This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR Research Protocols, is properly cited. The complete bibliographic information, a link to the original publication on https://www.researchprotocols.org, as well as this copyright and license information must be included.
Compared with White patients, minoritized patients (Black and Hispanic patients) have a higher incidence of advanced solid cancers and have a higher mortality. These patients also report poor patient-centered communication and worse pain assessment and management. Although many factors contribute to these disparities, physician implicit bias may be a contributing factor.
The primary goal of this study is to evaluate the role of implicit bias among oncologists and examine the impact on racial or ethnic differences in objective assessments of communication with minority patients with advanced cancer.
To accomplish this goal, we plan to recruit 65 oncologists and 325 patients (5 patients per oncologist) with advanced solid cancer from ambulatory cancer clinics within the diverse settings of the Mount Sinai Health System in New York City and the Duke University Health System in Durham, NC. We audio record patient-oncologist encounters during a postimaging visit, with 3 encounters for each of the patients. We will analyze the recorded visits and compare the patient-centered communication content of these conversations. Immediately after the recorded visit (no more than 2 weeks later, in order to minimize recall bias), patients are required to complete a follow-up survey to evaluate patient-centered outcomes. A 3-month follow-up survey is used to assess pain levels and control, use of analgesics, and psychological distress. A 6-month follow-up survey is used to assess psychological distress. We administer the Implicit Association Test to oncologists to assess their level of implicit bias toward patients who identify as Black or Hispanic after we finish recording patient encounters.
Funding from the National Cancer Institute was received in March 2021. Patient and oncologist recruitment began in March 2022. We have recruited all 65 oncologists in the study, and patient recruitment is ongoing. The study team plans to continue to enroll patients until March 2025. As of December 2024, we have enrolled 245 patients. We expect to publish the findings in October 2026.
In this paper, we outline the study methods, describe the development of a codebook to assess pain conversations being used to evaluate primary and secondary outcomes, and discuss challenges and lessons learned throughout the study.
DERR1-10.2196/66086
Compared with White patients, minoritized patients (Black and Hispanic patients) have a higher incidence of advanced solid cancer and have a higher mortality rate [
The impact of implicit bias on the outcomes of ethnic minority patients with advanced cancer may be associated with different mechanisms. Minority patients have reported poor patient-centered communication (eg, respecting and responding to patients’ wants, needs, and preferences), which can influence assessments, recommendations, and outcomes [
This study has been reviewed by the Icahn School of Medicine Protection of Human Subjects/Institutional Review Board (PPHS/IRB) and the Duke University Health System (DUHS) IRB, and it has been determined that the study procedures are in accordance with ethical standards for experiments on human subjects (Study-21-00396/IRB Pro00108633). We obtain informed written consent using an IRB-approved informed consent document. Subjects are free to withdraw from participation in the study at any time upon request. To preserve subject confidentiality, subjects are assigned coded study ID numbers. Using these study ID numbers, none of the collection forms will contain the names or medical record numbers of the subjects or other personal identifiers. We pay participants for their time and effort (enrollment [visit 1]: US $30, visit 2: US $15, and visit 3: US $15).
This study is a prospective 2-site observational trial that examines postscan encounters between oncologists and their patients with advanced solid cancer who have a prognosis of less than 2 years. We plan to enroll 65 oncologists and approximately 325 patients (4-7 patients per oncologist; with relatively equal proportions of those who identify as White, Black, and Hispanic) seen at outpatient cancer clinics at the Mount Sinai Health System (MSHS) and DUHS. We collect up to 21 audio recordings of encounters per oncologist, with 3 encounters for each of the 4-7 patients, and code the audio-recorded encounters.
Study recruitment has occurred in waves over the past 3 years. We have introduced oncologists to the study and approached them during clinician meetings or individually. Oncologists are eligible if they provide care to patients with advanced solid cancer at MSHS or DUHS for at least a half-day per week and will continue practicing at either facility for at least 1 year.
Before patient recruitment, the study staff had visited clinics to introduce the study to the clinic staff. This study includes patients with advanced solid cancer who are 21 years or older; are on active treatment (ie, chemotherapy, targeted therapy, and immunotherapy); identify as Black/African American, White, or Hispanic; are English-speaking; and are scheduled for a postscan visit. Patients are excluded if they are diagnosed with more than one malignancy (except for nonmelanoma skin cancer) in the last 5 years. Patients are identified via the electronic health record (EHR; both sites use Epic) from the schedules of consented oncologists and are sent a letter from their oncologist describing the study, with an opt-out number for declining to participate in the study. We attempt to recruit equal numbers of White, Black, and Hispanic patients.
We obtain oncologists’ consent either through paper consent in the clinic by the study staff or via email through REDCap. We collect oncologist baseline surveys (see
After sending recruitment letters, we call, email, or text potentially eligible patients to screen them. After the staff determine that patients are eligible (see
Our primary outcome is patient-centered communication, which we assess from coding audio-recorded encounters, open questions, reflective statements, empathic responses to empathic opportunities, and eliciting questions. Our secondary outcome involves discussions of pain, use of guideline-concordant pain management, and pain control, as well as satisfaction and psychological distress among patients with advanced solid cancer. We assess this through coding discussions of pain and goals of care conversations in audio-recorded encounters and postencounter surveys completed by patients after their postscan visit.
To assess if implicit bias causes, in part, racial disparities in patient-centered communication, we use multilevel structural equation modeling, which allows for simultaneously examining several regression-type relationships while accounting for latent variables. Specifically, we assess models of complete versus partial mediation by comparing constrained models (direct effects constrained to zero) with unconstrained models that allow for partial mediation [
We use 3 established codebooks and 1 newly developed codebook with definitions and examples for each code. These codes include measures of patient-centered communication, goals of care, and pain. Four teams of 2 coders each code all transcripts for our primary and secondary outcomes. Coders listen to the audio-recorded encounters between the oncologists and their consented patients to record instances of goals of care conversations, discussions of patient-perceived pain, discriminatory behaviors, patient participatory behaviors, and effective communication skills (eg, open questions and empathic statements). All 8 coders are trained by the investigator through reviewing audio recordings until interrater reliability is achieved. Interrater reliability is achieved by ensuring that each coder is coding with at least 80% similarity to the other coder on their team. All teams are monitored regularly by the investigators.
The first codebook includes aspects related to our primary outcome (open questions, reflective statements, empathic responses to empathic opportunities, and eliciting questions) [
To detect the effect size (relative risk of 0.80; proposed 20% difference) and assume an equal size of both groups, we need approximately 323 patients to find differences in our primary outcome, effective communication, which involves open-ended questions, reflective statements, empathic responses to empathic opportunities, and eliciting questions. We assume that enrolled patients of the same physician will be correlated, and the level of intraphysician or intracluster correlation will be low to moderate. We assume a 20% difference for the primary outcome of the 4 communication behaviors (aim 1) between minorities (Black and Hispanic people) and nonminorities based on our pilot and the literature demonstrating similar differences in patient-centered communication for Black and Hispanic people [
This study received funding from the National Cancer Institute in March 2021. Patient and oncologist enrollment began in March 2022. We have recruited all 65 oncologists in the study, and patient recruitment is ongoing. Due to a lack of diversity among oncologists at both MSHS and DUHS, almost all enrolled oncologists identify as non-Hispanic White. There are 8 clinics located at DUHS and 8 at MSHS: gastrointestinal, sarcoma, melanoma, neuro-oncology, thoracic, breast, genitourinary, gynecologic, and head and neck. The study team plans to continue to enroll patients until March 2025. As of December 2024, we have enrolled 245 patients. We expect to publish the findings in October 2026.
We anticipate that we will find that implicit bias among oncologists causes racial and ethnic differences in patient-centered communication and is associated with higher levels of psychological distress and lower levels of satisfaction with communication. We also anticipate that we will find that implicit bias among oncologists causes racial and ethnic gaps in pain assessment, which can lead to poor perception of pain control among minority patients with advanced solid cancer. Our findings will deepen our understanding of relationships among implicit bias, communication processes, management, and patient outcomes. This study is innovative because we assess implicit bias among oncologists who treat Hispanic patients and examine the goals of care over time using multiple postscan visits and audio recordings. No other team has assessed how the goals of care conversations evolve over multiple encounters. We outline the limitations and solutions below.
We planned to obtain 3 audio recordings per patient to assess how the goals of care conversations change over time as the disease potentially progresses. However, we encountered difficulties in recording 3 encounters as some patients entered hospice or died.
We assessed the goals of care conversations at the first audio-recorded visit for all participants and realized that we would only be able to examine changes over time in a subset of patients. Given the prevalence of patients becoming too sick to continue participating, we are not replacing recordings.
Staff at the MSHS experienced challenges identifying eligible participants, specifically those with advanced cancer coming for a postscan visit, as the data were not readily available in the patient’s EHR. Unlike the DUHS, where the EHR includes stage information, at the MSHS, the EHR does not include this information, and thus, a labor-intensive manual review of all scheduled patients is needed. The manual review not only introduces a considerable workload for the staff but also poses a risk of oversight or delays in identifying eligible participants. The complexity of this task is compounded by the inherent variability in how medical information is recorded and cataloged within the EHR, making the identification of specific criteria, such as cancer stage, a labor-intensive and error-prone process.
To address this challenge, we are implementing a machine learning–based solution leveraging large language modeling. This approach offers several advantages over manual review. First, it eliminates the need for labor-intensive manual chart reviews, reducing the workload for staff and mitigating the risk of oversight or delays in identifying eligible participants. Additionally, by leveraging natural language processing techniques, the model can account for the inherent variability in how medical information is recorded and cataloged within the EHR, enhancing accuracy and efficiency in identifying specific criteria such as cancer stage. Furthermore, by integrating this solution into existing workflow systems, such as appointment scheduling software, we can automate the identification process, providing real-time eligibility screening for scheduled patients. This not only streamlines the recruitment process but also enables proactive identification of eligible participants, allowing for timely intervention and enrollment in the research protocol.
Oncologists at both sites had postscan conversations with advanced practice providers (nurse practitioners and physician assistants) and oncology fellows in practice. During instances where advanced practice providers and fellows see patients along with the consented oncologists, the oncologists might not say much in the encounter. Having multiple health care professionals in these encounters impedes the study’s objective of specifically recording conversations between patients and their oncologists who are enrolled in the study.
When possible, the study coordinator emails oncologists in advance of the scheduled encounter. Often, oncologists agree to see the patients for the postscan conversation instead of their advanced practice providers or fellows. Third-party consents are obtained from advanced practice providers and fellows who may be a part of the recorded conversation.
Our goal is to recruit 5-6 patients per oncologist. The DUHS currently has an equal number of Black and White patient participants recruited for the study owing to a low number of Hispanic patients who meet the eligibility criteria. Some oncologists see more Black patients than White patients, while others see more White patients than Black patients. At the MSHS, there is a higher number of patients who identify as Hispanic.
Due to the low number of eligible Hispanic patients at the DUHS, we are oversampling Hispanic patients at the MSHS. At the DUHS, the study team has been recruiting 3 Black patients and 3 White patients per oncologist, where possible. There have been some oncologists who see more Black patients and those who see fewer. We compensate for this by overrecruiting Black patients for some oncologists.
This is one of the first studies to examine the goals of care over time in a diverse sample of patients with advanced solid tumors. A more in-depth understanding of the reasons underlying the disparities in care will be an important step toward improving the outcomes of over 65,000 minority patients diagnosed with advanced cancer each year. The knowledge generated by our project could inform both oncologist-level and patient-specific interventions.
This study aims to evaluate the role of oncologists’ implicit bias in racial and ethnic disparities in patient-centered communication, psychological distress, satisfaction with communication, pain assessment, and pain management among patients with advanced solid cancer. By collecting one of the largest datasets of conversations between oncologists and Black and Hispanic patients to date, this study lays the groundwork for developing interventions aimed at improving communication and care delivery. The significance of this study lies in its potential to deepen our understanding of the complex relationships among implicit bias, communication processes, management strategies, and patient outcomes. The findings generated by this project have the potential to inform both physician-level educational programs and patient-specific activation interventions, ultimately contributing to the reduction of disparities in cancer care.
In summary, this study has far-reaching implications for advancing equitable cancer care and underscores the importance of addressing implicit bias in health care delivery. By identifying and addressing barriers to effective communication and care, we can strive toward achieving better outcomes for all patients, regardless of race or ethnicity.
Clinician baseline survey.
Clinician postencounter survey.
Patient prescreener survey.
Patient baseline survey.
Patient postaudio survey.
Duke University Health System
electronic health record
Implicit Association Test
institutional review board
Mount Sinai Health System
This work was supported in part through computational and data resources and staff expertise provided by Scientific Computing and Data at the Icahn School of Medicine at Mount Sinai and supported by a National Cancer Institute grant (1R01CA251547-01A1) from the National Institutes of Health.
This study is being conducted in accordance with the following publication and data sharing policies and regulations:
National Institutes of Health (NIH) Public Access Policy, which ensures that the public has access to the published results of NIH-funded research. It requires scientists to submit final peer-reviewed journal manuscripts that arise from NIH funds to the digital archive PubMed Central upon acceptance for publication.
Writing – reviewing & editing: VCKD, KIP, CBS, BNR, ANA, RMT, MLA
Conceptualization: CBS, KIP
Writing – original draft: VCKD
Supervision: CBS, KIP, JAM, VCKD
Funding acquisition: CBS
Project administration: CBS, KIP
CBS has received speaker fees from GSK and AstraZeneca and honoraria from Elevance Health. All other authors have no conflicts of interest to disclose.