Targeting Bias to Reduce Disparities in End of Life Care

Significance

Scope of the problem:

US adults over 65 are more racially diverse than ever, and this trend will continue. The number of older Black Americans is projected to reach 12 million or 12% of older adults by 2060. Although the growth of palliative care has improved the quality of end-of-life care over the past two decades, racial and ethnic minority patients have not fully benefitted from these advances. There are well-documented disparities in access and uptake of palliative care and in the quality and intensity of end-of-life care. These inequalities affect many minority groups, each with its unique history, language and culture that may influence the underlying mechanisms of disparities. The current proposal focuses on Black Americans who experience disparities that are deeply rooted in complex historical and present-day interactions with the health care system.

Mechanisms underlying disparities in quality of end-of-life care

Clinician factors:

There is a major gap in knowledge regarding clinicians' contributions to end-of-life disparities. Most physicians in the US are White, and only 4% are Black. Patient-physician racial discordance is associated with decreased quality of communication. Clinician communication during racially discordant encounters is an important and modifiable factor in end-of-life care disparities.

End-of-life conversations are less likely to result in goal-concordant care for Black compared with White patients. Possible explanations for these differences include implicit bias as well as explicit beliefs. Stereotypical beliefs hinder development of true knowledge of the patient by shortcutting around knowledge-gathering communication. Implicit attitudes decrease the quality of communication affecting partnership, respect, affiliation and trust. Reduced access to quality communication could explain some of the increased use of aggressive care and reduced quality of end-of-life care for Black patients.

Bias and Communication:

Implicit bias refers to unconscious, automatic positive or negative attitudes. Clinicians rarely mention race and ethnicity as barriers to communication, suggesting that they are unable to name these barriers due either to the implicit nature of the bias or to social desirability bias in reporting. The Implicit Association Test (IAT) demonstrates unconscious racial bias among clinicians. Implicit bias predicts patient ratings in domains critical to quality end-of-life care including interpersonal treatment, communication and trust. Physician implicit bias results in poor communication patterns, and poor patient-doctor relationships. Some physicians express explicit beliefs regarding Black patients' "resistance" to hospice care. It is unknown whether explicit stereotypes about end-of-life preferences influence communication behavior.

Bias Mitigation:

Evidence-based strategies have been developed to reduce effects of implicit bias, but few have been systematically tailored to gaps in care and measured for effectiveness in the clinical setting. There is limited evidence to support specific strategies to reduce the effects of clinician bias on clinical care. Strategies that focus only on making clinicians aware of their biases are unlikely to reduce their effects in the long term without providing them with specific strategies to overcome them. A strategy based on transformative learning theory and incorporating critical reflection, guided dialogue, perspective taking exercises, role plays and strategy development has been successfully used with medical students.

Aims/Objectives:

The aim of this study is to assess the effects of clinician racial bias on end-of-life communication and to develop interventions to reduce the effects of implicit bias on quality of clinician communication. There are two study objectives: (1) Establish which communication behaviors in the end-of-life setting are most affected by clinician implicit racial bias and explicit stereotyping. (2) Pilot a theory-driven intervention to improve clinician communication and reduce disparities in quality end-of-life care. These objectives will be met in two study phases.

Methods/Design:

Phase 1:

Methods:

(1) The investigators will develop a case depicting a hospitalized 72-year-old with lung cancer and sepsis due to pneumonia on high-flow nasal cannula with my mentorship team of experts in palliative medicine and critical care. The investigators will design a clinical encounter with the caregiver to discuss goals of care. 2) The caregiver encounter will be piloted with critical care and hospitalist physicians at Montefiore to ensure realism and achieve standardization. Black standardized caregivers will be recruited from the pool of actors used by the Einstein Clinical Skills Center. (3) Standardized caregivers will be trained in the use of a communication assessment tool. (4) 50 physicians will be recruited to participate in a videotaped high-fidelity simulation of the encounter. The session will be followed by a 1-hour communication skills training. (5) Immediately after the simulation subjects will complete the IAT and a questionnaire probing for attitudes about race and end-of-life care along with demographic questions. The investigators will assess training needs by adapting a previously developed questionnaire measuring physicians' self-identified serious illness communication training needs by incorporating questions about bias mitigation skills. The IAT is publicly available here: https://implicit.harvard.edu/implicit/selectatest.html.

COVID-19 Concerns:

The Clinical Skills Center will be holding in-person simulations with standardized patients for the purposes of medical education starting in May 2021. The simulations in this study will not begin until after that time and will involve less risk than typical simulations because they will not entail a physical examination of the standardized caregiver. Actors and participants will be equipped with personal protective equipment including surgical masks. Trainers and participants in the training session will be required to wear provided surgical masks and remain six feet apart during the training sessions. If COVID-19 state or institutional guidelines prohibit these types of encounters at the time the simulations will be launched, the simulations and training will be adapted to a virtual format and nonverbal behavior coding will be adjusted. These changes will be submitted to the Institutional Review Board (IRB) as an amendment.

Data collection:

Encounters will be videotaped. The amount of time the physician and standardized caregiver speak will be recorded. A five-point verbal communication rating scale will be used to rate physician (1) informativeness, (2) supportiveness, and (3) partnership-building and a modified version of the Nonverbal Accommodation Analysis System (NAAS). The NAAS will be modified to include physical distance and open/closed body language which were shown to be important in serious illness encounters.

Another questionnaire will be administered at the end of the training. The investigators will ask physicians to rate their experiences with different educational modalities presented in the training session, such as feedback from observed encounters, communication drills and role-plays. In addition, the questionnaire will include the cognitive and affective empathy portions of the Interpersonal Reactivity Index. For participants in the active group only, several questions will assess the illusion of ownership of the virtual body in the VR experience. Open-ended questions will be used to gather formative information that the researchers may not have considered. This questionnaire will be distributed at the end of the communication skills training.

Analysis plan:

Bivariate associations of communication scores with IAT scores will be estimated with Pearson or Spearman rank correlation coefficients depending on the distribution of the data. Generalized linear models (GLM) will be fit to the data to assess the associations of the IAT and racial attitudes with primary and secondary communication outcomes, while controlling for potential confounding variables including clinician age, gender and specialty. The identity link will be used in the GLM for continuous communication outcomes. Subgroup analysis will be performed in Asian, Black and Hispanic physicians if there are sufficient numbers in these groups to assess heterogeneity of the associations by race.

Power analysis:

The target sample size of 50 participants would have >80% power to detect a minimum correlation between IAT score and verbal dominance of ρ ≥ 0.38 using a two-sided hypothesis test of a single correlation with a significance level of 0.05. This correlation is similar in magnitude to associations observed in prior studies; B=11.0 in one study and r = 0.32 in another.

Phase 2:

Methods:

This is a communication training session based on a culturally-based program developed with rural, southern Black patients and families and modified for an urban, northern population. A strategy of bias mitigation successfully used with medical students will be adapted for practicing clinicians using results of phase 1. This strategy is based on transformative learning theory and incorporates critical reflection, guided dialogue, perspective taking exercises, role plays and strategy development. In this study, a virtual reality (VR) experience (https://michigan.it.umich.edu/news/2023/07/31/education-andawareness- in-vr-new-experience-illustrates-the-effects-of-racial-discrimination/) will be used as the disorienting dilemma to spark the process of transformative learning. This brief experience embodies the participant in the experiences of an African American man over their life course. It is aimed to engender empathy and perspective taking. Following the VR experience, participants will debrief with the facilitator in small groups, discuss how implicit bias impacts the clinical encounter, brainstorm bias mitigation techniques and role-play strategies to mitigate bias in the clinical encounter. If specific communication behaviors are found related to bias and stereotyping in phase 1, these will be discussed and targeted using these techniques. Otherwise, these techniques will be used to address racial bias generally. The intervention will be incorporated within the communication training session.

Design:

Clinicians will be randomized 1:1 to the active intervention or a control communication training without bias mitigation techniques. Allocation concealment will be in place to ensure the individual enrolling the subject into the study has no a priori knowledge of group assignment. Block randomization will occur with randomly mixed block sizes of 2, 4, and 6. The allocator (research assistant) will hide block size from the executor (PI) in order to prevent the executor from predicting the next assignment. Randomization will be carried out by having a piece of paper that has the phrase "Intervention (Communication + Bias)" or "Control (Communication only)" placed inside an opaque envelope. The outside of the envelopes will be labeled with the sequence number. After a subject has been enrolled into the study and consented, the next sequence numbered envelope on the stack will be opened to determine the study group that the subject will enter.

Participants will be videotaped during high-fidelity simulations of encounters with caregivers described in phase 1, before and after receiving the intervention or control. The intervention and simulations will take place in one half-day workshop.

Study population and Recruitment:

A new group of intensivists, oncologists and hospitalists will be recruited. Physician subjects will be recruited face-to-face by the PI, through departmental email blasts and emails to contacts of the study team. As in phase 1, physicians will complete a screening questionnaire to exclude those that do not routinely encounter hospitalized patients with a life expectancy of less than one year. Potential subjects will be asked about the frequency and timing of previous communication and bias training. Physicians practicing Hospice and Palliative Medicine will be excluded because of their extensive prior training in communication.

Data collection:

Encounters will be videotaped. The investigators will use the five-point verbal communication rating scale and a modified version of the NAAS for nonverbal communication behaviors during the standardized encounter as described in phase 1 above. A research assistant who did not participate in the training sessions and is blinded to participant group will code communication behaviors. Subjects will complete the IAT and the questionnaire described in phase 1 immediately after the first encounter.

Analysis plan:

Initially, the mean change in verbal and nonverbal communication scores pre- and post-intervention will be evaluated within each intervention arm using paired t-tests or Wilcoxon tests if the data are not normally distributed. To assess whether communication scores improved more in the bias mitigation treatment compared with communication training only control, a generalized linear model will be fit to the data with post-intervention score as the outcome, and pre-intervention score and intervention arm as the main effects. In addition, the investigators will explore whether pre-intervention level of bias modifies the effect of bias mitigation training, e.g., those with more bias will have a greater difference in improvement between the intervention and control group, by including in the model an interaction term between IAT score and treatment group on communication scores. A positive interaction will indicate that implicit bias as measured by the IAT modifies the effect of treatment group on communication score. With 25 subjects per intervention arm, the study will have 80% power to detect an effect size of 0.8 between groups in pre-post change in communication scores with a two-sided Type 1 error rate of 5%. While this is considered a large effect size according to Cohen's criteria, this is a pilot study in which the primary goal is to assess feasibility and to generate preliminary data on the effect of the proposed communication training intervention for clinicians to mitigate the effects of implicit bias.

Data Management/Analysis:

Communication behavior codes and data from the questionnaires and IAT stored on Montefiore's secure Box drive and will only be accessible by members of the research team. Video recordings will be stored on a password-protected hard drive accessible only to the research team.

Risks/Benefits:

Potential risks: This research poses two main risks to subjects: 1) emotional distress and 2) risk to privacy. There may be some emotional distress caused by learning about one's unconscious racial biases. Emotional distress may also result from passive deception; subjects will not be told that racial bias is the focus of the study until after the intervention. Privacy risk arises because racial biases may be considered sensitive information about the subject.

Adequacy of Protection Against Risks Recruitment and Informed Consent: All subjects will be recruited by the PI through departmental email blasts, emails to contacts of the study team and departmental faculty meeting presentations. The consent process will take place in the Clinical Skills Center on the prior to the standardized patient encounter. Subjects will be informed of their right to participate or discontinue participation at any time without jeopardizing their employment, professional position or relationships with colleagues. Copies of the signed consent form will be given to the subject and placed in the subject's research folder in a locked file drawer in an office in the PI's office expressly designed for this purpose.

Incomplete disclosure of information during the consenting process: Subjects will be informed that they are participating in a study exploring physician cognition, cognitive biases and communication behavior. They will be informed that study participation includes completion of an on-line instrument after the standardized patient encounter. They will not be told during the consent process that implicit racial bias and stereotyping are the focus of the study.

Protections against risks: Personal identifiers not be collected in the on-line instrument; instead a unique study ID will be assigned to each participant. Video recordings will be maintained in a password protected digital folder on a hard drive in a locked office. This database of coded communication scores will be stripped of participant identifiers as soon as all data is collected, and a file linking the study ID and patient identifiers will be kept in a separate, password protected digital folder.

Debriefing: After completion of the on-line bias measurement instrument, the subjects will be shown a debriefing statement including full disclosure of the aims of the study and the intent to examine the relationship between implicit racial bias and stereotyping and communication behavior. This statement will include a reminder that they may discontinue participation at any time and request that their data be removed from the dataset without penalty. The PIs contact information will be provided.

Potential Benefits of the Proposed Research: Subjects are not expected to directly benefit from this research. Importance of the knowledge to be gained: The information from this study will provide knowledge about the role of implicit bias and stereotyping in communication behaviors that have profound effects on the quality of end-of-life care for Black patients. It will also inform efforts to mitigate the effects of this bias in future studies.

Necessity of withholding information in the consent process: Withholding of complete information from the consent process is only acceptable if there is no reasonably effective alternative to achieve the goals of the research. In this case, knowledge of the intent to study implicit racial bias is likely to affect the way physicians communicate in the standardized patient encounter, thus invalidating the results of the study.