Roter DL, Larson S, Fischer GS, Arnold RM, Tulsky JA. Experts Practice What They PreachA Descriptive Study of Best and Normative Practices in End-of-Life Discussions. Arch Intern Med. 2000;160(22):3477-3485. doi:10.1001/archinte.160.22.3477
Advance directives (ADs) are widely regarded as the best available mechanism to ensure that patients' wishes about medical treatment at the end of life are respected. However, observational studies suggest that these discussions often fail to meet their stated goals.
To explore best practices by describing what physicians who are considered expert in the area of end of-life bioethics or medical communication do when discussing ADs with their patients and to explore the ways in which best practices of the expert group might differ in content or style from normative practice derived from primary care physicians' discussions of ADs with their patients collected as part of an earlier study.
Nonexperimental, descriptive study of audiotaped discussions.
Outpatient primary care practices in the United States.
Eighteen internists who have published articles in the areas of bioethics or communication and 48 of their patients. Fifty-six academic internists and 56 of their established patients in 5 practice sites in 2 locations—Durham, NC, and Pittsburgh, Pa. Eligible patients were at least 65 years old or suffered from serious medical illness and had not previously discussed ADs with their physician. Expert clinicians had discretion regarding patient selection, while the internists chose patients according to a predetermined protocol.
Coders applied the Roter Interaction Analysis System (RIAS) to audiotapes of the medical visits to describe communication dynamics. In addition, the audiotapes were scored on 21 items reflecting physician performance in specific skills related to AD discussions.
Experts spent close to twice as much time (14.7 vs 8.1 minutes, P<.001) and were less verbally dominant (P<.05) than other physicians during AD discussions. When length of visit was controlled statistically, the expert physicians gave less information about treatment procedures and biomedical issues (P<.05) and asked fewer related questions (P<.05) but tended toward more psychosocial and lifestyle discussion and questions. Experts engaged in more partnership building (P<.05) with their patients. Patients of the expert physicians engaged in more psychosocial and lifestyle discussion (P<.001), and more positive talk (P<.05) than patients of community physicians. Expert physicians scored higher on the 21 items reflecting AD-specific skills (P<.001).
Best practices as reflected in the performance of expert physicians reflect differences in measures of communication style and in specific AD-related proficiencies. Physician training in ADs must be broad enough to include both of these domains.
WHILE NOT fail-proof,1,2 advance directives (ADs) are widely regarded as the best available mechanism to help ensure that patients' wishes about medical treatment at the end of life will be respected when they become incapable of making decisions.3,4 The advance care planning process has been portrayed in the bioethics literature as anticipatory decision making, similar to traditional informed consent with a focus on specific treatment options within varying scenarios of mental capacity and physical state.5- 7 Many have worried, however, that while such a task-focused approach may yield a consistent, even adamant preference statement, it may not produce patient insight or understanding of the options chosen.8,9 An alternative, process-focused model attempts to give meaning to choices by focusing on patient self-discovery through discussions in which patient values and experiences guide informed decision making.10,11
The few studies that have directly observed physician-patient conversations about end of life suggest that they often fail to meet either of the goals described previously.12- 16 For example, in studies of both residents with hospitalized patients12 and attending physicians with outpatients,13,14 descriptions of therapy options were too vague to ensure informed patient choice and patient preferences were not tied to specific situations or contingencies. Physicians and patients infrequently went beyond the narrow biomedical and legal technicalities of their discussion to the psychosocial context, rarely eliciting personal values or goals for care12,13 or patient motivation for particular decisions.15,16 Not surprisingly, physicians were rarely able to accurately predict patient choice after such discussions, and patients often had misconceptions about simple medical facts.14
It is unclear why the quality of AD conversations is so poor. It may be unrealistic to ask physicians and patients to talk about so profound an issue as death in care settings that are often characterized by time restrictions. Furthermore, many physicians receive inadequate training in interpersonal skills. These limitations would suggest that attempts to improve these conversations likely to be futile. Yet, the literature does describe positive models of end-of-life planning and suggests that AD discussions can be successful if structural elements of the visit, including the dedication of sufficient time to the task and necessary interviewing skills, are in place.8
This study was designed to explore best practices by describing what physicians who are considered expert in the area of end of life bioethics or medical communication do when discussing ADs with their patients. A second objective was to explore the ways in which best practices of the expert group might differ in content or style from normative practice derived from primary care physicians' discussions of ADs with their patients collected as part of an earlier study.13,14
Eighteen internists and family physicians, 9 nationally recognized as expert in medical ethics, 7 recognized as expert in physician-patient communication, and 2 recognized as expert in both areas, were recruited by us and participated in the study. To qualify as an expert medical ethicist, the physician must have published at least 2 articles on end-of-life decision making in peer-reviewed journals. Inclusion criteria for the communication specialists were fellowship in the American Academy on Physician and Patient, a national organization devoted to improving physician-patient communication, and publication of at least 2 articles on physician-patient communication. In the 10 years prior to participation in the study, the physician-participants published a total of 105 articles directly on the topic of ADs and an additional 54 articles on relevant areas of communication studies. Included in the list were 5 directors of university medical ethics programs, 5 directors of primary care programs, and authors of authoritative textbooks on medical ethics and communication. All physicians maintain an outpatient practice in settings across the country.
We compared the AD discussions of these expert physicians with previously recorded discussions of community primary care internists.13,14 This sample of physicians (referred to as the community physician sample) included internists at 5 practice sites in 2 locations—Durham, NC, and Pittsburgh, Pa. Fifty-six of the 60 eligible physicians agreed to participate. The practice sites included 2 university-based general medicine faculty practices, 2 Veterans Affairs general medicine faculty practices, and 1 university-based geriatrics practice. All physicians held medical school appointments, worked within an academic system, and were actively engaged in patient care.
Eighteen of the 20 expert physicians approached by investigators participated in the study. As listed in Table 1, expert physicians were advanced in their career; they averaged 49 years of age with a median of 20 years experience in medical practice. All of the expert physicians were white, and most (72%) were men. The physicians reported having a median of 6.5 outpatient AD discussions and 3 inpatient AD discussions over the 3-month period preceding the study. In contrast to the expert physician sample, the community sample physicians were considerably younger, averaging 37 years, and earlier in their career, having a median of 10 years experience in medicine. Unlike the expert sample, community physicians were almost equally as likely to be female (44%) as male (56%). The physicians reported having a median of 1 outpatient AD discussion and 2 inpatient AD discussions during the 3-month period preceding the study.
Patients were eligible for the study if they were at least 65 years old or suffered from a serious medical illness that made a discussion of ADs relevant, including cancer, prior cardiac arrest, human immunodeficiency virus infection, renal insufficiency (serum creatinine level >250 µmol/L [>3 mg/dL] or receiving chronic dialysis), or chronic obstructive pulmonary disease, congestive heart failure, or cirrhosis severe enough to cause 2 hospitalizations in the past year. They had to speak English, to be judged competent by their physician to make medical decisions, and to have not previously discussed ADs with their physician.
For the community study, the investigators randomly ordered a list of eligible patients scheduled for a given clinic session. Physicians were asked to discuss ADs with the first patient on the eligible list for whom such a discussion was considered appropriate by the physician within the context of their relation to the patient and the day's visit agenda. No specific instructions were given to the physicians other than encouragement to conduct the AD discussion in whatever way they preferred, and the investigators obtained informed consent from the patients.
Because of the logistical problems associated with data collection for the national expert sample, patient recruitment proceeded differently. The research team did not select eligible patients; rather the expert physicians themselves were asked to select 3 patients from their practices who met the selection criteria and for whom discussion of ADs was appropriate. No specific instructions were given to the physicians other than encouragement to conduct the discussion in whatever way they felt most comfortable. The expert physicians were asked to obtain informed consent from the patients and to record their medical visits on audiotape.
Eighty-nine patients were eligible for participation in the community physician study. In 8 cases, physicians opted to exclude an eligible patient from the study because they thought that the patient was emotionally unstable or had too few prior visits with them. Fifty-six (69%) of the 81 recruited patients agreed to participate in the study. Expert physicians directly recruited patients, and there were no reports of patient refusals for participation.
The expert physicians enrolled a total of 48 patients into the study; 6 physicians audiotaped 2 patients each and 12 audiotaped 3 patients each. As shown in Table 1, the patients averaged 74 years of age, were preponderantly white (78%), and were about as likely to be male (54%) as female. Most patients were well educated, with only 7 (14%) reporting less than a high school education. As a whole, patients viewed their health quite positively with a minority (23%) of self-ratings falling in the fair or poor category. In contrast to the patients of the expert physician sample, the community physicians' patients (n = 50) were more likely to be male (68%) and to have less than a high school education (66%) and had an average age of 72 years. The patients' self-ratings of health were less optimistic than that of the other patients, with 48% falling in the fair or poor category (Table 1).
Fifty audiotapes obtained from the community physician sample and 48 audiotapes obtained from 18 physicians from the expert sample were available for analysis in this study. Only the portion of the audiotaped visit directly relevant to the AD discussion was coded.
The unit of analysis for the study is the physician. This approach is straightforward for the community physician sample as each patient-physician encounter is independent of the others, with only 1 patient recorded with each physician. For the expert physician sample, however, this is not the case; individual physicians participated in more than 1 encounter (an average of 2.7 per physician), so that patient encounters are not independent of physician. We aggregated the expert physician data by physician so that our analysis is based on the average performance of each expert physician over their 2 or 3 patients.
The audiotapes were analyzed using the Roter Interaction Analysis System (RIAS), a widely used quantitative approach to audiotape coding. In addition to the RIAS, a content analysis focusing on key skills related to proficiency in the conduct of AD discussions was undertaken.
The RIAS places each complete thought, usually expressed as a phrase during the visit, by either patient or physician, into mutually exclusive and exhaustive categories. The communication categories relate broadly to the task-focused and socioemotional functions of the visit. Task-focused communication includes categories of patient education and counseling about the medical condition and symptoms, treatment regimens and procedures, directions and orientations, as well as lifestyle and psychosocial issues. Socioemotional communication includes responding to patient emotions, positive and negative exchange, and partnership building. Table 2 gives the individual code categories included within each of the communication groupings and examples of dialogue for each coding category.
Coding of the audiotapes is done directly from audiotapes without transcription, and coding averaged about 1.5 times the length of the visit. Two coders with more than 5 years of experience with the RIAS, shared the coding task. As in other studies, coding reliability was good.17- 19 Reliability was assessed by calculating a Pearson correlation coefficient for a 13% random sample of audiotapes (n = 13) selected throughout the coding period for all categories with frequencies greater than 1.0. The mean correlation representing reliability for the coding of physician categories is 0.86 (range, 0.77-0.99), and the reliability for coding of patient categories is 0.85 (range, 0.75-0.99).
A measure of verbal dominance during the medical visit was derived by calculating a ratio of the count of physician statements divided by a count of patient statements. Discussion length in seconds was derived from the audiotape. This measure excluded waiting time.
In addition to the more general stylistic elements of communication captured through the RIAS analysis, a number of critical content-specific elements reflective of proficiency in AD discussions were also coded. These elements were identified by us through a review of the AD literature and later refined and operationalized on a subset of the audiotapes, and a code manual with dialogue examples and coding rules was developed. Twenty-one items representing 5 broad content domains were scored as either present or absent with possible scores ranging from 0 (none present) to the number of items included in the domain (all present). (A listing of the items within each domain is included in Table 3.)
The content domains included the following: (1) Probes and elicits the patient's preferences referring to specific scenarios (4 items; Cronbach α = 0.58). (2) Probes and elicits the patient's values, beliefs, and experiences (8 items; Cronbach α = 0.64). (3) Provides support for the decision-making process (3 items; Cronbach α = 0.43). (4) Provides resources and encouragement toward decision resolution (4 items; Cronbach α = 0.58). (5) Effectively closes the visit: includes summarization at the close of the discussion (1 item) and solicitation of further questions or concerns in closing (1 item). The above skill domains were conceptualized as representing complementary areas of proficiency and were combined into a single 21-item scale with a Cronbach α of 0.68.
Coding was done directly from audiotape without transcription. Reliability of coding was assessed through double coding of a 13% (n = 13) random sample of audiotapes selected throughout the coding period. Agreement between coders on the individual items was good (average κ = 0.88; range 0.500-1.00).
Descriptive analysis of RIAS-coded communication and the content-specific AD skills variables was conducted using both 1-way analysis of variance and analysis of covariance (ANCOVA).
Overall, experts averaged 6.6 minutes longer in AD discussions than other physicians: 14.7 minutes for experts (median = 11.7; range, 7.4-26.3 minutes) and 8.1 minutes for community physicians (median = 6.0; range, 0.92-38.4 minutes) (t = 3.7; P<.001). Almost two thirds (n = 32; 62%) of the community physicians had discussions of less than 7 minutes, while all expert physicians' discussions exceeded 7 minutes. Both groups had a small minority of physicians whose discussions exceeded 20 minutes, with the longest of these visits lasting 26 minutes for the expert physicians and 38 minutes for the community physicians.
Typically, discussions about ADs were set apart from the routine business of the medical visit, either as the primary focus of the visit or as an addendum to the visit. Most experts (14/18 [78%]) used a single visit structure across all of their study patients. The most common model for expert physicians was to dedicate the entire medical encounter to discussion of ADs. Eleven (61%) of the expert physicians used this approach with all of their recorded patients, whereas only 2 of the community physicians did so. The dedicated visits averaged 15 minutes (SD = 6.8 minutes) for experts and 8.8 minutes (SD = 0.8 minutes) for community physicians. In contrast, more than 80% (n = 41) of the community physicians appended the AD discussion to the end of the visit after all other visit business was completed, but only 2 of the experts did so with all of their patients. The appended visits averaged 7.1 minutes (SD = 6.3 minutes) for community physicians and 9.7 minutes (SD = 0.5 minutes) for experts. Several community physicians (n = 7) and only 1 expert integrated discussion of ADs into the broader medical dialogue, such as discussing a test result or a prescription, or even conducting a short physical examination, in the midst of the AD discussion.
The expert physicians not only spent more time in AD discussions, but also the time was spent in more dialogue. Both experts and their patients talked more during visits than general physicians and their patients; however, the proportionate contribution of each was somewhat different. The ratio of physician to patient talk shows expert physicians to be significantly less verbally dominant in these discussions than their counterparts (1.3:1 for experts compared with 1.7:1 for community physicians [t = 2.2; P<.03]).
What physicians talked about during their time with patients is given in Table 4. As is shown, expert physicians demonstrated higher mean frequencies for all of the RIAS-coded categories of interaction, with the exception of information on treatment and procedures. Statistically significant differences were found in the task-focused categories of lifestyle information and psychosocial counseling and in the number of psychosocial questions asked. In addition, significant differences are evident in all of the socioemotional categories, including partnership building, emotional exchanges, positive talk, negative talk, and social talk. Since higher category frequencies are related to a longer length of visit and expert physicians conducted longer AD discussions, ANCOVA controlling for the length of visit was undertaken, and adjusted means are also displayed in the table. The adjusted means show higher discussion levels for community physicians of medical treatment and procedures and related medical counseling, as well as higher levels of biomedical questions and instructions and orientations. Higher levels of lifestyle information and psychosocial counseling, and socioemotional areas of exchange, including responding to patients' emotion, positive talk, and social talk, by expert physicians are less pronounced when the length of visit is controlled, but the patterns are similar. Partnership building and disagreements and criticisms (negative talk) remain significantly higher among expert physicians regardless of whether adjusted or unadjusted means are examined.
Table 5 shows the unadjusted and adjusted mean frequencies for coded categories of patient interaction during the AD discussions. As was true for the expert physicians, the patients of expert physicians show significantly higher frequencies of interaction for all categories when unadjusted means are examined. When adjusted means are examined, differences emerge showing higher levels of interaction related to biomedical topics and question asking for patients of community physicians. Higher frequencies for patients of expert physicians in the categories of psychosocial and lifestyle topics, emotional talk, and positive talk show similar patterns with both unadjusted and adjusted means.
Analysis of content-specific AD skills showed higher mean levels for expert physicians across the board; statistically significant differences were evident in 3 of the 5 dimensions, as listed in Table 3 and in the overall score. Expert physicians more frequently probed and elicited patients' values and experiences related to end of life, provided more resources and encouragement for decision making, and more effectively concluded the visit through summarization and final checking for additional questions and concerns. After statistically controlling for the length of visit, the differences in mean scores within dimensions were somewhat diminished, but the patterns were consistent. The difference in the overall score remained statistically significant.
We believe we have captured best practices of physicians considered expert in end-of-life issues and medical communication, as well as normative practice as reflected in the AD discussions of well-respected primary care physicians who provide care to patients in community settings. These differences are most obvious in the way in which the visits were structured and in the amount of time devoted to the discussions, as well as in the relative emphasis on biomedical vs psychosocial aspects of care. Differences were also evident in specific AD skills used in the discussions.
Experts spent substantially more time, close to twice as much, in AD discussions than community physicians did. Longer length of visit is associated with higher mean frequencies of almost all of the coded categories, and consequently we conducted analysis statistically controlling for length of visit to understand differences in communication dynamics over and simple above the provision of more time. However, the time differences evident in our analysis represent real effects—patients experience visits in real time and expert physicians spent more of it with their patients. For that reason, both unadjusted and adjusted mean frequencies of interaction categories were presented.
Differences in time devoted to AD discussions were most marked by the preponderance of very short exchanges for the community physicians rather than particularly long discussions for the expert physicians; none of the expert physicians conducted discussions of less than 7 minutes, while almost two thirds of the community physicians did so. Emanuel and colleagues20,21 estimate that a thorough AD discussion would take a median of 14 minutes. This estimate was based on their experience in AD discussions with 405 outpatients using the Medical Directive, a comprehensive advance care document focusing on paradigmatic scenarios defined by prognosis and disability of incompetent patients, to guide the discussion.24,25 While there was no indication that the physicians in our study used the Medical Directive document per se, the time spent in discussion is comparable. In a similar vein, expert physicians seem to agree with the suggestion that a full AD discussion may be best accomplished in a dedicated visit.21
While the length of AD discussion represents the contributions of both patients and physicians to the visit's dialogue, these are not of equal magnitude. Expert physicians were less verbally dominant in the recorded discussions than were community physicians. The more balanced dialogues of the expert physician visits were associated with physicians' use of partnership-building skills that invite the expression of patient opinion, expectations, understanding, and beliefs. Use of these partnership skills was furthermore associated with a greater psychosocial rather than biomedical emphasis in the visit. When patient communication is examined, a mirror image to physician communication emerges. Patients of expert physicians disclose more psychosocial information to their physicians, offer more emotional talk, and seem more positive. This is a likely reciprocation of the expert physicians' higher use of these categories.
Community physicians did as well as experts in probing and eliciting patient preferences in response to scenarios. This is perhaps the aspect of AD discussions that is most widely publicized. However, these probes were only done in about half of the recorded AD discussions, and the discussion was typically abstract (ie, "do everything" or "if I am a vegetable I do not want machines"). Experts were more proficient at what might be considered the higher order skills measured in the study, including probing and eliciting patients' values and experiences, and clarifying and summarizing the discussion at its close. It is in this realm that we think patient-centered communication skills complement and extend physicians' ability to address the broad spectrum of concerns and issues that patients face.22- 25 It may be worth rethinking the goals of advance care planning discussions to emphasize these tasks.
Several limitations should be considered in the interpretation of our findings. Differences in patient recruitment methods across the groups may have introduced a systematic bias to the study. The expert physicians maintained complete control over patient recruitment and had the flexibility of scheduling a patient visit specifically to discuss ADs. It seems clear that many expert physicians did just this. Furthermore, it is possible that they chose patients they considered more positively disposed toward an AD discussion or patients with whom they had especially good relationships. Nevertheless, the expert physicians reported that both the patients chosen and the AD discussions were representative of their usual practice. In contrast, study personnel identified and randomly ordered eligible patients from the daily list for the community physicians. Physicians were able to skip over patients they considered inappropriate for AD discussions but were required to consider each patient in order as identified by the investigators. Consequently, the community physicians did not have the opportunity to schedule a visit for study patients that would be dedicated to an AD discussion. Inasmuch as the community physicians reported that the patients and AD discussions that were included in the study were representative of their usual practice, it is unlikely that we missed any significant common practice that would include scheduling dedicated visits for the purpose of AD discussion.
There are other reasons that may lead to differences in patient selection. Experts as a group may consider patients appropriate candidates for AD discussions at earlier stages of chronic disease and when in better health than their community counterparts. Despite use of the same broad patient eligibility criteria, some sociodemographic differences between the 2 patient groups were evident; patients of expert physicians were better educated, were more likely to be female, and were more optimistic in their health ratings than patients of community physicians. We cannot judge the extent to which these differences are the result of selection bias in the ways described or simply a reflection of the patient populations served by physicians in the 2 groups. Our guess is that all of these factors may be in operation. Either way, we recognize that differing patient characteristics in the 2 groups may affect the results. Together, the fact that there are patient differences qualifies the interpretation of our findings to discussions of expert physicians as ideal or physician-controlled circumstances compared with the less ideal circumstances encountered by community physicians.
The expert group of physicians was recruited on the basis of their professional reputation and publications; however, there may be other ways in which the 2 groups of physicians differ. For instance, as one might expect, the expert physicians are considerably older than the others, perhaps marking greater professional maturity and experience. An expression of this maturity may well be in more patient-centered communication style. We captured multiple observations of AD discussions for the experts, but only one conversation each was recorded for the community physicians. Multiple visits are important when describing the behavior of an individual physician or when the numbers of physicians in a group are small, as is the case for our expert sample. In drawing group comparisons to the much larger sample of community physicians, the necessity of multiple observations per physician is less compelling. Consequently, limited numbers of observations per community physician were undertaken.
While it is true that we may have captured "best behavior" in our audiotapes, it is unlikely that this was systematically interpreted by physicians in a way that would jeopardize the interpretation of our findings. The issue of performance bias in response to audiotape recording has been addressed in several studies.26- 28 All have found that the effect is minimal. Included among these is a study of the content of videotape recordings of physicians who were and were not informed that recordings were being made that found no statistically significant differences in the length of visit or in the number or nature of the problems discussed.28
A strength of the study is the high participation rate of both community and expert physicians. Nearly every physician recruited into the community sample, which included physicians from 5 practices in 2 cities, agreed to participate in the study, as did almost all of the national experts.
We are unable to generalize our conclusions regarding differences between expert physicians and community physicians beyond best and normative practices because of the different procedures used for patient selection in each group. Nevertheless, there is much to learn from the experts' methods. In listening to these national experts, we have found that their communication reflects mastery of the content area related to end-of-life planning, but in equal weight, we found that they excel in use of a patient-centered communication style. We think this finding has important implications for how we might best improve physicians' ability to discuss end-of-life issues. It is not sufficient to simply train physicians in the narrow and specific elements of AD guidelines; if we want to fully prepare physicians in this arena, we will have to broaden training to include the array of patient-centered skills that serve physicians so well in all their clinical encounters.
Accepted for publication June 14, 2000.
This work was supported by a grant from COMSORT (Dr Roter) and grant HFP82-008 from the Veterans Affairs Health Services Research and Development; grant 5-P60-AG11268 from the National Institute on Aging Claude D. Pepper Older Americans Independence Center; the John A. Hartford Foundation; the Greenwall Foundation (Dr Tulsky); the R. K. Mellon Foundation; and the LAS Trust (Dr Arnold). Drs Arnold and Tulsky are Project on Death in America Soros Faculty Scholars, and Dr Tulsky is a Robert Wood Johnson Generalist Physician Faculty Scholar and an Veterans Affairs Career Development Awardee.
We are indebted to 178 patients and their physicians who graciously allowed us to observe them during a private moment. We are also grateful to Bernard Lo, MD, for assistance with design and implementation of the expert study and thoughtful comments on the manuscript.
Corresponding author and reprints: Debra L. Roter, DrPH, Johns Hopkins School of Hygiene and Public Health, 624 N Broadway, Baltimore, MD 21205 (e-mail: firstname.lastname@example.org).