi combination therapy. Overall, interassay variability (including Myriad myChoice) was low and could be further decreased by the adaption of optimal cutoff. This is consistent with published smaller studies performing similar comparisons.^20-23 Worth mentioning, these publications compared only one assay with Myriad myChoice. Our study is, to our knowledge, the first one comparing a multitude of assays. The data are well in range or even better than variability data for a variety of biomarker assays used in a daily manner for decades. These deviations can occur due to biological conditions such as tumor heterogeneity, which also occur in other studies.^24,25 Thus, although the QiaSeq-Assay, which is still under development, showed lower kappa values and slightly lower Pierce correlations, we believe that after improvement almost all of these assays can be used safely and effectively for routine diagnostic setting once the assay development and clinical validation is completed.

Furthermore, one has to take into account that applying a fixed cutoff to delineate HRD positive from negative samples is artificial in any scenario even when using Myriad myChoice. It is unlikely that a significant difference in response between a patient with a GI score of 11 or a score of 13 will be observed even if the cutoff which delineates HRD positivity is set to 12. It is much more likely that there will be a continuum of increasing response rate with increasing GI.

In our study, we ran a variety of assays twice in different qualified molecular diagnostic laboratories using the same DNA. With respect to the results, an extremely high concordance of HRD status was observed; however, there were also some cases in which a different HRD status would be reported leading to a different patient treatment and patient outcome. Again, this is an observation for many molecular tests used in daily practice. Interestingly, the observed low interlaboratory variation is almost exactly in the range of assay variation itself and also very similar to interlaboratory variations shown for the Myriad myChoice test in the central laboratory and an university-based trained laboratory.²⁶

One important additional aspect not covered in this study is the potential variation of GI scores due to intratumor heterogeneity, when different tumor areas are used for analysis. We cannot rule out that variation by this factor might be even higher than variations that occur through the performing laboratory and the assay used. Regarding tumor heterogeneity, only sparse data have been published. In one previous study, we were able to establish that the variation of Myriad myChoice scores was low when primary tumors and matched metastases were tested.¹⁴

Future works have to address the question, to which extent the variety in BRCA mutation calling as the second important parameter in determining HRD status would affect on our group stratifications. Since we fully wanted to concentrate on GIS, we decided to exclude BRCA calling in this manuscript.

In other tumor types, such as breast and pancreatic cancers and also others with frequent HRD in which PARPi treatment is established,^14,27,28 GIS is not yet a standard clinical predictor. In these entities, prediction is currently solely based on the presence or absence of BRCA1/2 (and in some scenarios other HRR gene) mutations. However, it seems likely that the principle of using GI scoring as a surrogate marker rather than relying on the presence of a specific mutation to determine HRD positivity also holds true in other cancer types. Data on this topic are scarce and urgently needed. A recent large pan-cancer analysis favored distinct HRD score cutoffs in different cancer types.²⁹ Whether the extremely high concordance of different assays in GI scoring reported here for ovarian cancer also holds true for other tumor entities will have to be addressed in future studies.

As the main conclusion, our data prove strong evidence that when performed in qualified molecular laboratories many molecular assays can be used to assess HRD and will produce results comparable with those achieved using Myriad myChoice.

AUTHOR CONTRIBUTIONS

Conception and design: Nicole Pfarr, Karin von Schwarzenberg, Dario Zocholl, Sabine Merkelbach-Bruse, Lukas C. Heukamp, Gustavo Baretton, Daniel Kazdal, Peter Wild, Claudia Vollbrecht, Michael Hummel, Elena I. Braicu, Wilko Weichert

Administrative support: Nicole Pfarr, Karin von Schwarzenberg, Karsten Kleo, Melanie C. Demes, Peter Wild, Peter Schirmacher, Jalid Sehouli, Michael Hummel, Wilko Weichert

Provision of study materials or patients: Karin von Schwarzenberg, Karsten Kleo, Ulrich Lehmann, Stephan Bartels, Thomas McKee, Peter Schirmacher, Reinhard Büttner, Jalid Sehouli, Claudia Vollbrecht, Michael Hummel, Elena I. Braicu

Collection and assembly of data: Nicole Pfarr, Karin von Schwarzenberg, Sabine Merkelbach-Bruse, Janna Siemanowski, Eva-Maria Mayr, Sylvia Herold, Karsten Kleo, Lukas C. Heukamp, Eva-Maria Willing, Michael Menzel, Ulrich Lehmann, Stephan Bartels, Melanie C. Demes, Daniel Kazdal, Peter Wild, Thomas McKee, David Horst, Reinhard Büttner, Albrecht Stenzinger, Jalid Sehouli, Claudia Vollbrecht, Elena I. Braicu

Data analysis and interpretation: Nicole Pfarr, Dario Zocholl, Sabine Merkelbach-Bruse, Janna Siemanowski, Lukas C. Heukamp, Eva-Maria Willing, Michael Menzel, Ulrich Lehmann, Stephan Bartels, Shounak Chakraborty, Claudia Döring, Daniel Kazdal, Jan Budczies, Roland Rad, Peter Wild, Yann Christinat, Peter Schirmacher, Albrecht Stenzinger, Jalid Sehouli, Claudia Vollbrecht, Michael Hummel, Elena I. Braicu, Wilko Weichert, Eva-Maria Mayr, Thomas McKee

Manuscript writing: All authors

Final approval of manuscript: All authors

Accountable for all aspects of the work: All authors

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated unless otherwise noted. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/po/author-center.

Open Payments is a public database containing information reported by companies about payments made to US-licensed physicians (Open Payments).