Patient-Centered Approach to Benefit–Risk Characterization Using Number Needed to Benefit and Number Needed to Harm: Advanced Non–Small-Cell Lung Cancer

Abstract

This work summarizes the benefit and risk of the results of clinical trials submitted to the US Food and Drug Administration of therapies for the treatment of non–small cell lung cancer (NSCLC) using number needed to benefit (NNB) and number needed to harm (NNH) metrics. NNB and NNH metrics have been reported as potentially being more patient centric and more intuitive to medical practitioners than more common metrics, such as the hazard ratio, and valuable to medical practitioners in complementing other metrics, such as the median time to event. This approach involved the characterization of efficacy and safety results in terms of NNB and NNH of 30 clinical trials in advanced NSCLC supporting US Food and Drug Administration approval decisions from 2003 to 2017. We assessed trends of NNB over time of treatment (eg, for programmed death 1 inhibitors) and variation of NNB across subpopulations (eg, characterized by epidermal growth factor receptor mutation, programmed death ligand 1 expression, Eastern Cooperative Oncology Group performance status, age, and extent of disease progression). Furthermore, the evolution of NNB of treatments for advanced NSCLC was charted from 2003 to 2017. Across subpopulations, NNB, on average, was 4 patients for approved targeted therapies in molecularly enriched populations, 11 patients for approved therapies in nonmolecularly enriched populations, and 23 patients for withdrawn or unapproved therapies. Furthermore, the NNB analysis showed variation for attributes of epidermal growth factor receptor mutations, level of programmed death 1 expression, Eastern Cooperative Oncology Group performance status, etc. When considering the best-case subpopulations and available drugs, the NNB frontier reduced from an estimated value of 7.7 in 2003 to an estimated value of 2.5 in 2017 at the estimated median overall survival—equal to 6 months—of an untreated patient.

INTRODUCTION

There were an estimated 538,243 people living with lung and bronchus cancer in the United States in 2016.¹ An estimated 228,150 new cases and 142,690 deaths related to lung and bronchus cancer are likely to occur in the United States in 2019.¹ Approximately 80% to 85% of lung cancers are non–small-cell lung cancers (NSCLCs).²

CONTEXT

• Key Objective

• To quantify the benefit and risk of non–small-cell lung cancer (NSCLC) therapies in a manner that has the potential to be patient centric and intuitive to clinicians while complementing previously reported metrics, such as overall survival, progression-free survival, and objective response rate. The quantification was demonstrated using estimates of the number needed to benefit (NNB) and the number needed to harm (NNH) on the basis of recent clinical trial results for NSCLC therapies.

• Knowledge Generated

• The results of this work suggest differences in estimated NNB according to therapy type. Notable improvements in NNB over the temporal history of NSCLC therapies were observed, particularly in the more recent clinical trial results of targeted therapies.

• Relevance

• The NNB and NNH metrics used to assess the clinical trial data have the potential to enhance patient-centric understanding of the benefit and risk of treatments and to further inform treatment decisions.

The prognosis for patients with NSCLC varies by the stage of disease at diagnosis. The 5-year relative survival rates of patients with stage IVA and IVB NSCLC are 10% and 1%, respectively.² Since the beginning of the 21st century, several treatment options have become available for NSCLC, for multiple types of patients (eg, nonsquamous, squamous, with genetic driver mutations, etc), with various mechanisms of action (eg, epidermal growth factor receptor [EGFR] inhibitors, ALK inhibitors, VEGF inhibitors, programmed death 1 [PD-1] inhibitors, etc), and over various lines of therapy. With the advent of targeted therapies, the prognosis for patients with advanced NSCLC has improved significantly from 2003 to 2014. In 2003, estimated median overall survival (OS) of patients—no subpopulations segregated—with NSCLC who were treated with chemotherapy was 10 months (vinorelbine plus cisplatin, control in cetuximab FLEX study).³ In 2014, median OS of previously untreated EGFR mutation +ve patients who were treated with erlotinib was estimated to be 23 months.⁴

There are advantages to shared decision making about treatments by patients and doctors and taking a patient-centered approach—that is, taking into account the patient’s history, genotype, prior treatments, preexisting conditions, state of disease, etc.⁵ This complex decision making can be facilitated by presenting knowledge in an intuitive manner to patients about the benefit and risk of various drugs learned from clinical trials.

US Food and Drug Administration (FDA) regulatory approval decisions are intended to be based on the evaluation of the benefit and risk of therapies.^6,7 As part of the Prescription Drug User Fee Act and the US Food and Drug Administration Safety and Innovation Act,⁶ the FDA committed to implement a structured benefit–risk approach in assessments of new drug applications and biologic license applications.^8,9

Assessment of the approvability of oncologic drugs typically relies on the following: clinical trial end points, such as objective response rate (ORR), progression-free survival (PFS), and OS (the benefits); and rates and types of adverse events, such as fatalities, nonfatal serious, and common adverse events potentially attributable to the drug product (the risks). Time-to-event metrics, such as PFS and OS, are characterized in terms of hazard ratio (HR) and median time-to-event metrics—for example, median PFS and median OS. Metrics that are closely related to HR are relative risk reduction and absolute risk reduction (ARR). The limitations of ARR are discussed in Cook et al,¹⁰ with similar limitations for relative risk reduction. These limitations include the potential nonintuitiveness to patients and health care practitioners, given the probabilistic nature of these metrics. This work analyzes the benefit and risk of drug clinical trials using metrics^11,12 that have been reported to be potentially more patient centric and more intuitive to medical practitioners than metrics, such as HR, and potentially valuable to medical practitioners in complementing other metrics, such as ORR, median PFS, and median OS. We used number needed to benefit (NNB) and number needed to harm (NNH) metrics and applied them to the clinical trial results of new drug applications recently submitted to the FDA for the treatment advanced NSCLC. Additional information on the context and calculations of the number needed to treat (NNT), NNB, and NNH is provided in the Data Supplement. Health technology assessments of drug products typically involve the estimation of quality-adjusted life years gained as a result of the drug product compared with a standard of care or control of a randomized clinical trial. Limitations of the quality-adjusted life years–based approach, from ethical, methodologic, and context-specific perspectives, are well documented (eg, Pettitt et al¹³).

In prior work, a quantitative benefit–risk analysis approach, building on the FDA’s qualitative approach, was described¹⁴ and used for analysis of regulatory decision making in new drug applications and biologic license applications for the treatment of patients with NSCLC,¹⁵ multiple myeloma,¹⁶ and melanoma.¹⁷ This approach mapped the FDA’s characterization of the efficacy in terms of clinical end points, such as ORR, PFS, and OS, and characterization of safety in terms of fatal, nonfatal, and common adverse events to area under the quality-of-life curve. Benefit was expressed in terms of median OS and PFS, and risk as a summation of the product of seriousness and probability of a patient dying or undergoing nonfatal serious adverse events or only common adverse events. The seriousness of each of the above categories—fatal, nonfatal serious, and common adverse events—of risk was an estimate of the area under the quality-of-life curve lost by the corresponding patient.¹⁴

This work quantifies benefit and risk of drug products in terms of the NNT metric, specifically NNB and NNH. Furthermore, it uses the NNT metric to capture the evolution of benefit and risk of drug products for NSCLC since the beginning of the 21st century.

METHODS

Candidate Treatments and Regulatory Decisions

A major focus of this work was to develop an understanding of the extent to which NNB and NNH analyses align with recent decisions made by the FDA regarding new drugs to treat NSCLC. This analysis was based on publicly available details of selected clinical trials, the results of which were published between 2003 and 2017 and which formed a basis for the assessment of the approval of drug products for NSCLC by the FDA. Thirty trials were identified as detailed and shown in the Data Supplement. These trials were selected to include drug products belonging to a variety of therapeutic classes (eg, chemotherapy and targeted therapies, including EGFR inhibitors, ALK inhibitors, PD-1 inhibitors, VEGF inhibitors, etc); populations who were molecularly enriched or otherwise; various lines of therapies; various primary end points (eg, ORR, PFS, and/or OS); and regulatory statuses (eg, accelerated approval, regular approval, withdrawn submissions, etc). Clinical trial information was available either within the FDA Medical Reviews^18-27 or in journal publications.^3,4,28-51

NNT, NNB, and NNH

NNT is the expected number of patients who have to be treated to prevent one additional event—for example, death, serious adverse event, etc.^10-12 NNT is the reciprocal of ARR. ARR, in the context of a randomized clinical trial, is the difference between the probabilities of the occurrence of the event of interest—for example, death, serious adverse event, etc—in the experimental drug arm and control arm.

When the event of interest is considered to be a benefit to the patient—for example, a patient surviving in a clinical trial arm beyond a given time point—the NNT is the NNB, which is the reciprocal of the difference between the probabilities of the patient surviving beyond the given time point in the experimental drug arm and control arm. When the event of interest is considered to be a harm to the patient—for example, a patient undergoing a serious adverse event during the clinical trial—the NNT is the NNH, which is the reciprocal of the difference between the probabilities of the patient undergoing a serious adverse event in the experimental drug arm and control arm.¹² Additional information on the context and calculations of NNT, NNB, and NNH is provided in the Data Supplement.

Assessing the Impact of the Evolution of NSCLC Treatments on Clinical Outcomes for Patients With NSCLC

The period from 2003 onward has seen the approval of a number of therapies targeted to various types of NSCLC—for example, EGFR+ve, ALK+ve, BRAF+ve, PD ligand 1+ve, etc. Clinical trial results (summarized in the Data Supplement) show that these therapies have demonstrated substantial improvements in efficacy over chemotherapy. In this work, we try to assess the impact of the above developments in the treatment of NSCLC on clinical outcomes for patients with NSCLC. Specifically, we try to capture the impact on clinical outcomes for patients who are newly diagnosed with NSCLC and expected to experience progression through multiple lines of therapy. We do this using two approaches for patients newly diagnosed with some form of NSCLC: by estimating the growth in median OS over time, and by estimating the reduction in NNB—compared with an untreated patient—over time. The approach to estimating the growth in median OS over time for a newly diagnosed patient has been described in prior work,^16,17 and its application to NSCLC is discussed below in Evolution of Estimated Median OS and Corresponding NNB of NSCLC Therapies. The approach to estimate the reduction in NNB—compared with an untreated patient—over time is discussed below.

An NNB frontier, with respect to a given treatment alternative—for example, targeted therapy, chemotherapy, or placebo—for a set of treatments is the lowest estimated NNB from among the treatments within the set. The set of treatments for which the frontier NNB is calculated can be selected in many ways—for example, on the basis of subpopulation of patients (eg, EGFR+ve NSCLC), line of therapy (eg, first, second, third line of therapy, etc), or available treatments at a given point in time (eg, available first-line treatments for NSCLC in the year 2005, etc). In this work, we estimate the frontier NNB for patients with newly diagnosed NSCLC at various time points. The frontier NNB, with respect to an untreated patient, for a patient with newly diagnosed NSCLC at a given point in time can be estimated in the following ways: on the basis of an estimated median OS of various treatments at various lines of therapy, as discussed and illustrated below in Evolution of Estimated Median OS and Corresponding NNB of NSCLC Therapies, or on the basis of HRs of various treatments in various lines of therapy, as discussed and illustrated in the Data Supplement. Thus, we define the NNB frontier for a given year as the best possible NNB over an untreated patient—possibly of a targeted subpopulation—aggregated over all lines of therapy in the given year. We chart the evolution of NNB over multiple years (2003 to 2017) as the reduction in the NNB frontier over the given years. The NNB frontier for the baseline year (2003) is the NNB of chemotherapy over an untreated patient in the baseline year. Aggregation of NNB over lines of therapy is based on compounding of HRs of event-free survival—PFS or OS—at the median OS of the untreated patient (50% surviving untreated patients). The NNB frontier would shift lower if it is calculated at a lower percentile—lower than the 50^th percentile—of event-free survival of an untreated patient. The NNB frontier is calculated at 2 levels: stratified by types of patients (eg, EGFR+ve, ALK+ve, BRAF+ve, squamous cell, etc), and regardless of patient type. The latter is calculated as the lower bound of all patient types for each year in the horizon of analysis (2003 to 2017). Formulas for estimating the NNB frontier over time are given in the Data Supplement.

RESULTS

NNB and NNH for Selected Trials

The estimated NNB and NNH as described in the approach, associated with each of the NSCLC drugs along with their CIs, are listed in Table 1. A graphical representation of the NNB and NNH associated with NSCLC drugs is shown in Figure 1. For each clinical trial, the NNB computed based on the ORR reported could be different from that computed based on the PFS reported, which, in turn, could be different from the NNB computed based on the OS reported. Any such difference could be a result of various factors, including patients discontinuing clinical trial, deaths from disease progression or toxicity, crossover, and subsequent treatments. Figure 2 shows the NNB and NNH of selected clinical trials grouped by therapeutic class of the drug product and molecular enrichment of the patient population. It can be seen that NNB is:

TABLE 1.

Summary of NNB and NNH for Selected Clinical Trials (see the Data Supplement for the number of patients taken into account for NNB and NNH estimation)

FIG 1.

(A) Number needed to benefit (NNB) of selected clinical trials (first line). (B) NNB of selected clinical trials (non–first line). (C) Number needed to harm (NNH) of nonfatal serious adverse events of selected clinical trials (first line). (D) NNH of nonfatal serious adverse events of selected clinical trials (non–first line). ORR, objective response rate; OS, overall survival; PFS, progression-free survival.

FIG 2.

(A) Number needed to benefit (NNB) of selected clinical trials grouped by therapeutic class of the experimental drug and molecular enrichment (Mol. Enr.) of patient populations (and within each group ordered chronologically by the date of publication; first line). (B) NNB of selected clinical trials grouped by therapeutic class of the experimental drug and molecular enrichment of patient populations (and within each group ordered chronologically by the date of publication; non–first line). (C) Number needed to harm (NNH) of nonfatal serious adverse events of selected clinical trials grouped by therapeutic class of the experimental drug and molecular enrichment of patient populations (first line). (D) NNH of nonfatal serious adverse events of selected clinical trials grouped by therapeutic class of the experimental drug and molecular enrichment of patient populations (non–first line). ALK, anaplastic lymphoma kinase; EGFR, epidermal growth factor receptor; ORR, objective response rate; OS, overall survival; PD-1, programmed death 1; PFS, progression-free survival.

• Relatively low for approved targeted therapies in molecularly enriched patient populations (eg, dabrafenib plus trametinib in BRAF mutation+ve patients; crizotinib, ceritinib, and alectinib in ALK+ve patients; erlotinib and afatinib in EGFR+ve patients; and pembrolizumab in PD ligand 1+ve patients), and higher for approved therapies in nonmolecularly enriched populations (eg, erlotinib, atezolizumab, nivolumab, necitumumab, bevacizumab, and pemetrexed)

• Highest for therapies for which the submissions were withdrawn or not approved (eg, cetuximab and gefitinib for non–first line use)

Also, it can be seen that NNH is relatively small with a negative sign—better than control—for approved targeted therapies in molecularly enriched populations.

NNB Analysis Over Time and Subpopulations

NNB analyses over time for selected trials are shown in Figure 3. It can be seen that:

FIG 3.

(A) Number needed to benefit (NNB) over time and epidermal growth factor receptor (EGFR) mutations subpopulations stratified by the presence of EGFR mutations in the gefitinib IPASS trial.³⁰ (B) NNB over time and subpopulations in the nivolumab CheckMate 057 trial⁴¹ stratified by programmed death ligand 1 (PD-L1) expression. (C) NNB over time and subpopulations in the crizotinib 1007 trial³⁶ stratified by the presence of brain metastasis. (D) NNB over time and subpopulations in the crizotinib 1007 trial³⁶ stratified by the Eastern Cooperative Oncology Group performance status (ECOG PS). OS, overall survival; PFS, progression-free survival.

• The overall NNB over time for gefitinib in the IPASS trial³⁰ (Fig 3A) is a superposition of the negative NNB over time for EGFR−ve patients and positive NNB over time for EGFR+ve patients.

• Patients with PD-1 expression of 10% or greater in the Checkmate 057 trial⁴¹ of nivolumab (Fig 3B) benefit early from treatment compared with patients with PD-1 expression of less than 10%.

• Patients with brain metastasis have a lower NNB compared with patients with no brain metastasis in the crizotinib 1007 trial³⁶ (Fig 3C).

• Patients with ECOG performance status of 2 have a lower NNB compared with patients with ECOG performance status of less than 2 in the crizotnib 1007 trial³⁶ (Fig 3D).

In Figures 3A and 3B, the points at which the NNB curve crosses the time axis correspond to the time points when the corresponding Kaplan-Meier curves are estimated to cross each other. NNB analysis for subpopulations in the nivolumab Checkmate 057 trial⁴¹ (Fig 4A), the alectinib ALEX trial⁴⁹ (Fig 4B), and the afatinib LUX-Lung 8 trial⁴² (Fig 4C) indicate that, generally, the point estimate of NNB is lower for patients with lower ECOG performance status. An exception to this general observation is seen in crizotinib 1007 trial³⁶ (Fig 3D)—that is, the apparent lower NNB of crizotinib over chemotherapy for patients with higher ECOG performance status could be a result of their lower tolerance to the toxicity of chemotherapy. This possibility is supported by the higher NNB for higher ECOG performance status in trials in which the experimental drug and control were both considered to be relatively toxic, as in the nivolumab trial Checkmate 057⁴¹ (Fig 4A), or relatively less toxic, as in the alectinib ALEX trial⁴⁹ (Fig 4B) and the afatinib LUX-Lung 8 trial⁴² (Fig 4C).

FIG 4.

(A) Number needed to benefit (NNB) over subpopulations in the nivolumab CheckMate 057 trial⁴¹ stratified by Eastern Cooperative Oncology Group performance status (ECOG PS). (B) NNB over subpopulations in the alectinib ALEX trial⁴⁹ stratified by ECOG PS. (C) NNB over subpopulations in the afatinib LUX LUNG 8 trial⁴² stratified by ECOG PS. OS, overall survival; PFS, progression-free survival.

Evolution of Estimated Median OS and Corresponding NNB of NSCLC Therapies

Figure 5A shows the evolution of the estimated median OS of NSCLC treatments from 2003 to 2017 for various patient types. The thick dark blue line in Figure 5A—labeled Overall—shows the evolution of the highest estimated median OS of NSCLC treatments among all types of patients considered. Figure 5B shows the evolution of the NNB frontier for NSCLC treatments from 2003 to 2017 for various patient types. The thick dark blue line in Figure 5B shows the evolution of the overall NNB frontier—that is, the best NNB across all types of patients considered—for NSCLC treatments.

FIG 5.

(A) Evolution of estimated median overall survival (OS) of non–small-cell lung cancer (NSCLC) drug products. (B) Evolution of number needed to benefit (NNB; estimated on the basis of median OS) of NSCLC drug products. FL, first line.

To estimate median OS, the approach and ORR–OS and PFS–OS correlations used are the same as in previous work.¹⁵ In Figure 5A, the evolution of the estimated median OS of NSCLC treatments for each patient type is calculated as in previous work.^16,17

The frontier median OS—highest estimated median OS—among all types of patients considered (shown by the thick dark blue line in Fig 5A) is the maximum of the estimated frontier median OS of all patient types considered in each year. For example, the estimated frontier median OS in 2017 is 39.7 months, corresponding to the EGFR+ve patient type, which is higher than that of the other patient types considered (37.9 months in ALK+ve patients, and 23.2 months in BRAF+ve, nonsquamous, nondriver mutation, and patients with squamous NSCLC).

Figure 5B shows the evolution of the NNB frontier for each patient type derived from the corresponding median OS frontier shown in Figure 5A. The NNB frontier for each year is derived from the median OS frontier for the same year by comparing the frontier median OS with the median OS of an untreated patient—estimated as 6 months.⁵² The HR corresponding to the frontier median OS in each year to that of an untreated patient is estimated as the median OS of untreated patient/frontier median OS. The corresponding NNB—termed the frontier NNB for the given year—is then derived from the HR of the frontier treatment with respect to an untreated patient using the formula NNB = 1/(0.5^HR − 0.5), as previously discussed in Methods. For example, in 2017, the estimated HR corresponding to the frontier median OS of 39.7 months = 6/39.7 = 0.15. The corresponding frontier NNB is estimated as = 1(0.5^0.15 − 0.5) = 2.5 patients. Similarly, the frontier NNB for each year in Figure 5B is calculated from the frontier median OS shown in Figure 5A for the corresponding year.

The evolution of the NNB frontier has also been estimated using an approach of compounding HRs over lines of therapy and time, as discussed and illustrated for EGFR+ve patients in the Data Supplement. The Data Supplement discusses the evolution of the NNB frontier estimated using the approach of compounding HRs for the types of patients considered in Figure 4A. The Data Supplement also shows the evolution of NNB frontiers estimated using the approach of compounding HRs over all types of patients considered. The Data Supplement shows the potential treatment sequences for various types of patients—for example, as in Doroshow et al⁵³—over time. The Data Supplement shows the year-by-year NNB frontier for EGFR+ve patients estimated using the approach of compounding HRs.

It can be seen in Figure 5A that the frontier median OS has increased from 9 months in 2003 to 39.7 months in 2017, and Figure 5B that the frontier NNB has reduced from 7.7 patients in 2003 to approximately 2.5 patients in 2017 for patients with NSCLC for the most advanced frontier of any patient type. This improvement is partially correlated with the advent of targeted therapies administered to molecularly enriched populations.

DISCUSSION

Whereas HR is extensively used to assess the approvability of oncologic therapies, it might not be intuitive to patients and health care professionals trying to make decisions in a clinical setting.¹⁰ NNB and NNH are metrics that may be intuitive complements to HR. Moreover, the NNT metric, as it is a function of the absolute reduction of risk, is reflective of the clinical importance of a probability of a specific event—for example, death or nonfatal serious adverse event.¹⁰ As the NNB metric on the basis of time-to-event metrics draws on the same common source of data—Kaplan-Meier curves of OS or PFS—as such metrics as HR or median OS or PFS, NNB can be expected to be largely consistent with HR or median OS or median PFS in assessing the benefit of the experimental drug product over the control.

The analysis in this work shows that NNB and NNH for the selected trials are consistent with the regulatory decisions that are based on these trials. Thus, the communication of clinical trial results and drug product approvability decisions in terms of NNB and NNH, in addition to currently used metrics, such as HR, might enable better communication of the results of clinical trials in practice.

In this work, NNT has been computed for trials with various scenarios: primary end points, therapeutic classes of drug products, lines of therapy, and regulatory statuses of drug products. Whereas in those scenarios—for example, primary end point, line of therapy, etc—NNT provides a quantitative metric of how much better the experimental drug is compared with the control, NNTs might not be comparable across the scenarios. For example, the NNT for OS for a product could be different from the NNT for PFS and the NNT for ORR for a product in the same clinical trial. Therefore, the purpose of NNT estimation—for example, selection of a therapy in a clinical setting—has to be taken into account while assessing the comparability of NNT estimates from various scenarios.

As baseline risk—for example, the proportion of patients who die as a result of the disease at a given point in time since the start of treatment—reduces, the maximum potential ARR also reduces—as the maximum value of ARR is no more than the baseline risk—and, therefore, minimum potential NNT tends to increase. Thus, for relatively small baseline risk, minimum potential NNT tends to be relatively large. For example, if the baseline risk of fatality is 1%, then the maximum potential ARR is 1% and the minimum potential NNT is 100. This property of NNT makes the NNT metric less sensitive to improvements in relative risk—for example, if the risk in the experimental drug arm is 0.5%, then the improvement in relative risk is 50%, whereas ARR is only 0.5%; however, this same property of NNT is also considered to be advantageous when trading the benefit of treatment against the risk of toxicity for small values of baseline risk.¹⁰

As ARR tends to zero, NNT becomes high—tends to infinity—and CI for NNT tends to become wide. This could be interpreted in terms of NNT as the less effective a drug product is compared with its control, the less the certainty that the product is more effective compared with the control.

With regard to the use of NNT for statistical inference, for the case in which NNT is calculated on the basis of the difference in ORR (as described in the Data Supplement), if the difference in ORR is statistically significant—that is, the lower CI of the difference in ORRs is not opposite in sign to the upper CI—then the corresponding NNT is also statistically significant. Similarly, for the case in which NNT is calculated on the basis of HR for time-to-event metrics, such as PFS or OS (as described in the Data Supplement), if the ARR for a given point in time—for example, median PFS or OS in the control arm—estimated on the basis of HR is statistically significant, then the corresponding NNT is also statistically significant.

Despite the limitations discussed above, reporting of clinical trial results including NNT could further enable patient-centric, shared decision making by doctors and patients about treatment options, taking into account the performance of drug products in clinical trials, subpopulation effects on drug product performance, variation of drug product performance over time of treatment administration, and the specific attributes of the patient being treated.

AUTHOR CONTRIBUTIONS

Conception and design: All authors

Administrative support: Janet Woodcock

Financial support: Janet Woodcock

Collection and assembly of data: Gokaraju K. Raju, Sean Khozin, Karthik Gurumurthi, Reuben Domike

Data analysis and interpretation: All authors

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/rwcorascopubs.org/cci/author-center.

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

Gokaraju K. Raju

Consulting or Advisory Role: Pfizer (Inst), Sun Pharma (Inst)

Sean Khozin

Employment: Johnson & Johnson

Leadership: Johnson & Johnson

Stock and Other Ownership Interests: Johnson & Johnson

Travel, Accommodations, Expenses: Johnson & Johnson

No other potential conflicts of interest were reported.