The study is a cluster RCT conducted in multiple oncology practices and cancer centers in the Rochester/Buffalo, NY and Sacramento, CA regions, designed to evaluate the effects of a theory-based intervention to improve communication between oncologists and patients with advanced cancer and their caregivers. The methods were developed in collaboration with Phyllis Butow, Martin Tattersall, Adam Walczak and colleagues, who are conducting a parallel study in Sydney, Australia called Conversations with Your Doctor: Making the Most of Medical Consultations. That study incorporates similar interventions but a different study design. The VOICE RCT incorporates many elements of effectiveness studies, such as broad eligibility criteria, usual-care controls, tailored interventions, patient-oriented outcome measures, and intention-to-treat analyses. The study design and all consent forms have the approval of the Institutional Review Boards at each institution.
The study procedures (see Figure ) are separated into two phases. Phase 1 involves preparation, physician recruitment, piloting, and pre-randomization data collection of physicians’ communication behaviors. Three Phase 1 patients and their caregivers are recruited for each physician. Each of these patients has one office visit audio recorded, allowing us to assess physicians’ baseline communication behaviors for potential use as a covariate in the RCT analyses. Phase 1 patients and caregivers also complete pilot versions of some study measures pre- and post- office visit to inform finalized versions of the measures for the RCT.
Phase 2 is the cluster RCT. Physicians are the unit of randomization and are randomly assigned to the intervention or control condition. Up to seven Phase 2 patients (and caregivers) are recruited per physician. Patients and their caregivers are randomly assigned by proxy: patients of physicians in the intervention condition are assigned to the intervention, and patients of physicians in the control condition are assigned to the control group. Patients complete measures at study entry, participate in the intervention or control condition, agree to have an oncology office visit audio recorded, and complete follow-up measures immediately after the office visit, approximately 2–4 days after the office visit, and every 3 months for up to 3 years or until death. Caregivers, where available, also participate in the study and complete measures periodically, including measures one-month post-mortem.
Following completion of Phase 2, data analysis will ensue. After the study has closed, physicians from the control group will be offered the opportunity to receive the intervention, both as an incentive to participate and to further their professional development. In addition to the primary VOICE trial, additional studies are using the procedures described here to examine patient, caregiver, and oncologist emotional processes [46
] that drive decision making, and the effect of the intervention on caregiver bereavement outcomes. Prior research [24
] suggests that improvements in end-of-life care can mitigate bereavement-related morbidity.
Selection of study sites
In the Rochester region, oncologists (n
25) were recruited from academic and private practices in western New York; recruitment is complete. In the Sacramento region, all oncologists are recruited from the UC Davis Comprehensive Cancer Center (n
14), and recruitment is nearly complete. One Rochester physician withdrew from the study prior to any data collection due to lack of time. We anticipate a total of 40 oncologists.
Eligibility, recruitment, consent, and randomization
Ethical approval was obtained from the IRBs of the five affiliated institutions where the study is being conducted. All participants (i.e., patients, caregivers, and oncologists) complete written informed consent documents. Specific inclusion and exclusion criteria are presented in Table , and participant eligibility is verified immediately prior to completing baseline study measures.
Inclusion and exclusion criteria for oncologists, patients, and caregivers
Medical oncologists who care for patients with solid (non-hematologic) cancers are solicited for participation through presentations about the study at grand rounds and faculty meetings, or through personal contacts via cancer center directors or project investigators. Interested oncologists then meet with study personnel (e.g., project manager, research assistant, or research oncologist) to achieve a clear understanding of all study components, provide written consent, and complete the baseline surveys. The consent document seeks each oncologist’s agreement to (a) guide recruitment efforts among their patients by determining eligibility, (b) participate in an educational program to enhance their communication skills, (c) audiotape one clinic encounter per enrolled patient, and (d) complete brief surveys at baseline, after each audio-recorded office visit, and the study conclusion. Oncologists are told “the purpose of this study is to assess the impact of brief educational interventions on clinical care and outcomes of patients with cancer and their caregivers. Outcomes of interest include communication between patients, caregivers, and physicians, as well as patient well-being and health services utilization.” Oncologists receive $600 for completing surveys and identifying eligible patients for the study.
Research assistants identify all patients by working closely with participating oncologists and their clinic staff to review clinic rosters in detail to ascertain that all potentially eligible patients are identified. Depending on the site and differing IRB requirements, potentially eligible patients are either (a) approached by physicians or practice nurses and then by the research assistant or (b) sent a letter, a study brochure, and an opt-out card. Patients approached by physicians or clinic staff are asked if they would be willing to speak to a research assistant to learn more about the study. Patients who receive recruitment materials by mail are contacted via phone by a research assistant if an opt-out card is not returned to the research office within 7 days. The research assistant provides them with study information, and gives them time to deliberate about participation and ask questions. The screening and consent process continues until the research assistant feels that the potential participant fully understands all aspects of study involvement. The research assistant then schedules a time to meet and obtains written consent from those who voluntarily wish to enroll, using IRB-approved consent forms. Patients are reminded that they can opt out at any time. All patients are asked to complete baseline surveys and to give permission to have one of their oncology office visits audio recorded. The research assistant orally administers the baseline surveys as well as other study materials as needed. If consent documentation and baseline survey measures are completed in separate visits, eligibility criteria are reviewed immediately prior to survey administration to ensure continued eligibility.
Phase 1 patients complete two sets of orally-administered surveys, one before and one after the audio-recorded office visit with their oncologist. Phase 2 patients (a) have one audio-recorded office visit with their oncologist, (b) complete surveys prior to and immediately after their office visit, then 2–4 days after the office visit, and every three months for up to three years, and (c) give permission for research staff to access their medical records to ascertain their use of health services such as emergency department visits, hospital admissions, cancer treatments, and community-based nursing services. The initial survey takes up to 60 minutes with later surveys taking anywhere from 10 to 30 minutes. Patients receive $15 for each set of surveys, for a maximum of $30 in Phase 1 and $180 in Phase 2.
All patients are asked to identify a “family member, partner, friend or someone else who is involved with your health care issues, for example, someone who you talk to about personal issues including medical decisions or who comes to doctor appointments with you. This person may also help with routine day-to-day activities, like transportation or paperwork.” The term “caregiver” is used to describe these persons for scientific purposes, and patients are not required to self-identify individuals as such. Patients are asked to identify up to three potential caregivers, rank them in order of the likelihood that they will attend oncology office visits, and provide permission to contact the primary caregiver (or others caregivers, if needed) and provide them with a study brochure.
Once identified, the research assistant independently approaches the caregiver in person or by telephone and asks if they would be willing to participate and provide written consent. Caregivers often accompany patients at the time of the patient consent. The research assistant reviews the consent form in detail with the caregiver and advises them that they can opt out of the study at any time. Caregivers are told the same study details as the patients and complete surveys at approximately the same time intervals as the patients. Caregivers are asked for permission to be contacted in the event that the patient dies within three years of enrollment in order to meet with a research assistant (or speak on the telephone if preferred) to complete a survey. The survey is completed approximately one month after the death of the patient. The research assistant responds to caregivers’ questions until it is clear that the caregiver has full comprehension of the consent form and their involvement in the study. Caregivers receive $15 for each set of surveys, for a maximum of $30 in Phase 1 and $180 in Phase 2.
A stratified block-randomization scheme is used to assure balanced assignment by clinic site and cancer focus. Oncologists are grouped into a site according to their health center, clinic, or practice of employment. Within each site, oncologists are randomly assigned approximately evenly across the treatment and control conditions. Sites with a single oncologist are grouped with a similar site for randomization purposes. Oncologists are also categorized by their cancer focus, including breast-cancer oncology (≥50% of patients have diagnoses of breast cancer) and non-breast cancer oncology groups. Within these areas of focus, oncologists are randomly assigned approximately evenly across the treatment and control conditions. This accounts for any biases that may be introduced by the low prevalence of breast cancer in men and the potential that breast cancer patients may be more “activated” than patients with other cancers. For each site/focus combination in the study, separate sequences of random numbers have been generated for use in assigning oncologists to the intervention and control groups. To preserve blinding, assignment to the treatment or control conditions is maintained by the study statistician and not explicitly revealed to transcriptionists or coders of the audio-recorded office visits.
For the initial baseline surveys, a time is arranged to meet with the patient and caregiver, either together or separately, in a private area based on the participant’s comfort and preference. This may include a meeting room in the cancer center, the infusion suite, the participant’s home, or a local coffee shop with private areas. Once both the patient and caregiver have completed these surveys, the research assistant answers any questions and reviews the next steps of the study, the audio-recorded office visit and post-visit survey. For that visit, the research assistant usually meets the patient and caregiver in the physician’s waiting room and accompanies them to the clinic room, turns on and places two audio recorders in the room, and leaves. If another person attends the visit, the research assistant obtains verbal permission to record them. The research assistant collects the recorders after the office visit and meets with the patient and caregiver to administer brief surveys. This completes Phase 1. In Phase 2, the scenario is identical, except that the patients and caregivers in the intervention arm receive a 1-hour coaching session prior to the office visit. After the audio-recorded office visit, the patient completes a survey, and the research assistant explains to the patient and caregiver that they will be contacted in 2–4 days by phone to answer some additional questions. The patient and caregiver are also called three months after the audio-recorded visit to complete surveys and every three months thereafter for up to three years.
Description of the intervention and control conditions
Oncologists randomized to the intervention arm participate in a multifaceted tailored educational intervention involving standardized patients instructors (SPIs). Patients and caregivers complete a coaching intervention to facilitate prioritizing and discussing questions and concerns.
Oncologists completing the intervention meet with SPIs for two in-office educational outreach sessions [47
], including a 60-minute training session and a 45-minute booster session [48
]. At the first session, the SPIs show them a 15-minute DVD created by the study investigators specifically for this project, which presents actual clinical examples to outline key skills in discussing prognosis and treatment choices with patients with advanced cancer and their caregivers. Oncologists receive a copy of the DVD to keep and receive a 1-page summary of evidence-based guidelines for communication in advanced cancer [13
]. They are also given a Communication Guide
“reminder” card prompting them to discuss topics, such as prognosis and symptoms that appear on the patients’ Question Prompt List, such as prognosis and symptoms (see below, QPL).
After viewing the DVD, the oncologist is asked to select one or two of the key skills to practice with two SPIs who portray a 60-year-old man with incurable cancer and his wife. A few days prior to the session, the oncologist is sent a medical “chart” to review in anticipation of the SPI visit, containing a comprehensive prior consultation note which includes radiographic and laboratory results. In the “chart,” the patient is described as having metastatic colon cancer with progression despite one course of state-of-the art chemotherapy, and intolerance of a second course of chemotherapy. Current symptoms include severe diarrhea, nausea, and fatigue. The SPIs present as facing important treatment decisions over the next few weeks, while having an unclear idea of prognosis. The SPIs engage the oncologist in a role play exercise with specific tailored feedback on the key communication skills described in the DVD; oncologists are then given the opportunity to rehearse areas of difficulty. Oncologists also receive a follow up letter from the SPIs that includes a version of the communication guide with individually-tailored comments. The guide shows the four key skills discussed in training and summarizes what was agreed to by the oncologist and SPIs during the session, such as the physician’s demonstrated areas of strength, as well as those areas jointly identified as challenges for further growth. One month later, oncologists complete a 45-minute reinforcement session, which uses a similar format. Specifically, the same SPIs return for a simulated follow-up visit, where the cancer has “progressed” despite third-line treatment. Key communication skills highlighted in the DVD and SPI feedback (see Table ) were chosen based on ecological theory and evidence that they (a) promote discussions of prognosis and treatment choice, (b) can be taught in brief interventions, and (c) are associated with patient trust and lower anxiety [13
Oncologist communication behaviors targeted in the tailored educational intervention
Patient and caregiver coaching
Patients and caregivers completing the intervention meet with a coach trained to facilitate health communication. During the coaching session, the coach gives each patient and caregiver a Question Prompt List (QPL), which is organized in a booklet called My Cancer Care
. The QPL includes questions about diagnosis, prognosis, treatment options, symptom management, transitions in care, self-care, and family needs. Sample questions include, “What are the pros and cons of further treatment for my cancer?” and “How can I help my family and children understand what is happening?” The QPL was developed in Australia, adapted for the United States, and piloted simultaneous in both countries [60
]. Using the QPL as a guide, patients are coached to (a) identify and prioritize 2–3 personally relevant questions on the list, (b) ask these questions during the visit, (c) ask their oncologist for clarification when they do not understand, (d) express desire to participate more actively in discussions about prognosis and treatment choices, and (e) prepare for the future [22
]. These skills promote the same goals as the physician intervention. The coach makes follow-up phone calls at monthly intervals for up to three months to reinforce the coaching intervention and address patients’ concerns [62
]. The coaches include a nurse and social workers with healthcare backgrounds, similar to prior studies [61
]; they do not provide disease-specific information. All Rochester and UC-Davis coaches participate in local training, video conferences, and a 3-day intensive training, using methods and materials similar to our previous studies [63
]. All intervention sessions are audio-recorded and progress notes are written after each of the 3 coaching follow-up telephone calls. The coaches hold weekly conference calls to fine-tune, discuss, and review their coaching sessions.
The control group involves care as usual. Oncologists as well as patients/caregivers meet with the research assistants to complete the same surveys as intervention participants but receive no training.
The audio-recorded office visits for all participants are coded for each of the four domains of communication behaviors hypothesized to be affected by the intervention: Engaging
, and Framing
(see Table ). The Active Patient Participation Coding (APPC) [32
] scheme is used to measure Engaging
communication behaviors, such as patient assertive behaviors and oncologist facilitative behaviors, that promote patient participation in the decision-making process. The Verona VR-CoDES system [64
] is used to code for sequences of Responding
to emotion, such as patient expression of emotional cues and concerns and oncologist responses to these expressions of emotion. The Prognostic and Treatment Choices (PTCC) [65
] system is used to code Informing
behaviors, such as the oncologist addressing patients’ wishes for information regarding prognosis and treatment choices. The Optimism/Pessimism subscale of the Framing of Prognostic Information (FPI) [6
] system is used to code Framing
behaviors, such as oncologists’ balance in expressions of optimistic and pessimistic perspectives about “what to expect.” Several exploratory measures assess the level of patient and caregiver question asking [26
] and the degree of shared decision making [66
Patient survey measures
The survey measures completed by patients in Phase 2 are outlined in Table . Several scales measure aspects of the quality of interactions between physicians and patients. The quality of the Physician
is measured with The Human Connection (THC) [68
] survey. Characteristics of the Physician
are measured with the Health Care Communication Questionnaire (HCCQ) [69
], the Mishel Uncertainty in Illness Scale (MUIS) [70
], and the Information Preference Scale (IPS) [71
]. Patient Communicational Self
is measured using an adaptation of the Perceived Efficacy in Patient-Physician Interactions (PEPPI) [43
] survey. The Physician
survey assesses topics discussed in recent oncology office visits, such as prognosis, end-of-life care, and emotional issues, supplementing information obtained from the single audio-recorded oncology office visit.
Survey measures completed by patients in the RCT
Several scales relate to patient attitudes, values and beliefs. Patients’ Preferred Decision Role
is measured with the Control Preferences Scale [72
], which is compared to an adapted version assessing patients’ Actual Decision Role
in oncology office visits. The Treatment Preferences
survey measures preferences for experimental treatments, life support, and palliative care, in the event that no further anti-cancer treatments would be helpful. Illness Acceptance
is measured with the Peaceful Acceptance subscale of the PEACE [7
]. Prognostic Forecasting
is measured using items [5
] assessing patients’ beliefs about their chances of living two years and chances of being cured.
is measured with the Global Quality of Life, Psychological Well-being, and Existential Well-being subscales of the McGill Quality of Life (MQOL) [73
] survey as well as the Physical Well-being and Social/Family Well-being subscales of the Functional Assessment of Cancer Therapy – General (FACT-G) [74
Each of these surveys has been tailored to improve clarity and relevance, reduce respondent burden, and reflect American English. Most of these measures have been piloted with patients in Phase 1 and refined as needed to reduce ceiling and floor effects or further improve item wording.
Caregiver survey measures
The survey measures completed by caregivers in Phase 2 are outlined in Table . Measures of the Physician
, and Caregiver Communicational Self
have been adapted from patient versions of the same scales. Using measures adapted from patient versions, caregivers also provide informant ratings of Patient Treatment Preferences
, Patient Illness Acceptance
, and Patient Well
, and complete a measure of Prognostic Forecasting
. They report on Patient Quality of Death
via qualitative questions about the healthcare decision-making process, the Quality of Life Near Death (QOLND) [75
] survey, and items from the Quality of Death Long-Term Care – Cognitively intact (QOD-LTC-C) [76
] survey. As with the patient measures, each of these surveys has been tailored for the current study.
Survey measures completed by caregivers in the RCT
Oncologist survey measures
The survey measures completed by physicians are outlined in Table . Communication Skills
are measured with a pilot-tested survey derived for this study, which assesses perceived skills in discussing prognosis and end-of-life issues, such as giving bad news, expressing empathy, and discussing referrals to palliative care. Decision Making Skills
are measured with a survey adapted from a prior measure [72
] in order to assess physician comfort with decision making across varying levels of patient involvement (e.g., physician makes the decision, patient makes the decision, shared decision making). Several measures have been adapted from the patient self-report surveys: physician beliefs about Patient Treatment Preferences
, physician beliefs about Patient Illness Acceptance
, and Prognostic Forecasting
about patients’ prognoses (e.g., chances of a cure, chances of living two years, patients’ understanding thereof). Patient Disease Status
is assessed via several survey items, which supplement information from patient medical records.
Survey measures completed by physicians
Medical chart abstraction
The death of participating patients is ascertained by checking in regularly with participating oncologists, through scheduled follow-up sessions with patients and caregivers, by reviewing death notices (obituaries) in the local press, and by periodically reviewing electronic health records. Once a death is identified, study staff identify all emergency department visits or hospital overnight stays occurring in the last 30 days before death and abstract the corresponding medical records for the following data elements: (1) date of death, (2) dates and circumstances of attendance at a hospital, emergency department, and intensive care unit, (3) use of intubation, cardiopulmonary resuscitation (CPR), hemodialysis, permanent enternal feeding tube, and chemotherapy, (4) involvement of hospice and palliative care services, and (5) completion of Do Not Resuscitate/Do Not Intubate (DNR/DNI) forms and Advanced Directives. Outpatient records are abstracted to identify any chemotherapy agent given in the last 14 days before death and any new chemotherapeutic regimen started in the last 30 days before death, the absence of which indicate higher quality end-of-life care [77
Both the oncologist and patient/caregiver components of the intervention are monitored for fidelity. For the oncologist training sessions, all audio recordings are reviewed by all SPI trainers. For the patient/caregiver component, audio recordings are reviewed for each coach’s first five coaching sessions, every subsequent session until >95% fidelity, and at least every fifth session thereafter, including in-person and phone sessions. To assess fidelity and ensure standardization of survey administration procedures, each research assistant’s first five sessions are monitored by direct observation and reinforced every 4–6 months during data collection. In addition, audio-recordings of qualitative survey items are reviewed thereafter as needed.
Sample size determination
This is a stratified cluster randomized study, with the physician as the unit of randomization. Based on prior studies, we have made the following assumptions: physician attrition 0-3%, patient attrition <5% for audio-recordings, and 10%, 30% and 35% for the 2–4 day, 3-month, and 6-month post-visit patient surveys [78
], 80% patient mortality during 3-year follow up, availability of 85% of caregivers for post-death interviews [3
], availability of 90% of medical records for audit [3
], no differential attrition between the intervention and control groups, and an intraclass correlation coefficient (ICC) of .10 or less for within-physician clustering on patient and caregiver survey measures [81
Prior work found that activation training increases physician communication behaviors two to three fold [26
]. This equates to an effect size of 2.0 standard deviation (SD
) units. Thus, for our primary outcome (Aim 1a), we expect a 2.0 SD
improvement across each communication measure; 0.5 SD
is clinically significant. Power is based on a single measurement of communication during the oncology office visit. With at least 19 oncologists in each group, 7 patients per oncologist, and an ICC of .10, the minimum detectable effect size is 0.50 SD
For our survey outcomes (Aims 1b & 2), we rely on observational data [83
] indicating effect sizes of 0.40 to 0.70 SD
for the relationship between communication and patient well-being. With an intervention we would expect these differences to be larger. The power analysis takes advantage of a repeated-measures design [84
]. For patient data, we assume one pre-intervention measure and two post-intervention measures, and an average correlation among repeated measures of .50. Our proposed sample size is sufficient to detect an effect size of 0.40 SD
with power of .80. For our utilization outcomes, we rely on data that suggest 3- to 8-fold differences in use of aggressive treatments during the final week of life between patients who have had discussions compared to those who have not [3
]. Thus, the study will be adequately powered, even considering attrition.
Planned analytic approach
This is a cluster-randomized trial, where our primary communication outcomes (Aim 1a) are measured at the level of the physician-patient dyad and our secondary outcomes (Aims 1b, 2, & 3) are measured at the level of the patient. Analyses are based on published guidelines for group (cluster) RCTs [85
], in this case, clustering at the site and physician level. Prior to hypothesis testing, preliminary analyses will examine whether random assignment produced comparable groups in terms of patient disease status, patient well-being, and pre-RCT (Phase 1) physician communication styles. Any confounders will be included in remaining analyses. Hypothesis testing will involve comparisons of the two randomized groups. Since the patients are nested within oncologists, methods for panel data will be applied, and random effects for physicians may be added to account for the within-physician correlations of each dyad. We will use both generalized estimating equations (GEE) and generalized linear mixed-effects models (GLMM). The primary outcomes (Aim 1a) are unlikely to require substantial attention to attrition, since measurement occurs shortly after randomization, and will not require methods for repeated measures. Hierarchical linear models with nested random coefficients will be applied where secondary outcomes that are measured repeatedly (Aims 1b, 2, & 3). As missing values are likely, we will examine the nature of the missing data and use weighted GEE (WGEE), or multiple imputation and sensitivity analyses, if necessary. In assessing significance across multiple tests, we will use procedures to control the false discovery rate (FDR) at 5% within sets of related analyses [86