Statistical Analysis Plan
eTable 1. Patient and physician characteristics (no statistically significant differences between intervention and control for Pre-Randomization or Cluster RCT)
eTable 2. Physician and patient-caregiver interventions
eTable 3. Health care utilization in the final 30 days of life
eTable 4. Raw scores for each of the component communication measures
eFigure 1. VOICE STUDY Cluster RCT Physician/Patient CONSORT Flow Diagram
eFigure 2. Quality of Life, Intervention vs Control: Terminal Decline Model
eFigure 3. Health care utilization in the final 30 days of life
eFigure 4. VOICE Study CONSORT diagram: pre-randomization sample
Customize your JAMA Network experience by selecting one or more topics from the list below.
Epstein RM, Duberstein PR, Fenton JJ, et al. Effect of a Patient-Centered Communication Intervention on Oncologist-Patient Communication, Quality of Life, and Health Care Utilization in Advanced CancerThe VOICE Randomized Clinical Trial. JAMA Oncol. 2017;3(1):92–100. doi:10.1001/jamaoncol.2016.4373
Copyright 2016 American Medical Association. All Rights Reserved.
Can communication between patients with advanced cancer and their oncologists be improved?
This cluster randomized clinical trial of communication training for oncologists paired with previsit coaching for patients showed clinically and statistically significant improvement in patient-centered communication.
Paired communication training for patients and oncologists helps achieve patient-centered care in advanced cancer: engaging patients in consultations (asking questions, expressing preferences), responding to emotions, and providing information about prognosis and treatment choices.
Observational studies demonstrate links between patient-centered communication, quality of life (QOL), and aggressive treatments in advanced cancer, yet few randomized clinical trials (RCTs) of communication interventions have been reported.
To determine whether a combined intervention involving oncologists, patients with advanced cancer, and caregivers would promote patient-centered communication, and to estimate intervention effects on shared understanding, patient-physician relationships, QOL, and aggressive treatments in the last 30 days of life.
Design, Setting, and Participants
Cluster RCT at community- and hospital-based cancer clinics in Western New York and Northern California; 38 medical oncologists (mean age 44.6 years; 11 (29%) female) and 265 community-dwelling adult patients with advanced nonhematologic cancer participated (mean age, 64.4 years, 146 [55.0%] female, 235 [89%] white; enrolled August 2012 to June 2014; followed for 3 years); 194 patients had participating caregivers.
Oncologists received individualized communication training using standardized patient instructors while patients received question prompt lists and individualized communication coaching to identify issues to address during an upcoming oncologist visit. Both interventions focused on engaging patients in consultations, responding to emotions, informing patients about prognosis and treatment choices, and balanced framing of information. Control participants received no training.
Main Outcomes and Measures
The prespecified primary outcome was a composite measure of patient-centered communication coded from audio recordings of the first oncologist visit following patient coaching (intervention group) or enrollment (control). Secondary outcomes included the patient-physician relationship, shared understanding of prognosis, QOL, and aggressive treatments and hospice use in the last 30 days of life.
Data from 38 oncologists (19 randomized to intervention) and 265 patients (130 intervention) were analyzed. In fully adjusted models, the intervention resulted in clinically and statistically significant improvements in the primary physician-patient communication end point (adjusted intervention effect, 0.34; 95% CI, 0.06-0.62; P = .02). Differences in secondary outcomes were not statistically significant.
Conclusions and Relevance
A combined intervention that included oncologist communication training and coaching for patients with advanced cancer was effective in improving patient-centered communication but did not affect secondary outcomes.
clinicaltrials.gov Identifier: NCT01485627
The National Cancer Institute, the National Academy of Medicine, the American Society for Clinical Oncology, and the National Priorities Partnership all call for improved patient-physician communication in the context of serious and life-limiting illnesses, citing effects of good communication on quality of care and quality of life (QOL) and an ethical mandate that patients be offered participation in informed decisions regarding their care.1-4 In advanced cancer, inadequate communication about prognosis and treatment choices is common5-7 and is associated with unrealistic patient expectations regarding curability,8 provision of aggressive treatment that is not concordant with patients’ wishes and enrollment in hospice too late to deliver discernable benefit.8-12 Critical conversations typically do not happen, or happen in hospital shortly before a patient’s death.13
Making high-quality conversations happen is difficult. More than 90% of patients with advanced cancer say they want to be actively involved in their care and value frank and sensitive conversations about QOL, prognosis, and treatment choices.10,14,15 Yet patients are often reluctant to be assertive, ask questions, request clarification, express emotions directly, or state opinions and preferences.16 As death approaches, patients express considerable ambivalence about end-of-life discussions, often indicating that now is “not the right time.”17-20 Similarly, while most clinicians indicate willingness to have these discussions “now,” few follow through.21,22 More typically, clinicians wait for patients to signal interest, then offer prognostic information that is optimistically biased.23,24
Prior attempts to improve patient-physician communication in advanced cancer have had limited impact. Intensive workshops inconsistently improve trainee communication behaviors25,26 and are impractical for busy clinicians. Brief individualized in-office interventions using expert feedback on video recorded encounters improved oncologists’ empathy and patient-reported trust,27 but this approach has not been applied to other communication behaviors, such as delivery of prognostic information. Patient coaching increased discussion of cancer pain,28 and, in palliative care, question prompt lists (QPLs) have increased question-asking29—but only when physicians encouraged patients to ask questions.
The Values and Options in Cancer Care (VOICE) study30 combined 2 interventions, a brief individualized oncologist skill-based training, and individualized patient and caregiver coaching incorporating a QPL. Based on an ecological model of patient-clinician communication, both interventions were designed to promote the involvement in care that patients and families desire but rarely request and emphasized the same communication skills and topics identified in prior research: engaging patients to participate in the consultation, responding to patients’ emotions, informing patients about prognosis and treatment choices, and framing information in a balanced manner.6,18,29,31-35 The primary outcome was patient-centered communication in these domains. Secondary outcomes were shared understanding, patient-physician relationships, QOL, and health care utilization at the end of life.33
We conducted a multisite cluster randomized clinical trial (RCT) of an intervention to improve communication between patients (and caregivers when available) and oncologists. Oncologists randomized to the intervention arm participated in individualized communication training using standardized patient-instructors (SPIs), while their patients (with caregivers) participated in an individualized communication coaching session with follow-up telephone calls. After a prerandomization phase designed to assess baseline communication patterns of participating oncologists, we enrolled participants (from August 2012 through June 2014) and followed them until October 2015. We obtained institutional review board approval for all study sites, and participants provided written informed consent and received $15 per questionnaire completed. See the published study description36 and the eAppendix in Supplement 1 for the study protocol and descriptions of the interventions and Supplement 2 for the statistical analysis plan.
We conducted the study in community-based cancer clinics (4), academic medical centers (3) and community hospitals (3) in Western New York and Sacramento, California.
We recruited medical oncologists who care for patients with nonhematologic cancers at practice meetings at participating clinics. The mean physician age was 44 years; 27 (71%) were male; 17 (45%) were white, 16 (42%) were Asian, and 5 (13%) were of other race.
With clinic staff, research assistants reviewed clinic rosters of enrolled physicians to contact potentially eligible patients age 21 years or older, able to understand spoken English and provide written informed consent, and who had either stage IV nonhematologic cancer or stage III cancer and whose physician “would not be surprised” if the patient were to die within 12 months.37,38 We excluded inpatients and those in hospice. We first recruited 3 to 4 “prerandomization” patients per physician who agreed to have 1 office visit audio recorded and complete questionnaires before and after the office visit. After physician randomization, we recruited a new cohort of patients, up to 10 per physician, for the cluster RCT (eTable 1 in Supplement 3) until we reached the target sample size of 265 patients. Cluster RCT patients also agreed to an audio recorded office visit and previsit and postvisit questionnaires; in addition, they agreed to participate in intervention or control conditions, complete questionnaires quarterly for 3 years, and to have their medical records abstracted. Patients were blinded to study arm assignment until completion of baseline measures.
Research assistants asked patients to identify “a family member, partner, friend, or other individual involved with your health care issues, preferably someone who comes to physician appointments with you.” Eligible caregivers were 21 years or older and able to understand spoken English and provide written informed consent.
The experimental intervention36 included (1) a 2-session in-office physician training (1.75 hours) using a brief video, feedback from standardized patients portraying roles of patients with advanced cancer who also critiqued up to 2 audio recorded study patient visits, and (2) a single 1-hour patient and caregiver coaching session incorporating a question prompt list to help patients bring their most important concerns to their oncologist’s attention at an upcoming office visit, plus up to 3 follow-up phone calls (Table 1; eTable 2 in Supplement 3). Trainers and coaches underwent 3-day on-site trainings. To promote patient-centered communication about disease course, prognosis, treatment decisions and end-of-life care, physician and patient interventions focused on the same 4 key domains of patient-centered communication.36 Intervention sessions were audio recorded and reviewed by lead trainers and investigators using a fidelity checklist. Fidelity was 94% or higher.
All intervention physicians completed both training sessions. All intervention patients received in-person coaching; of the 52% who responded to a mailed survey, 87% “would recommend coaching to other patients with cancer”; and 85% were able to ask “all” or “most” of their “most important” questions. Of the 130 coached patients, 94% participated in ≥1 follow-up call (≥2 calls, 78.7%; 3 calls, 58.3%); reasons for nonparticipation were death and/or illness (47.1%), unreturned phone calls (47.1%) and refusal and/or withdrawal (5.8%). Control physicians and patients received no training.
We audio recorded the first physician visit after the coaching session (for intervention) or after study entry (control). The primary outcome was a composite of 4 prespecified communication measures matched to the goals of communication training, described in detail in Table 1—engaging patients in consultations (Active Patient Participation Coding [APPC]39), responding to patients’ emotions (Verona VR-CoDES41,42), informing patients about prognosis and treatment choices (Prognostic and Treatment Choices [PTCC] Informing subscale44) and balanced framing of decisions (PTCC Balanced Framing subscale44). The composite score derived from these scales were designed to capture key elements of 6 interrelated functions of communication outlined by the National Cancer Institute: fostering healing relationships (APPC), exchanging information (PTCC), managing uncertainty (PTCC), making decisions (APPC, PTCC), responding to emotions (Verona), and enabling patient self-management (APPC).47 Coding of the 4 measures was performed by teams of trained university students who were audited continuously and blinded to study hypotheses and group assignment. We transformed each of the 4 component scores to z scores based on the prerandomization phase sample means (SDs):
z = (Raw Score – Prerandomization Phase Mean)/Prerandomization Phase SD
The 4-component z-scores were averaged to form the primary outcome, a composite measure with better overall precision and sensitivity than the individual components for assessing intervention effects on the multiple targeted communication goals.
We assessed patient-physician relationships using The Human Connection (THC) scale,48 the Health Care Climate Questionnaire (HCCQ),49 and the Perceived Efficacy in Patient-Physician Interactions (PEPPI) scale50 shortly after the audio recorded visit. Physicians and patients were also asked to estimate 2-year survival and curability of the patient’s cancer on a 7-point scale (100%, about 90%, about 75%, about 50/50, about 25%, about 10%, 0%, don’t know); discordance was defined as 2 or more categories of difference.
We administered QOL questionnaires at 3-month intervals from study entry for up to 3 years and prespecified a composite QOL score to be the average of 5 z-scored subscales: McGill QOL scale single item, McGill Psychological Well-Being subscale, McGill Existential Well-Being subscale, FACT-G Physical Functioning subscale, and FACT-G Social Functioning subscale51,52; all are widely used in research in advanced cancer. Fewer than 3% of follow-up questionnaires were missing.
Trained nurses and physicians abstracted utilization data from medical records at relevant hospitals, offices, and hospice organizations. Based on a review of the literature,9,53-55 we prespecified a composite utilization score of 3 indicators of aggressive treatment in the last 30 days of life (chemotherapy, potentially burdensome interventions, emergency department [ED]/hospital admission) and hospice utilization (eTable 3 in Supplement 3).
We randomized by physician and stratified by site (New York or California) and oncologist subspecialty (≥50% vs <50% of patients with breast cancer) to balance sex and other unmeasured patient characteristics that might be associated with the communication outcomes. Within strata, we randomly assigned physicians at a 1:1 ratio to intervention or control. We recruited, obtained consent, and enrolled patients based on the arm to which their physician was assigned. We oversampled patients with caregivers to achieve recruitment goals for a companion study of caregiver bereavement. Only the study statisticians were aware of the random number sequences and treatment assignment, preserving blinding among transcriptionists, coders, and abstractors.
To affect utilization and patient outcomes meaningfully, we felt a moderately large effect on communication would be needed. To account for attrition and variance inflation arising from cluster randomization, we used standard formula as well as simulation studies using SAS statistical software to determine that a target sample size of 38 physicians and 265 patients would yield the effective sample needed to provide at least 80% power (2-sided testing, α = .05) to detect standardized effects of 0.50 for the primary communication outcome.36
Between-group comparisons on communication and utilization outcomes were conducted using Wald-type tests from prespecified mixed-effects linear regression models (for continuous outcomes) and generalized estimating equations for binary outcomes, specified to account for the nesting of patients (the units of analysis) within physicians (the units of randomization). For all regression analyses, covariates for study site and breast cancer subspecialty were included to account for the stratified randomization, as well as patient-level covariates to adjust for demographic and cancer characteristics. Between-group comparisons of QOL trajectories were performed using the terminal decline model (TDM) of Li et al56 that accounts for mortality by jointly modeling QOL and survival using piecewise linear regression and exponential hazards regression models, respectively. The TDM parameterizes time counting backward from patient time of death and is specified with 2 periods for each component, the “terminal decline” period nearest death and the more remote period before then; our model extends on Li et al56 to permit the inclusion of study covariates. We chose 9 months and 12 months as the duration of the terminal decline period for QOL and mortality, respectively, based on the observed change point in the data.
For the analysis of communication outcomes, we used the prerandomization data to adjust for between-physician differences among the 38 physician clusters. Hence, the data sets for these mixed models analyses included observations from prerandomization and postrandomization audio recordings, and the models included fixed-effects terms for phase (prerandomization vs postrandomization), study arm, and the interaction of phase and arm. Intervention effects were estimated as the between-arm difference in adjusted mean difference from prerandomization to postrandomization samples. For the other continuous outcomes, only postrandomization data were included, and intervention effects were estimated as adjusted mean differences. In model validation and exploratory analysis, heterogeneity of treatment effects was assessed by adding interaction terms to regression models to compare intervention effects across prespecified subgroups. Residual plots were also used for model validation.
Statistical analyses (Supplement 2) were conducted in version 9.4 of the SAS System.
Of 52 physicians contacted, 43 enrolled and 38 participated in the cluster RCT (eFigure 1 in Supplement 3). Of the 265 participating patients, 194 (73.0%) had an enrolled caregiver. Patient characteristics across study arms were well matched (eTable 1 in Supplement 3); mean age was 64.4 years, 55.0% were female, 11.5% were nonwhite, 28.0% had high school education or less, and 19.0% reported income of $20 000 or less. The mean follow-up for patients was 15 months; by study closing (October 1, 2015), 151 patients had died, 18 had withdrawn, and 1 was lost to follow-up (eFigure 1 in Supplement 3). We abstracted all decedents’ medical records.
In fully adjusted models, the composite communication score showed a significant intervention effect (estimated adjusted intervention effect, 0.34; 95% CI, 0.06-0.62; P = .02) (Table 2; eTable 4 in Supplement 3). The sample standard deviation of the composite from the prerandomization cohort was 0.53, hence the estimated intervention effect of 0.34 corresponds to a standardized effect of 0.64, corresponding to 5.7 additional “engaging” statements (a 44% increase), 0.6 additional responses to emotion (a 71% increase), and 1.4 additional statements regarding prognosis and treatment choices (a 38% increase).
Of the individual communication component measures, only the engaging measure (APPC) was statistically significant. There were no statistically significant effects of the intervention on the PEPPI, THC, or HCCQ scales, or on 2-year survival and curability estimates; 2-year survival discordance was 59% in the intervention group vs 62% for control; corresponding figures for curability discordance were 39% and 44%. Quality of life was stable until 6 to 9 months prior to death, with a terminal decline (eFigure 2 in Supplement 3); overall, QOL differences between intervention and control were not statistically significant. We observed no intervention effects on health care utilization.
None of the prespecified candidate effect modifiers were associated with heterogeneity in treatment effects on communication outcomes. Median survival was 16 months: 19 months in the intervention group and 14 months in the control (hazard ratio, 0.84; 95% CI, 0.61-1.15) (eFigure 3 in Supplement 3).
In this study, a brief combined intervention targeting physicians, patients with advanced cancer, and their caregivers (when available) promoted patient-centered communication in the near term, with clinically meaningful increases in engaging patients in discussions, responding to emotions and discussions of prognosis and treatment choices. These communication domains are linked; provision of information or emotional support, for example, may depend on a patient cue or request as well as a clinician’s willingness and capacity to respond. Of the 4 communication domains, the most fundamental, engaging patients as active partners in care—being assertive, asking questions, requesting clarification, expressing opinions and preferences to a greater degree than control patients—was independently significant in secondary analyses. Our approach was individualized and tailored to participants’ educational needs; it was theory-based, highly rated by patients, caregivers, and oncologists, and focused on important domains of patient-centered communication but whose incorporation into practice remains elusive.1,9-11 While prior reports suggest that activated patients and those receiving bad news may rate their physicians more harshly,8,57 we did not observe these effects, perhaps because our intervention focused on aligning patient, caregiver, and physician expectations.
Despite calls for improved patient-centered communication between oncologists and their patients and evidence that linking end-of-life discussions with more realistic prognosis estimates, better QOL and reduced utilization of aggressive treatments,8-12 there has been little headway over the past 20 years. Many patients hold unrealistically optimistic prognostic estimates,8,11,21,58,59 which they mistakenly believe their physicians share60; future studies can unravel how to interrupt the temporarily adaptive but ultimately dysfunctional pas de deux, in which physicians, caregivers, and patients avoid, euphemize, or misinterpret these discussions.61-64 Oncologists need better training in the provision of information to patients with varying levels of health numeracy and literacy as well as “terror management,” a defense mechanism that may prompt some patients (and physicians) to respond to fear of death through avoidance and selective attention.65-67 Venues already exist for communication and awareness training during residency and fellowship,25,68,69 and interventions such as ours are feasible for practicing oncologists.
Consistent with prior data,70,71 QOL is remarkably stable during the course of cancer, until the terminal decline. It is possible that the timing of the intervention may have not have been optimized to affect QOL trajectories. Future patient and caregiver interventions might be targeted to key junctures in the clinical course, such disease progression, symptoms, or early declines in QOL. Because patient-centered timing of interventions poses logistic and methodological difficulties, training existing office personnel to coach patients might better adapt to patient needs. While outpatient palliative care consultations may improve QOL, widespread implementation for all patients with advanced cancer is unrealistic; oncologists still need to communicate disease-related information clearly, respond to emotions, help patients make choices, and facilitate referrals when indicated.72
We observed no intervention effects on utilization. The 2 study sites have moderate to high use of palliative care and low to moderate use of aggressive interventions,73 possibly limiting room for improvement. As expected, utilization outcomes clustered by physician, suggesting that addressing underlying physician attributes74and institutional norms75-77 might also be needed to address utilization of aggressive interventions and hospice.
In addition to the study limitations addressed herein, the choice of only 2 study sites may limit generalizability, and the use of more than 1 audio recorded office visit and different measures of patient-centered communication57 may have revealed patterns that were not observed here. Median survival in this study was 16 months, longer than anticipated, during which time the intervention effects may have waned. Lengthier physician interventions may have reinforced skills more effectively, but at a price: longer training could limit participation to only the most motivated physicians. Similarly, longer or more intensive patient interventions might not be feasible for patients who are symptomatic or close to death.
Although clinician-patient communication patterns are difficult to change,78 an intentionally brief communication intervention was effective in improving patient-centered communication in advanced cancer but requires refinement in focus, delivery, support, or timing to promote shared understanding, QOL, and appropriate use of health care at the end of life. The current productivity-oriented practice environment also presents barriers to effective communication. Changes are needed in medical education and health systems to provide communication skills training for physicians, meaningful support for them to participate, and trained personnel to coach patients so that their voices can be heard.
Corresponding Author: Ronald M. Epstein, MD, Center for Communication and Disparities Research, Department of Family Medicine, University of Rochester Medical Center, 1381 South Ave, Rochester, NY 14620 (email@example.com).
Published Online: September 9, 2016. doi:10.1001/jamaoncol.2016.4373
Author Contributions: Dr Epstein had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Concept and design: Epstein, Duberstein, Fiscella, Hoerger, Tancredi, Gramling, Franks, Kaesberg, Back, Butow, Walczak, Tattersall, Venuti, Hoh, Kravitz.
Acquisition, analysis, or interpretation of data: Epstein, Duberstein, Fenton, Fiscella, Hoerger, Tancredi, Xing, Gramling, Mohile, Franks, Kaesberg, Plumb, Cipri, Street, Shields, Butow, Sullivan, Robinson, Lewis, Kravitz.
Drafting of the manuscript: Epstein, Duberstein, Xing, Sullivan, Robinson.
Critical revision of the manuscript for important intellectual content: Epstein, Duberstein, Fenton, Fiscella, Hoerger, Tancredi, Gramling, Mohile, Franks, Kaesberg, Plumb, Cipri, Street, Shields, Back, Butow, Walczak, Tattersall, Venuti, Hoh, Lewis, Kravitz.
Statistical analysis: Epstein, Duberstein, Hoerger, Tancredi, Xing, Gramling, Franks.
Obtained funding: Epstein, Duberstein, Fiscella, Hoerger, Kravitz.
Administrative, technical, or material support: Epstein, Duberstein, Hoerger, Kaesberg, Plumb, Cipri, Street, Back, Tattersall, Venuti, Robinson, Hoh, Lewis, Kravitz.
Study supervision: Epstein, Duberstein, Fenton, Hoerger, Mohile, Plumb, Cipri, Butow, Sullivan, Kravitz.
Conflict of Interest Disclosures: None reported.
Funding/Support: This work was supported by a grant from the National Cancer Institute (R01 CA140419-05; co–principal investigators: Drs Epstein Kravitz). Dr Duberstein is supported by a grant from the National Cancer Institute (R01CA168387). Dr Hoerger is supported by grants from the National Institutes of Health (T32MH018911, U54GM104940).
Role of the Funder/Sponsor: The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Additional Contributions: We thank Gary Morrow, PhD (University of Rochester [UR]), Timothy Quill, MD (UR), and Ted Wun, MD (University of California, Davis [UCD]), for their input on study design; Amy Abernethy, MD (Duke University), for her help with developing the intervention video; Rimaben P. Cabrera, MSW (UCD), Gisela Escalera, MSW (UCD), Iwona Grzela-Juskiewicz, MD (UR), and Martha Tuttle, BA (UR), for recruiting, administering surveys to and following the study patients and their caregivers; Andrew Cyrus Marcy, RN (UR), for medical record abstraction; Don Gaudion, MBA (UR), and Blair Leatherwood, MFA (UCD), for their role in physician training sessions; Charles Kamen, PhD (UR), Rachel Rodenbach, MD (UR), and Josef Bartels, MD (UR), for intervention fidelity assessments; Emma Perry, BA (Texas A&M), and Lindsay N. Fuzzell, MA (Purdue), for supervising the coding of audio recordings and transcripts; Patrick Davis, BS (UR), and Emma Pollock (UR) for coding transcripts and data management; Joseph Duckett (UR) for programing and database management; and Jennifer H. Kim, BA, Morgan McDonald, Jacob Paulson, BS, Syed M. Sajjad, BS, and Juliet Wu, BS (all at UR), for data entry and management. All received payment for their work on the study with the exception of Drs Morrow, Kamen, and Wun. In addition, we thank Mary Lou Pollock (UR) and Carolyn Coleman, BS (UCD), who provided financial and administrative oversight, and Donna Geil, AAS (UR), and Mahala Ruppel, BS (UR), for administrative support. Many thanks to the oncologists, their staff, the patients, and their caregivers who participated in this study.
Previous Presentations: This study was presented at the ASCO Palliative Care in Oncology Symposium; September 9, 2016; San Francisco, California.