Effectiveness of First‐Line Bevacizumab in Metastatic Colorectal Cancer: The Observational Cohort Study GRETA

The GRETA study, reported here, was designed to compare the overall survival (OS) of metastatic colorectal cancer patients treated with first‐line bevacizumab‐containing chemotherapy (B+CT) with the OS of those receiving chemotherapy (CT) alone.

Abstract

Background.

Scant real‐world data exist on the clinical outcomes associated with the use of bevacizumab‐containing chemotherapy (B+CT) in patients with metastatic colorectal cancer (mCRC). The primary objective of the GRETA cohort study was to compare the overall survival (OS) of patients with mCRC treated with first‐line B+CT versus chemotherapy (CT) alone, in an Italian clinical practice setting.

Materials and Methods.

Incident patients with mCRC were identified during the period 2010–2012 from five population‐based cancer registries in Italy. Cases were linked to regional health care utilization databases to obtain the entire spectrum of health services provided to each patient. Patients starting a first‐line treatment with B+CT or CT alone within 90 days from the diagnosis were included in the study cohort. A propensity score (PS) method was applied to account for residual confounding.

Results.

Of 480 patients with mCRC included in the study cohort, 21.0 received first‐line B+CT, and 79.0% received CT. Patients receiving B+CT were younger (p < .001) and underwent surgery more frequently (p = .001). The median OS was 22.5 and 14.6 months for B+CT and CT, respectively (p = .011). The corresponding hazard ratios adjusted by multivariate modeling and PS matched analysis were 0.82 (95% confidence interval [CI], 0.62–1.08) and 0.86 (95% CI, 0.56–1.33), respectively. Similar results were observed after subgrouping by age and surgery.

Conclusion.

In this Italian real‐world setting of unselected mCRC, the OS of patients treated with B+CT was consistent with previous observational and patient‐registry studies. However, definitive evidence of an improvement in OS cannot be drawn.

Implications for Practice.

Bevacizumab is a well‐established first‐line treatment for metastatic colorectal cancer. However, there is scarce evidence in the literature about its effectiveness in clinical practice. Evaluating this topic should be of interest for both clinicians and regulatory agencies. In this study, the median overall survival of the bevacizumab cohort was strikingly coherent with that reported in large observational series of unselected patients, thus suggesting a consistent and reproducible effect of the drug in clinical practice. Although consistent results were observed both in the overall population and in age and surgery subgroups, the present study did not offer definitive evidence of an improvement in OS.

Introduction

Bevacizumab (B; Avastin; Genentech, Inc., South San Francisco, CA), is a humanized recombinant monoclonal antibody that inhibits vascular endothelial growth factor, a common target that plays an important role in the process of tumor angiogenesis [1]. It was the first antiangiogenic agent approved for the first‐line treatment of patients with metastatic colorectal cancer (mCRC) by both U.S. (Food and Drug Administration, February 2004) and European Union (European Medicines Agency, January 2005) agencies. According to the results of randomized clinical trials (RCTs) [2], [3], [4], B plus standard regimens is currently recommended as first‐line treatment of mCRC. Despite these findings, however, there is still a relative paucity of data on the clinical outcomes associated with the use of B‐containing therapy in large, unselected, general clinical practice populations, including elderly patients and patients with less favorable prognostic factors than those enrolled in RCTs.

The “Generating Real‐world Evidence on Therapy of metastatic colorectal cancer with bevacizumab‐Avastin” (GRETA) study is an observational cohort study designed to compare the overall survival (OS) of patients with mCRC treated with first‐line B plus chemotherapy (B+CT) with the OS of those receiving chemotherapy (CT) alone, in an Italian clinical practice setting. Secondary objectives comprised the assessment of the baseline characteristics of patients assigned to the different therapeutic schemes (B+CT vs. CT alone) and of the treatment patterns in clinical practice, including therapy duration.

Materials and Methods

Data Source

The cohort of incident patients with mCRC was identified during the period 2010–2012 from five population‐based cancer registries from northern (provinces of Varese, Mantova, and Cremona) and southern (provinces of Palermo and Ragusa) Italy, covering an overall target population of more than 3 million people (nearly 5.4% of the whole Italian population).

To obtain the complete history of the health care services provided by the National Health Service to all cohort members, the records of the mCRC cases reported by cancer registries were linked with the corresponding records drawn from Regional Healthcare Utilization (HCU) databases. Information was retrieved from 3 years prior to the date of mCRC diagnosis until December 31, 2015. A variety of information is stored in the HCU databases, including the following: (a) outpatient dispensations of high‐cost drugs (including bevacizumab), recorded according to the Anatomical Therapeutic Chemical (ATC) codes; (b) inpatient ICD9‐CM diagnostic and procedure codes recorded by public or private hospitals; and (c) outpatient services, including diagnostic procedures and radiotherapies.

The data of patients with mCRC retrieved from the cancer registries were reviewed by the clinical oncologist of our team (S.B.), and patients with morphological characteristics of the tumor (coded through the ICD‐O‐3M classification [5]) inconsistent with mCRC were excluded from the analysis.

In order to preserve privacy, the patients’ identification codes were deleted from the database provided to the unique data analyst (M.F.). The study was approved by the Ethical Committee of the University of Milano‐Bicocca in June 2016.

Exposure to Antineoplastic Agents

Exposure to first‐line bevacizumab was defined by a two‐stage algorithm. Firstly, starting from the date of mCRC diagnosis, the first prescription of any antineoplastic drug approved for the treatment of mCRC was selected (the list of drugs approved in Italy for treating mCRC and the corresponding ATC codes are given in the Appendix). Secondly, because the information on chemotherapy (i.e., irinotecan, oxaliplatin, capecitabine, and fluorouracil) of inpatients was not available in our database, all the hospitalizations subsequent to the date of mCRC diagnosis reporting a code of CT, either as diagnosis or as procedure, were selected. Thus, the start date of therapy (index date) was defined as the earliest of the dates of the first hospitalization reporting a code of CT or the first prescription of an antineoplastic agent. Finally, starting from the index date, all the prescriptions of antineoplastic drugs in the next 21 days (i.e., the plausible duration of a CT cycle, that is, every 14 or 21 days) were selected. The cohort members were classified into two mutually exclusive categories. If at least one prescription of bevacizumab was dispensed within 21 days after the index date, the patient was considered exposed to bevacizumab‐containing therapy (i.e., patients starting on B+CT). If neither bevacizumab nor other antineoplastic biological drugs (i.e., cetuximab and panitumumab) were dispensed within 21 days after the index date, the patient was considered as starting a first line on CT alone.

Patients receiving biologics different from bevacizumab (i.e., cetuximab and panitumumab) within 21 days after the index date, as well as those who did not receive any therapy within 90 days after the index date (e.g., because they were never treated or received antineoplastic therapy afterwards), were excluded from the final cohort.

Follow‐Up

Cohort members accumulated person‐years of follow‐up from the date of the index date until the earliest date among the following: death; loss to follow‐up; December 31, 2015; or 3 years after the index date (because the number of patients who survived longer was too small).

Covariates

Baseline information included patient age, gender, year of mCRC diagnosis, surgery, tumor size and lymph node status (coded through the TNM Classification of Malignant Tumors [6]), grading, anatomical site (coded through the ICD‐O‐3T classification [5]), and histological characteristics (coded through the ICD‐O‐3M classification [5]). In addition, the Charlson comorbidity index score [7] was calculated using the diagnostic information in the 3 years preceding the index date, available from inpatient charts. The outpatients’ procedures recorded during the follow‐up were also evaluated, such as the number of computed tomography scans of the abdomen, magnetic resonance exams, x‐rays of digestive system, radiotherapies, and surgical procedures (the corresponding codes are given in the Appendix).

Statistical Analysis

The calculation of the sample size was based on the following a priori assumptions: (a) in Italy, nearly 52,000 cases of CRC are diagnosed every year [8]; (b) approximately 20%–25% have already presented metastases at the time of diagnosis [9]; and (c) cancer registries contributing to the current study cover a population of 3,187,359 inhabitants, representing about 5.4% of the entire Italian population. Therefore, the number of expected mCRC cases per year was 400 and, by considering a 3‐year period of recruitment (2010–2012), the total expected number was approximately 1,200 patients. Moreover, by assuming that one third of patients were untreated and that 20% of treated patients received B+CT, and by considering a one‐tail first type error of 0.05, a power of 0.80, and a median OS in the CT‐alone group of 15.6 months [2], the study would have been able to appreciate a gain in terms of median OS from 15.6 months in the CT group to 20.0 months in the B+CT group. The corresponding hazard ratio (HR) detectable under the aforementioned conditions was 0.78. The sample was calculated according to the crude comparison, based on the log‐rank test. However, the effect estimates of the therapy on the outcome were adjusted by several covariates (using a Cox regression model), implicating a reduction of the precision of the estimates and, consequently, a reduced ability of the study to appreciate a gain in the OS [10]. For this purpose, the variance of the parameter associated to the effect treatment estimated by the adjusted model (β = ln[HR]) was increased up to 50% compared with the variance of the parameter associated to the effect treatment estimated by the unadjusted model. Considering that using the unadjusted model the appreciable gain in OS was 4.4 months (from 15.6 to 20.0 months), the scenarios described above allowed us to detect at most a gain in OS of 4.5 months.

Differences in the baseline characteristics of the two treatment groups were tested by the chi‐square test, its version for the trend, or Fisher's exact test, where appropriate. OS, i.e., the primary endpoint of the study, was calculated by means of the Kaplan‐Meier estimator, and the log‐rank test was used for testing survival differences between patients starting on B+CT or CT alone.

Predictors of OS were evaluated by fitting a Cox proportional hazard model. The model included, other than the baseline exposure to antineoplastic agents (i.e., B+CT vs. CT alone), the above‐listed covariates measured during the follow‐up and included in the model as time‐dependent covariates. Results were expressed as HR and corresponding 95% confidence intervals (CI).

Sensitivity Analyses

To verify the robustness of our findings, the following three sensitivity analyses were performed. In the first sensitivity analysis, a longer time window was used to define the first‐line treatment (i.e., a 42‐day period, instead of 21 days as in the main analysis). In the second analysis, because information on cancer size, lymph node status and grading was missing in about 24% of cohort patients, a multiple imputation technique (i.e., the fully conditional specification) was used to account for missing data [11]. Finally, to address the issue of residual confounders influencing the first‐line treatment choice, a propensity score 1:1 matched analysis was performed [12]. In this analysis, propensity scores were calculated using, besides the covariates listed above, the following: diabetes, hypertension, cerebrovascular/ischemic heart/respiratory/renal diseases, time to treat (i.e., duration of time between mCRC diagnosis and start of first‐line treatment), number of hospitalizations, outpatient services, and drug prescriptions in the 3 years prior to the index date. This propensity score 1:1 matched sample was then used to fit the propensity score‐adjusted Cox model.

For all the tested hypotheses, two‐tailed p values less than .05 were considered statistically significant.

Results

Patients

During the study period, 1,118 incident patients with mCRC were identified. This figure is very close to the expected 1,200 patients with mCRC. However, the number of patients included in the final cohort (480) was lower than the expected (800 patients) because of exclusion criteria (i.e., patients treated with other drugs and those who started the first‐line treatment more than 90 days after the diagnosis of mCRC). Of the 480 patients selected, 101 were treated with B+CT and 379 with CT alone (Fig. 1 ). Age at diagnosis was a strong predictor of being untreated. Indeed, around 80% of patients aged more than 80 years did not receive any pharmacological treatment. Moreover, 197 (47.5%) untreated patients died within 3 months from the diagnosis of mCRC, indicating a terminal stage of the disease.

Figure 1.

Flow chart of cohort selection.

Abbreviations: CRC, colorectal cancer; CT, chemotherapy; mCRC, metastatic colorectal cancer.

The baseline characteristics of the study cohort are given in Table 1. Compared with patients using CT alone, those on B+CT were younger and underwent a surgical intervention more frequently. There was no statistical evidence that other baseline features were associated with the use of bevacizumab.

Table 1. Baseline characteristics of 480 patients with mCRC included in the study cohort.

^a According to the chi‐square test (gender, Charlson comorbidity, tumor site, nodal status), its version for the trend (age class, year of mCRC diagnosis), or Fisher's exact test (tumor grading and size).

^b Including one patient with diagnosis of anal cancer.

Abbreviations: B, bevacizumab; CT, chemotherapy; mCRC, metastatic colorectal cancer.

Treatment Patterns

The median duration of first‐line bevacizumab treatment was 176 days. Among the 101 patients starting on B+CT, 17 (16.8%) switched to cetuximab and 11 (10.9%) to panitumumab during the follow‐up; the median time to switch from the index date was of 300 and 648 days, respectively. Among 379 patients starting on CT alone, 105 (27.7%) switched to bevacizumab during the follow‐up, after a median of 218 days; 36 (9.5%) switched to cetuximab, after a median of 288 days; and 10 (2.6%) switched to panitumumab, after a median of 586 days.

Overall Survival

Overall, 360 (75%) patients died during the follow‐up: 68/101 (67.3%) in the B+CT group and 292/379 (77.0%) in the CT‐alone group. The proportions of survivors at 1, 2, and 3 years were 71%, 48%, and 32%, respectively, in the B+CT group and 55%, 34%, and 23% in the CT‐alone group (Fig. 2). The median OS was 22.5 and 14.6 months among patients who started on B+CT and CT alone, respectively (p = .011), with a crude HR of 0.71 (95% CI, 0.55–0.93; Table 2). However, after adjusting for the covariates measured at baseline (multivariable Cox model; Table 3), the resulting HR was 0.82 (95% CI, 0.62–1.08), leading to a non‐statistically significant result (Table 2). Among the covariates included in the model, younger age and surgery either at baseline or during the follow‐up had a major impact on the HR, decreasing by approximately 50% the probability of death. Therefore, to assess whether the effect of bevacizumab on OS was independent of these variables, we subgrouped patients by age (<70 years vs. ≥70 years) or surgery (yes vs. no) and then by first‐line treatment (B+CT vs. CT alone). As summarized in Table 2, among patients aged <70 years, the median OS was 25.1 months in the B+CT group and 16.8 months in the CT‐alone group, with an HR of 0.75 (95% CI, 0.53–1.05). In elderly patients, the median OS was 15.8 and 11.5 months, respectively, with an HR of 0.77 (95% CI, 0.50–1.18). A similar trend was observed in patients who underwent surgery and received B+CT versus CT alone (median OS, 24.4 and 20.9 months, respectively; HR, 0.88; 95% CI, 0.63–1. 21) and in those who did not undergo surgery and were given B+CT versus CT alone (median OS, 16.8 and 10.2 months, respectively; HR, 0.59; 95% CI, 0.36–0.95; Table 2).

Figure 2.

Kaplan‐Meier survival estimates of 480 patients with metastatic colorectal cancer included in the study cohort, stratified by first‐line treatment.

Abbreviations: B, bevacizumab; CT, chemotherapy; OS, overall survival.

Table 2. Median OS and corresponding HRs in the overall population and age and surgery subgroups, by treatment.

^a Estimated by multivariable Cox proportional hazard model.

Abbreviations: —, not applicable; B, bevacizumab; CT, chemotherapy; HR, hazard ratio; OS, overall survival.

Table 3. Hazard ratios of death and 95% confidence intervals, estimated by a multivariable Cox proportional hazard model.

^a Estimated by multivariable Cox proportional hazard model.

^b At least one service experienced during follow‐up.

Abbreviations: CI, confidence interval; HR, hazard ratio.

Finally, in contrast to age and surgery, an increment in the number of radiotherapies and imaging procedures (magnetic resonance and x‐ray) during the follow‐up were associated with a higher HR of death, probably indicating that patients followed up more frequently are those with worse health conditions.

Sensitivity Analyses

One hundred twenty‐four and 351 patients were classified as starting on B+CT and CT alone, respectively, when a period of 42 days was considered to define the first‐line treatment (5 patients belonging to the initial cohort used other biologics in first‐line and thus, according to the exclusion criteria, were not included in this analysis). Median OS was 18.5 and 14.4 months, respectively (p = .044). Crude and adjusted HRs of death associated with starting on B+CT versus CT alone were 0.78 (95% CI, 0.62–0.99) and 0.90 (95% CI, 0.70–1.16) respectively (data not shown).

The estimated effect of starting on B+CT did not change substantially after multiple imputations of missing values, the HR being 0.83 (95% CI, 0.61–1.14; data not shown). Similarly, the analysis of 94 propensity score‐matched couples of B+CT and CT patients showed similar results, with an HR of 0.86 (95% CI, 0.56–1.33). The number of deaths was 64 (68.1%) in the B+CT group and 70 (74.5%) in the CT‐alone group, and the median OS was 23.4 and 17.5 months, respectively. The OS curves of the matched analysis are shown in Figure 3.

Figure 3.

Kaplan‐Meier survival estimates of 94 propensity score‐matched couples of patients with metastatic colorectal cancer, stratified by first‐line treatment.

Abbreviations: B, bevacizumab; CT, chemotherapy; OS, overall survival.

Discussion

The present cohort study, carried out by linking data from five Italian population‐based cancer registries and health care utilization databases and including 480 unselected patients with mCRC, ranks among the largest studies published so far that have analyzed the outcome of treatment with bevacizumab combined with CT in the real‐practice setting, in Italy. Compared with the pivotal RCT [2], patients included in our cohort were older (median age, 68 vs. 59 years), and no exclusion criteria were applied on performance status, length of life expectancy, hematologic, hepatic and renal function, comorbidities, and cotreatments. Therefore, patients in the present study closely represent a typical mCRC population seen in daily practice, treated with CT regimens chosen by the physician under the specifications of the Italian health care system in the period 2010–2012.

In our study, the cohort members who started treatment on B+CT had a median OS of 22.5 months. Importantly, this is in line with the findings from previous large observational cohort studies conducted in the period from B approval to 2012, reporting a median OS between 17.3 and 25.3 months [13], [14], [15], [16], [17], [18], [19], [20], and from a pooled analysis of first‐line trials with FOLFIRI‐based CT+B, reporting a median OS of 23.7 months [21]. This is a relevant result, as it confirms the reproducibility of first‐line B+CT combination across several countries and clinical practice settings, regardless of the CT backbone.

Because of the fact that, at baseline, the two treatment groups differed for age and frequency of surgical procedures, we adjusted the HR for these covariates by multivariate modeling and found 0.82 (95% CI, 0.62–1.08). Next, we assessed the effect of first‐line B after subgrouping patients by age (<70 years vs. ≥70 years) or surgery (yes vs. no) and then by first‐line treatment (B+CT vs. CT alone). Similar results were observed in all comparisons. The discrepancy between the crude and adjusted results may be at least partly explained by the limited sample size included in the GRETA study, which is approximately half of the sample size initially planned.

A number of previous large observational studies compared the effectiveness of first‐line B+CT versus CT alone. Meyerhardt et al. recruited a cohort of 2,526 patients with mCRC from 2002 to 2007 [18]. First‐line bevacizumab was associated with significant improved median OS (adjusted HR, 0.85; 95% CI, 0.78–0.93), and the effect was more apparent with irinotecan‐ versus oxaliplatin‐based CT. However, after restricting the study cohort to years 2004–2007 (i.e., from the year of the approval of bevacizumab), during which the available treatment options changed, the OS improvement did not reach statistical significance (adjusted HR, 0.93; 95% CI, 0.84–1.02). Hammerman et al. compared 1,052 patients with mCRC on first‐line bevacizumab between 2006 and 2009 with 687 historical controls diagnosed between 2001 and 2004 (i.e., when bevacizumab was not available yet) [19]. They found an adjusted HR of 0.75 (95% CI, 0.68–0.84) [19]. Similarly, Renouf et al. observed that the 448 patients with mCRC diagnosed in 2006 (i.e., when bevacizumab started to be used for treating patients with mCRC) had an improved OS compared with the 969 patients with mCRC diagnosed before, the corresponding OS being 17.3 and 13.8 months (p < .001) [20]. However, whether the protective effect observed from the latter two studies may be attributed to bevacizumab or to changes in mCRC clinical patterns over the years is unclear.

In our case, the sample size was smaller than in the above‐reported studies. Moreover, GRETA describes the real‐world use of medications between 2010 and 2012 in five Italian provinces: although the management of mCRC changed in 2014 [22], we were not able to retrieve the information on the CT backbone used in association with bevacizumab and, therefore, to track the possible changes in therapeutic combinations.

A beneficial effect of bevacizumab in elderly patients was reported by a recent meta‐analysis of RCTs and observational studies [23]. However, the present study does not allow to draw any solid conclusion, as, among 211 patients aged ≥70 years, only 31 (14.7%) had received B+CT. Still, these data are important as they provide evidence of undertreatment among the elderly, which was a common practice in that period. Accordingly, 0/39 patients aged >80 years were treated with B+CT, whereas 70/269 (26%) of those younger than 70 years were treated with B+CT (data not shown).

Our study has several strengths. At present, GRETA is the only study performed in Italy to compare the effectiveness of two therapeutic strategies by using both cancer registries and HCU databases. This represents a new opportunity because (a) cancer registries allow performance of population‐based studies (i.e., all the cases occurred in a given population are potentially captured) characterized by high diagnostic validity (i.e., mCRC cases were signaled from cancer registries accredited by the Italian Association of Cancer Registries [24], [25]) and (b) HCU data allow the retrieval of the entire history of health care services provided to cohort members. Furthermore, our study involves distinct geographical areas from northern to southern Italy, thus reflecting the heterogeneity of clinical practice in treating broad and different populations. As patients were not selected by age and previous comorbidities, the representativeness of the routine clinical practice is guaranteed.

Finally, the sensitivity analyses generated similar results, making sources of systematic uncertainty unlikely.

This study has also some limitations. First, the administrative purpose for which the HCU databases are designed limits the completeness of the data recorded. In particular, information about clinical features (e.g., performance status and K‐RAS gene mutation) and therapeutic patterns (e.g., FOLFIRI or FOLFOX) were not available. Secondly, the sample size was calculated by assuming that about 20%–25% of CRC cases were metastatic at diagnosis [26]. However, in our sample we observed a lower proportion of CRC presenting metastasis at diagnosis (about 15%), probably as a consequence of colorectal cancer screening programs [27]. Moreover, based on the exclusion criteria, patients who were untreated, were treated with other drugs, and had started the first‐line treatment more than 90 days after the diagnosis of mCRC, accounting for a total of 615, were excluded from the final cohort. Finally, as with any observational study, we cannot rule out the possibility that patients receiving bevacizumab differ from those who did not receive bevacizumab for some unmeasured features that may be associated with mortality risk.

Conclusion

In the real‐world setting of unselected patients with mCRC treated in first line between 2010 and 2012 in five Italian provinces, the OS of patients treated with B+CT was consistent with that reported in previous studies, suggesting its reproducibility across several countries and clinical practice settings. However, definitive evidence of an improvement in OS cannot be drawn.

List of drugs approved in Italy for the treatment of mCRC:

Fluorouracil. ATC L01BC02

Capecitabine. ATC L01BC06

Oxaliplatin. ATC L01XA03

Irinotecan. ATC L01XX19

Bevacizumab. ATC L01XC07

Cetuximab. ATC L01XC06

Panitumumab. ATC L01XC08

List of ICD9 codes of selected procedures:

Computerized tomography of the abdomen: 88.01

Magnetic resonance: 88.91, 88.92, 88.93, 88.94, 88.95, 88.97

X‐ray of digestive system: 87.6

Radiotherapy: 92.23

Surgery: 32.3, 34.59, 42.86, 44.39, 45.41, 45.43, 45.49, 45.51, 45.61, 45.62, 45.71–45.76, 45.79, 45.8, 45.90–45.94, 46.01–46.04, 46.10, 46.11, 46.13, 46.14, 46.20–46.23, 46.40, 46.43, 46.51, 46.52, 46.76, 46.79, 46.85, 46.93, 46.94, 48.33, 48.35, 48.49, 48.5, 48.62, 48.63, 48.69, 48.79, 50.22, 50.29, 50.3, 54.12, 54.4, 54.51, 54.59, 54.61, 68.8.

Chemotherapy: 99.25, 99.28, V58.1, V58.11

Author Contributions

Conception/design: Matteo Franchi, Antonietta Caputo, Giovanni Corrao

Provision of study material or patients: Giovanna Tagliabue, Paolo Ricci, Walter Mazzucco, Rosario Tumino

Collection and/or assembly of data: Matteo Franchi

Data analysis and interpretation: Matteo Franchi, Sandro Barni, Antonietta Caputo, Giovanni Corrao

Manuscript writing: Matteo Franchi, Giovanni Corrao

Final approval of manuscript: Matteo Franchi, Sandro Barni, Giovanna Tagliabue, Paolo Ricci, Walter Mazzucco, Rosario Tumino, Antonietta Caputo, Giovanni Corrao

Disclosures

Antonietta Caputo: Roche S.p.A. (E); Giovanni Corrao: Novartis, GlaxoSmithKline, Roche, Amgen, Bristol‐Myers Squibb (RF), Roche (H). The other authors indicated no financial relationships.

(C/A) Consulting/advisory relationship; (RF) Research funding; (E) Employment; (ET) Expert testimony; (H) Honoraria received; (OI) Ownership interests; (IP) Intellectual property rights/inventor/patent holder; (SAB) Scientific advisory board