Extracting Electronic Health Record Neuroblastoma Treatment Data With High Fidelity Using the REDCap Clinical Data Interoperability Services Module

Abstract

PURPOSE

Although the International Neuroblastoma Risk Group Data Commons (INRGdc) has enabled seminal large cohort studies, the research is limited by the lack of real-world, electronic health record (EHR) treatment data. To address this limitation, we evaluated the feasibility of extracting treatment data directly from EHRs using the REDCap Clinical Data Interoperability Services (CDIS) module for future submission to the INRGdc.

METHODS

Patients enrolled on the Children's Oncology Group neuroblastoma biology study ANBL00B1 (ClinicalTrials.gov identifier: NCT00904241) who received care at the University of Chicago (UChicago) or the Vanderbilt University Medical Center (VUMC) after the go-live dates for the Fast Healthcare Interoperability Resources (FHIR)–compliant EHRs were identified. Antineoplastic drug orders were extracted using the CDIS module. To validate the CDIS output, antineoplastic agents extracted through FHIR were compared with those queried through EHR relational databases (UChicago's Clinical Research Data Warehouse and VUMC's Epic Clarity database) and manual chart review.

RESULTS

The analytic cohort consisted of 41 patients at UChicago and 32 VUMC patients. Antineoplastic drug orders were identified in the extracted EHR records of 39 (95.1%) UChicago patients and 26 (81.3%) VUMC patients. Manual chart review confirmed that patients with missing (n = 8) or discontinued (n = 1) orders in the CDIS output did not receive antineoplastic agents during the timeframe of the study. More than 99% of the antineoplastic drug orders in the EHR relational databases were identified in the corresponding CDIS output.

CONCLUSION

Our results demonstrate the feasibility of extracting EHR treatment data with high fidelity using HL7-FHIR via REDCap CDIS for future submission to the INRGdc.

A path to unlocking high-quality EHR treatment data for neuroblastoma research using REDCap CDIS.

INTRODUCTION

The International Neuroblastoma Risk Group Data Commons (INRGdc) is a cloud-based ecosystem developed to democratize access to neuroblastoma data and accelerate research.¹ The INRGdc includes patient demographics, biomarkers, outcome data, and clinical trial arm assignment (if applicable) for more than 25,000 patients with neuroblastoma enrolled on cooperative group research studies around the world.¹ The data set includes more than 12,000 patients enrolled on the Children's Oncology Group (COG) neuroblastoma biology study ANBL00B1 (ClinicalTrials.gov identifier: NCT00904241), which serves as a registry, banking protocol, and infrastructure for risk classification.^2,3 The cross-disease Pediatric Cancer Data Commons (PCDC)⁴ houses the INRGdc.⁵ The PCDC Data Portal enables cohort discovery and survival visualization for hypothesis exploration, and patient-level data are available to the research community through an application process. Although the INRGdc has enabled seminal large cohort studies,^6-11 the research has been limited by a lack of real-world electronic health record (EHR) data, including detailed information about treatment.

CONTEXT

• Key Objective

• Is it feasible to extract accurate neuroblastoma treatment data directly from electronic health records (EHRs) using the REDCap Clinical Data Interoperability Services (CDIS) module?

• Knowledge Generated

• This pilot demonstrates that Fast Healthcare Interoperability Resources–based EHR data extraction using REDCap CDIS can successfully derive high-fidelity medication orders from children with neuroblastoma. Given the ubiquity of REDCap, this approach has the potential to accelerate the pace of EHR-based automated data extraction for research.

• Relevance (J.L. Warner)

• While necessarily small due to the rarity of the specific cancer studied, this work demonstrates the utility of the REDCap CDIS module. REDCap users with EHRs should be encouraged to implement the CDIS to support similar work at larger scales.*

• *Relevance section written by JCO Clinical Cancer Informatics Editor-in-Chief Jeremy L. Warner, MD, MS, FAMIA, FASCO.

Traditionally, real-world treatment data are obtained through manual chart review from the EHR. Since manual review can be time- and cost-prohibitive, many research efforts have relied on querying databases with cleaned and transformed EHR data.^12-14 However, variations in data schemas and terminologies used make harmonizing data from EHR systems at different sites challenging. In the context of oncology research, data harmonization refers to the process of integrating and standardizing data from diverse multimodal sources (eg, clinical trials, EHRs, genomic and imaging studies) into a common data schema with controlled terminologies to ensure consistent, compatible, and interoperable data. This type of harmonization is critically important as it provides shared semantic grounding for concepts whose meaning may otherwise be ambiguous or highly context-dependent (eg, stage II can vary considerably in meaning without an additional context or a proper definition).^5,15 Moreover, real-world databases require significant infrastructure and expertise to maintain, which limits participation in EHR-dependent multisite research to larger cancer centers or those with better technical capabilities.

In this proof-of-concept study, we evaluated the feasibility of extracting antineoplastic data from EHRs of patients with neuroblastoma at the University of Chicago (UChicago) and the Vanderbilt University Medical Center (VUMC) using the REDCap Clinical Data Interoperability Services (CDIS) module. CDIS provides seamless data exchange between the EHR and REDCap via a Fast Healthcare Interoperability Resources (FHIR) application programming interface (API).¹⁶ Data access APIs (such as the Epic FHIR APIs used in this study) are required for compliance of EHR systems with the 21st Century Cures Act Final Rule from the Office of the National Coordinator.¹⁷ By representing EHR data in a standardized format, the FHIR standard solves a variety of interoperability challenges. The primary advantage of FHIR is that it allows EHR data to be represented and transferred in an EHR-agnostic fashion, thereby facilitating sharing of health information in electronic form regardless of the EHR used.^18,19 Therefore, if proven to be viable, this data exchange could enrich the data sources in the INRGdc for the research community by enabling automated collection of real-world data at any institution with enabled REDCap CDIS functionality that provides care for patients with neuroblastoma.

METHODS

Antineoplastic drug orders and medication administration records were extracted from EHRs of patients with neuroblastoma at the UChicago and the VUMC who were enrolled on ANBL00B1 (ClinicalTrials.gov identifier: NCT00904241). Medical record numbers (MRNs) and the associated COG participant identifiers (COG ID) for patients were identified at each institution through institutional clinical trials registration records. The study start dates were based on the go-live dates for the current FHIR-compliant EHRs at each institution. The data extraction cutoff date was December 31, 2021. Patient and tumor characteristics required for COG risk group classification² and neuroblastoma treatment data were abstracted manually from EHRs by two of the authors (A.V.D. and D.J.B.). UChicago and VUMC patient characteristics were compared using chi-square tests when any cell count was >5 or Fisher's exact tests when at least one cell count was <5. The study was approved by the institutional review boards at the UChicago, the primary study site, and the VUMC according to the US Common Rule ethical guidelines.

Data Extraction

Medication orders and administration records were extracted from EHRs using the FHIR-based REDCap CDIS module. CDIS-enabled REDCap projects were created, configured to include their respective patient populations (identified by MRN and COG ID), and set to extract all active, completed, on hold, or discontinued medication orders for the included date range. By default, CDIS-extracted data contain all medication orders for a specified date range rather than just the subset of neuroblastoma treatment medications of interest (Appendix Table A1), necessitating postprocessing of the data. For this, a medications of interest list was curated by clinician subject matter experts. The curated list included all chemotherapeutic and immunotherapeutic agents used in COG risk-based clinical trials during the time period covered by this study (2010-2021) that are currently considered the standard of care. Data postprocessing was performed using Python and R code with three primary functions: (1) connecting to the REDCap API and exporting a subset of fields for all records (ie, fields of interest), (2) subselecting medications of interest through free-text matching of medication names, and (3) deidentifying the record set (Fig 1). The code initially extracts records from REDCap through its API, including the following subset of fields: dob, medication_label, medication_date, medication_status, mrn, and cog_id. Once retrieved, the data set is filtered such that only those records where medication_label contains one of the medication labels in the curated list are retained. Finally, the data are deidentified by (1) calculating the age in days at medication order (age_at_medication) through subtracting dob from medication_date and (2) dropping dob and medication_date columns.

FIG 1.

FHIR-based EHR data extraction services using the REDCap CDIS module to identify medications and postprocessing used to identify antineoplastic drug. CDIS, Clinical Data Interoperability Services; EHR, electronic health record; FHIR, Fast Healthcare Interoperability Resources.

To create a benchmark for comparison, the same medication orders and medication administration records for the study cohort were also identified in the EHR relational gold standard databases at each institution (UChicago's Clinical Research Data Warehouse [CRDW] and VUMC'S Epic Clarity database). For the CRDW, data are pulled from a broad range of internal and external data sources, including EHRs, billing records, the cancer registry, the National Death Registry, and laboratory systems. CRDW data elements include patient demographics, laboratory values, procedure and diagnosis codes, medications, and visit information. CRDW data are aligned across all these systems, checked for discrepancies, and harmonized to a consistent data model. Clarity is a reporting database supporting the Epic clinical information system that stores abstracted data that are updated daily from the production data associated with day-to-day workflows. The benchmark medication order data were filtered such that only those records where the medication label contained one of the medication labels in the curated list were retained, and patient age in days at medication order was calculated. The resulting data set included MRN, medication label, age in days at medication order, and order status for only neuroblastoma treatment medications. A schematic comparing the methods required to extract the benchmark data stored in an EHR relational database (blue arrows) versus the data captured via REDCap CDIS (red arrow) for import into a REDCap project is shown in Figure 2.

FIG 2.

Comparison of methods to import data into a REDCap project using EHR relational databases (CRDW or Clarity; blue arrows; stopwatch indicates delay in transfer) versus REDCap CDIS (red arrow). CDIS, Clinical Data Interoperability Services; CRDW, Clinical Research Data Warehouse; EHR, electronic health record; SQL, Structured Query Language.

Data Validation

Data completeness and accuracy were assessed by comparing the REDCap CDIS records with the benchmark records queried through the CRDW or Clarity and manual chart review. The CDIS output was compared with the benchmark medication orders by aligning on MRN, medication label, and age at medication order. Records were considered a match if the following conditions were met: (1) the MRNs were the same, (2) the medication names were the same, and (3) the patient's age in days at medication order was the same (±1 day).

Records for patients with missing or discontinued extracted treatment data were manually reviewed to confirm that antineoplastic agents were not administered during the study dates. Patient risk group was determined by manually abstracting the established biomarkers from the EHR.² To evaluate frontline treatment, antineoplastic drugs included in standard-of-care low-, intermediate-, and high-risk regimens¹ were identified in the CDIS output for each patient (Fig 3). Identification of antineoplastic agents unique to up-front high-risk regimens (including cisplatin) and review of the schedule of drug administration (including topotecan) were used to distinguish high-risk therapy from intermediate- and low-risk treatments. Manual review of records was performed to confirm that the treatment was frontline.

FIG 3.

Schematic of treatment regimens for (A) low-, (B) intermediate-, and (C) high-risk neuroblastomas. GM-CSF, granulocyte-macrophage colony-stimulating factor. ^aBridge therapy: irinotecan, temozolomide, dinutuximab (not considered the standard of care. May be given on the basis of disease response at the end of induction). ^bIrinotecan, temozolomide (not considered the standard of care).

RESULTS

Patient Characteristics

The analytic cohorts comprised 41 patients with neuroblastoma at the UChicago who received care between January 1, 2010, and December 31, 2021, and 32 patients cared for at the VUMC between November 1, 2017, and December 31, 2021. The patient characteristics in each cohort were not significantly different (Table 1). The majority of patients in both cohorts had International Neuroblastoma Staging System,²⁰ stage 4 disease and were classified as high-risk.

TABLE 1.

Clinical and Biologic Characteristics by Cohort

Characteristic	UChicago (n = 41), No. (%)	VUMC (n = 32), No. (%)	P a
Age at diagnosis, months			.2802
<18	13 (31.7)	15 (46.9)
≥18	28 (68.3)	17 (53.1)
Sex			.1360
Female	16 (39.0)	19 (59.4)
Male	25 (61.0)	13 (40.6)
Race			.5853b
Asian	2 (5.1)	2 (6.5)
Black	5 (12.8)	4 (12.9)
White	29 (74.3)	25 (80.6)
More than one race	3 (7.7)	0 (0)
Unknown	2	1
Ethnicity			.2404b
Hispanic or Latino	7 (17.5)	1 (4.2)
Not Hispanic or Latino	33 (82.5)	23 (95.8)
Unknown	1	8
INSS stage			.5735b
1	3 (7.3)	4 (12.5)
2 (A or B)	5 (12.2)	3 (9.4)
3	7 (17.1)	4 (12.5)
4	26 (63.4)	19 (59.4)
4S	0 (0)	2 (6.2)
Histology			.5008
Favorable	11 (28.2)	12 (38.7)
Unfavorable	28 (71.8)	19 (61.3)
Unknown	2	1
MYCN status			.7476
Amplified	10 (26.3)	6 (20)
Nonamplified	28 (73.7)	24 (80)
Unknown	3	2
Ploidy			.2697
Diploid	9 (25.0)	9 (42.9)
Hyperdiploid	27 (75.0)	12 (57.1)
Unknown	5	11
COG risk classification (2006)			.3950b
Low	4 (9.7)	7 (21.9)
Intermediate	10 (24.4)	7 (21.9)
High	27 (65.9)	18 (56.2)

Abbreviations: COG, Children's Oncology Group; INSS, International Neuroblastoma Staging System; UChicago, University of Chicago; VUMC, Vanderbilt University Medical Center.

^aChi-square test unless otherwise specified. Unknown categories not included in P value calculations.

^bFisher's exact test.

EHR-Extracted Antineoplastic Drug Orders

Medication orders and administration records for antineoplastic agents were identified in the CDIS output for 38 (92.6%) of the 41 UChicago patients and 26 (81.3%) of the 32 patients in the VUMC cohort. Manual review of the data confirmed that patients with missing administration records did not receive chemotherapy or immunotherapy during the timeframe of the study. Three patients were observed without any treatment; five were treated with surgery alone, and one completed high-risk chemotherapy treatment at an outside hospital.

High Fidelity of Treatment Data Extracted Through REDCap CDIS From EHRs

To validate the accuracy of the antineoplastic drug data extracted through REDCap CDIS, patient-level medication lists were compared with those identified in the institutional EHR relational databases. At the UChicago, of the 1,859 antineoplastic drug orders identified from the CRDW, 1,845 (99.3%) corresponded to orders in the REDCap CDIS records. Further investigation revealed that all 14 missing orders corresponded to one patient. For this analysis, the patient's age in days at medication order was classified as matched to the CRDW record if the date in extracted EHR data was ±1 day. This range in age was used because the age at medication order was 1 day greater for 11.7% of the medication orders in the CRDW record compared with the corresponding CDIS-extracted record for a subset of patients. At the VUMC, of the 1,595 orders for antineoplastic agents identified from Clarity, 1,591 (99.7%) orders were identified in the CDIS output. Three of the four missing orders corresponded to one patient.

Frontline Risk-Based Neuroblastoma Treatments

All antineoplastic agents administered during the timeframe of this study, including frontline regimens and, in some cases, relapsed therapy, were included in the REDCap CDIS records. To specifically evaluate frontline treatment, we determined the risk classification for each patient using established biomarker data manually abstracted from the EHR and compared the CDIS record medication list with the drugs used in standard-of-care risk-based treatment regimens (Fig 3). Although some antineoplastic agents are included in low-, intermediate-, and high-risk therapies, a subset of unique agents (including cisplatin) was used to distinguish high-risk from non–high-risk regimens. We also evaluated the schedule of topotecan administration to discriminate high-risk from non–high-risk treatment. In COG high-risk regimens, topotecan is included in cycles 1 and 2 of induction therapy,²¹ whereas the agent is administered in a subset of intermediate-risk patients after cycle 8.²² In the combined UChicago and VUMC cohorts, 44 of the 45 high-risk patients were treated with high-risk regimens. The single exception was a patient who received high-risk treatment at an outside hospital. Among the 17 intermediate-risk patients, 15 received drugs that are used in standard-of-care intermediate-risk regimens. One of the two exceptions was an infant initially classified as intermediate-risk who did not receive any chemotherapy after developing a stage 4S pattern of metastatic disease.²³ Another patient was treated with high-risk therapy on the basis of physician preference. Among the 11 low-risk patients, three received immunosuppressive therapy with chemotherapy to treat opsoclonus-myoclonus-ataxia syndrome (OMAS).²⁴ Another low-risk patient was initially treated with chemotherapy as part of immunosuppressive therapy for OMAS in addition to surgery at an outside institution. This patient subsequently transferred to UChicago when she developed disease progression and was treated with second-line high-risk therapy. Seven low-risk patients were treated with surgery alone or were observed without therapy.

DISCUSSION

In this study, we successfully deployed FHIR-based EHR data extraction services using the REDCap CDIS module to derive EHR chemotherapy and immunotherapy data, including drug names and administration dates, from patients with neuroblastoma at two institutions. Among the 73 patients analyzed, antineoplastic drug orders were extracted from the EHRs of 64 (87.6%) patients. The accuracy of the extracted EHR administration orders was validated by manual chart review, which confirmed that none of the nine patients with missing or discontinued antineoplastic drug orders received chemotherapy and/or immunotherapy during the study dates. Furthermore, patient line-level comparisons between the benchmark antineoplastic medication records queried through the CRDW or Clarity and the corresponding institutional CDIS output demonstrated high fidelity.

Frontline treatment for children with neuroblastoma is stratified according to risk group assignment.^1,25 High-risk patients receive intensive multimodal therapy,^21,26 and intermediate-risk patients are treated with moderate-dose chemotherapy regimens,²² whereas low-risk patient are either observed or treated with surgery alone.^25,27 As expected, the CDIS output demonstrated that 44 (97.8%) of the 45 high-risk patients received up-front high-risk therapy, 15 (88.2%) of the 17 intermediate-risk patients received chemotherapeutic agents that are standard for intermediate-risk regimens, and no antineoplastic agents were identified in CDIS records for eight of the 11 low-risk patients.

Real-world data derived from EHRs include information about routine clinical care, providing a valuable data source for medical research beyond the confines of traditional clinical trials. A number of approaches for extracting EHR data are being developed including oncology-specific FHIR implementation guides (eg, mCode), third-party SMART-on-FHIR applications available to clinicians via EHR-hosted app galleries which provide custom point-of-care functionality not available in the standard EHR installation, CRDWs that aggregate data from multiple source systems and present them in bespoke or common data models (eg, Observational Medical Outcomes Partnership common data model oncology extension), and custom software packages that map from local terminologies to standards for aggregation (eg, ExtractEHR). A significant advantage of the REDCap CDIS module for EHR extraction is the ubiquity of the adoption of REDCap. As of June 2023, it has been used for 1.8 million projects at 6,500 institutions in 154 countries and has been cited over 32,000 times (REDCap²⁸).²⁹ The Health Insurance Portability and Accountability Act (HIPAA)–compliant REDCap CDIS module provides a scalable solution for data extraction using the widely supported and EHR-agnostic HL7 FHIR standard, meaning that this solution to EHR extraction can be used with any major EHR. Furthermore, REDCap CDIS includes built-in security, including support for OAuth2,³⁰ supports the HL7 FHIR standard,³¹ is HIPAA-compliant, and can be configured to adhere to 21-CFR part 11 compliance for submission to regulatory agencies.

A limitation of this study is that data were extracted from a small number of patients from two institutions. In addition, both institutions used the same EHR system (Epic). Although the feasibility of expanding this methodology to additional institutions with other EHR systems remains to be tested, the CDIS module has been used to extract data from other major EHR systems.¹⁸ However, a limitation of this data extraction approach is that REDCap CDIS can only access FHIR resources and profiles as implemented by EHR vendors. Thus, it may not be possible to extract data at institutions with EHR systems that do not have access to FHIR resources and profiles. Even among supported data elements, not all available useful information is passed to REDCap CDIS by local EHRs. For example, our experience demonstrated that the current versions of the EHRs used at the UChicago and at the VUMC do not support the MedicationAdministration FHIR resource, which is a much more reliable source of drug administration than medication orders. Even for resources that are supported, extracting standardized RxNorm codes required a subquery, which constrained our ability to leverage data standards to identify and categorize medications unambiguously. While installation of the module is technically straightforward and supported by thorough documentation, it requires collaboration between research and clinical information technology teams and adds to information technology support overhead. Furthermore, raw EHR data must be processed into an analytically usable form as granular row-level medication order often includes details (eg, canceled orders) that are not generally useful for understanding the treatment received. In oncology, this is highly relevant as patients are typically treated with multiagent regimens that consist of drug combinations administered at a scheduled sequence and interval, often with unexpected and unpredictable delays or modifications on the basis of toxicities or logistical constraints.

The task of inferring an up-front or relapse treatment regimen from a list of chemotherapy administration records requires knowledge of available treatment regimens, the time frame of drug administration, and the clinical status of the patient. A comprehensive, up-to-date catalog of treatment regimens relevant to pediatric oncology such as that being developed by HemOnc.org³² may serve as a useful reference for identifying specific drugs patients receive on or according to clinical trials. Machine learning clustering techniques may prove to be an ideal method for labeling the treatment regimen a patient has received. For pediatric patients with acute myeloid leukemia, the feasibility of using machine learning to identify treatment courses has been demonstrated using an administrative database,³³ and other methods such as ExtractEHR have been used to identify chemotherapy toxicity and adverse events on the basis of laboratory results extracted from the EHR.^34,35

We are currently expanding the collection of neuroblastoma treatment data to other COG institutions that have implemented REDCap CDIS. In the future, we plan to submit the extracted EHR treatment data to the INRGdc using the COG Data and Statistical Center as the honest broker, a strategy that is already in place for transferring COG data to this data commons. The availability of detailed treatment data in the INRGdc has the potential to transform our understanding of the impact of specific therapeutics on patient survival. Analyses of these data may also identify drugs that are effective in rare subsets of patients with tumors that harbor molecular aberrations. Large cohorts will be needed to conduct these types of studies, and our long-term goal is to extract EHR treatment data from institutions around the world to expand the data in the INRGdc. Similar to National Cancer Institute's (NCI) Genomics Data Commons,³⁶ the aims of the INRGdc are to democratize access to data and foster sharing of these data to ultimately improve the outcome of children with neuroblastoma.

Strategies using real-world EHR-extracted data to develop adult cancer registries for research are being leveraged by companies including Flatiron Health and CancerLinQ.^37,38 For pediatric patients, the NCI's Childhood Cancer Data Initiative is focused on systematically collecting and sharing data from every child, adolescent, and young adult with cancer to accelerate the development of new and more effective therapies and improve patient outcomes.³⁹ This and other pediatric cancer initiatives, such as the PCDC,⁵ will require scalable methods for data collection, which are enabled by data standards. As the value of international cross-institution data sharing is increasingly recognized, scalable methods to extract data of interest are vitally important. REDCap CDIS provides a plug-and-play mechanism for easily extracting clinical data of interest from a user-friendly interface. In this study, we have demonstrated the feasibility of extracting medication records of patients with neuroblastoma using REDCap CDIS. Because additional EHR data (eg, laboratory results) can be extracted using the CDIS module, this approach can be used to evaluate real-world data beyond treatment. Just as REDCap has transformed the practice of clinical research by facilitating standardized data capture,⁴⁰ we anticipate that REDCap CDIS will accelerate the pace of EHR-based automated data extraction for research ecosystems including the INRGdc, enabling new discoveries that could further improve the outcome for children with neuroblastoma.

APPENDIX

TABLE A1.

Antineoplastic Agent Names Used in Filtering Medication Orders From REDCap CDIS

Agent Name
Cyclophosphamide
Topotecan
Etoposide
Cisplatin
Doxorubicin
Vincristine
Busulfan
Melphalan
Carboplatin
Isotretinoin
Vincristine
Busulfan
Melphalan
Carboplatin
Isotretinoin
Dinutuximab
Sargramostim
Aldesleukin
Aldesluekin
Irinotecan
Temozolomide
Iobenguane
Thiotepa
Accutane

Abbreviation: CDIS, Clinical Data Interoperability Services.

AUTHOR CONTRIBUTIONS

Conception and design: Brian Furner, Alex Cheng, Daniel J. Benedetti, Debra L. Friedman, Samuel L. Volchenboum, Susan L. Cohn

Financial support: Susan L. Cohn

Administrative support: Susan L. Cohn

Provision of study materials or patients: Alex Cheng, Debra L. Friedman, Susan L. Cohn

Collection and assembly of data: Brian Furner, Alex Cheng, Ami V. Desai, Daniel J. Benedetti, Debra L. Friedman, Susan L. Cohn

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/rwc or ascopubs.org/cci/author-center.

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