What is the Risk of Distress in Palliative Care Survey Research?

Transcription

1 Vol. 28 No. 6 December 2004 Journal of Pain and Symptom Management 593 Original Article What is the Risk of Distress in Palliative Care Survey Research? Jennifer Takesaka, BA, Roxane Crowley, BA, and David Casarett, MD, MA Division of Geriatric Medicine (J.T., R.C., D.C.); The VA Center for Health Equity Research and Promotion (D.C.); the Center for Bioethics (D.C.); and the Leonard Davis Institute of Health Economics (D.C.), University of Pennsylvania, Philadelphia, Pennsylvania, USA Abstract To determine whether caregivers believe that interviews about end-of-life care are distressing and to identify patient and respondent characteristics associated with an increased risk of distress, distress was assessed in four studies that used family interviews. The setting was four Medicare certified hospice organizations, the University of Pennsylvania Memory Disorders Clinic, and two nursing homes, and participants included 296 family members of seriously ill or recently deceased patients. For three of the studies, respondents described their distress on a 5-point scale. Distress was reported as either present or absent in the fourth study. Sixty-four respondents (22%) reported experiencing distress. Intensity of distress was higher for younger respondents (Spearman rho 0.16; P 0.013), younger patients (Spearman rho 0.28; P 0.001), and family members of patients with cancer (mean 0.55 vs. 0.24; Rank sum test P 0.001). In a multivariable model, after adjusting for study population, younger patient age and cancer diagnosis were independently associated with the severity of distress. Sensitive questions about death and dying are unlikely to cause distress for family members. Although the likelihood of distress is low overall, investigators recruiting from these populations may improve the research by incorporating methods to assess and manage distress. J Pain Symptom Manage 2004;28: U.S. Cancer Pain Relief Committee. Published by Elsevier Inc. All rights reserved. Key Words Distress, end-of-life care, research methods Introduction Recent years have seen a dramatic increase in interest in palliative care, and particularly in efforts to evaluate the quality of patients endof-life care. 1 7 Empirical research and Quality Improvement (QI) activities within this field have increasingly asked patients friends and Address reprint requests to: David Casarett, MD, MA, 3615 Chestnut Street, Philadelphia, PA 19104, USA. Accepted for publication: March 26, U.S. Cancer Pain Relief Committee Published by Elsevier Inc. All rights reserved. family members to describe a patient s death or care near the end of life. 5,6,8 10 These strategies offer a valuable way to define the elements of care that are most important to patients and family members and to identify ways that a health care system can be improved. Despite their value, these data collection strategies may raise ethical concerns about the risk of distress that they pose for participants. Some studies collect data from bereaved family members who are still actively grieving, 11,12 often within months after the patient s death /04/$ see front matter doi: /j.jpainsymman

2 594 Takesaka et al. Vol. 28 No. 6 December 2004 Other studies collect data from the families of patients who are seriously ill. Many Institutional Review Boards (IRBs) believe that family members who participate in these studies could experience distress when they are asked to remember or to anticipate the patient s last days. However, it is not known how likely respondents are to experience distress as a result of their participation in these sorts of studies. Most existing data describe the qualitative experiences (especially positive) of research participation The best existing quantitative estimate of the incidence of and correlates of distress come from a single small study of bereaved parents who had experienced the death of a child. 16 These parents were asked to reflect back on a study in which they had participated in the past. It is not known what proportion of respondents would report experiencing distress if they are asked at the time of the interview, nor is it known what patient or respondent characteristics are associated with an increased risk of distress, although data from one study suggest that associations may exist. 16 This latter question is particularly important because when a study population can be characterized in advance, these risk factors could help investigators and clinicians to more accurately assess the likelihood of distress for a particular study or QI activity and make support available (e.g., access to a bereavement counselor). Therefore, the goals of this study were to determine whether family members find questions about end-of-life care to be distressing and to identify respondent and patient characteristics that are associated with an increased likelihood and intensity of distress. Methods Distress was assessed in four studies that used surrogate interviews with family members. The data for this study were compiled from secondary analysis of two published studies 17,18 and two as-yet unpublished studies that collected data from family members of patients. Each of these studies is described below. Study 1 used mailed surveys sent to primary caregivers of consecutive patients who had died in the care of a single urban hospice (Wissahickon Hospice, Philadelphia, PA). 17 Surveys were sent either 2 or 6 weeks after the patient s death, according to a blocked randomization scheme. These surveys asked families to describe the patient s last 24 hours of life and to identify both positive and negative aspects of the care that he/she received. Study 2 used telephone interviews with respondents drawn from consecutive deaths at three hospices chosen for their geographic diversity (San Diego Hospice and Palliative Care, San Diego, CA; Hope Hospice and Palliative Care, Ft. Meyers, FL; Home and Hospice Care of Rhode Island, Providence, RI). Bereavement counselors at each of the three hospice facilities recruited family members and then an interviewer contacted those who agreed to participate. Telephone interviews were conducted 2 months after the patient s death and ranged in length from 30 minutes to one hour. An interviewer followed an interview guide that asked family members to describe the decision-making process that led to hospice enrollment and the features of hospice that they and the patient considered in deciding whether or not to enroll. Study 3 used telephone interviews of caregivers of patients with Alzheimer s Disease (AD). 18 The caregivers of the AD patients were recruited from the Memory Disorders Clinic of the University of Pennsylvania. Caregivers were contacted by telephone and asked to participate. Scripted interviews asked caregivers to describe their concerns and needs for care and support, and then asked them to describe the relative importance of various hospice services. Study 4 consisted of telephone interviews with family members of nursing home residents. Family members were identified by chart review and notified about the study by a letter that offered a refusal option. Two weeks later the family members were recruited by telephone. These interviews defined family members perceptions of residents needs for care and their preferences for life-sustaining treatment. For all 4 studies, respondents were asked at the close of the interview or mail survey whether they had experienced any distress ( Were these questions distressing to you? ). For three studies (Studies 1, 2, and 4), respondents who reported distress were also asked to rate the intensity of the distress that they had experienced on a 4-point scale ( A little, Somewhat, Very, and Extremely ).

3 Vol. 28 No. 6 December 2004 Risk of Distress in Research 595 Possible relationships between patient and respondent characteristics and distress were examined by using distress as 5-point ordinal variable (0 4, in which no distress was coded as 0 ) and as a dichotomous variable (0 1: did/ did not report distress). Because of differences in study populations and small numbers of respondents reporting distress, pooled data (Studies 1 4) were used for all analyses. All bivariate analyses used either the χ 2 test or appropriate non-parametric tests (e.g., Wilcoxon Rank sum test, Kruskal Wallis test, Spearman correlation coefficient). Logistic and ordinal logistic regression models were used to identify patient and respondent characteristics that were independently associated with the presence and intensity of distress, respectively. For both models, patient and respondent characteristics were considered for inclusion if bivariate tests reached a significance of Candidate independent variables were removed from the preliminary model sequentially, and restored to the model if their removal altered the significance of remaining variables or if their removal altered the performance of the model as indicated by a comparison using the likelihood ratio test. 20 Stata for Windows (Version 8.0, Stata Co., College Station, TX) was used for all statistical analysis. Each study was approved by the University of Pennsylvania s Institutional Review Board (IRB). Oral informed consent was obtained for interview studies (Studies 2 4), and this requirement was waived for the mail survey study (Study 1). Results Response rates were 87% for Study 1 (112/ 207), 54% for Study 2 (100/115), 71% for Study 3 (45/63), and 91% for Study 4 (39/43). Characteristics of the 296 respondents are summarized in Table 1. Respondents were predominantly white (n 201; 68%) and female (n 211; 72%), and the mean age for patients and respondents was 73 years and 60 years, respectively. For most respondents, the index patient was a spouse (105; 35%) or a parent (124; 42%). Of the 296 respondents, 290 completed the distress question and of these, 64 (22%) reported that study questions had been distressing Table 1 Patient and Respondent Characteristics a Respondent Age: mean (range) 60 (25 88) n (%) Respondent Sex: Female (n 295) 211 (72) Respondent Relationship to Patient (n 296) Child 124 (42) Spouse 105 (35) Sibling 13 (4) Niece/Nephew 10 (3) Child-in-law 10 (3) Other 11 (4) Parent 9 (3) Friend 6 (2) Stepchild 3 (1) Grandchild 3 (1) Partner 2 (1) Respondent Ethnicity (n 294) White 201 (68) African American 78 (27) Other 8 (3) Hispanic 5 (2) Asian 2 (1) Patient Age: mean (range) 73 (18 101) Patient Diagnosis (n 292) n (%) Cancer 134 (46) Dementia 76 (26) Cardiovascular Disease 39 (13) Neurological Disease 10 (3) End-Stage Renal Failure 8 (3) Stroke 8 (3) Pulmonary Disease 5 (2) Diabetes 5 (2) Acquired Immunodeficiency 3 (1) Syndrome Cirrhosis 2 (1) Failure to Thrive 1 (0.3) 1 (0.3) a Differing numbers reflect missing data (Table 2). The incidence of distress was significantly different among the 4 studies (χ 2 P 0.001) and was greatest in Study 1 (Table 2). Across all studies, respondents were more than twice as likely to be distressed if the index patient had cancer, compared to other diagnoses (44/130, 34% vs. 20/157, 13%; χ 2 P 0.001). Respondents who reported distress were younger (median age 56 vs. 61 years; Rank sum test P 0.011) and had been caring for younger patients (median age 72 vs. 77 years; Rank sum test P 0.002). Distress was not related to ethnicity or to the respondent s relationship to the patient. To adjust for confounding of these variables by study participation, the variables above were adjusted for study. Of the variables that were associated with the incidence of distress in bivariate analysis, cancer

4 596 Takesaka et al. Vol. 28 No. 6 December 2004 Table 2 Incidence of Distress Study 1 Study 2 Study 3 Study 4 (n 107) a (n 100) (n 45) (n 38) All 4 Studies n(%) n(%) n(%) n(%) (n 290) Any distress 48 (45) 9 (9) 3 (7) 4 (11) 64 (22) Degree of distress (n 71) A little 27 (38) 5 (7) N/A 1 (1) 33 (11) Somewhat 15 (21) 4 (6) N/A 3 (4) 22 (8) Very much 4 (6) 0 (0) N/A 0 (0) 4 (1) Extremely 2 (3) 0 (0) N/A 0 (0) 2 (1) a Distress response was missing for 6 respondents. (odds ratio 3.71; P 0.001), patient age (odds ratio 0.97; P 0.012) and respondent age (odds ratio 0.97; P 0.010) remained significantly associated with the likelihood of distress. In order to identify patient and respondent characteristics that were independently associated with the experience of distress, a logistic regression model was developed and refined. In this model, after adjusting for study and considering other variables for inclusion in the model, only a cancer diagnosis remained significantly associated with a greater incidence of distress (odds ratio 3.61, 95% confidence interval ; P 0.001). Similar relationships were observed when intensity of distress was examined as an ordinal variable (0 4 scale). The intensity of distress was also significantly different among the 3 studies for which intensity was assessed (Kruskal-Wallis test P 0.001) and was also greatest in Study 1 (Table 2). Intensity of distress was inversely associated with respondent age (Spearman rho 0.16; P 0.013) and patient age (Spearman rho 0.28; P 0.001). It was also higher for respondents when the index patient had cancer than it was for other diagnoses (medians 0 in both groups, mean 0.55 vs. 0.24; Rank sum test P 0.001). As with the dichotomous distress variable that defined the presence or absence of distress, to adjust for confounding of these variables by study participation the variables above were adjusted for study. Again, cancer (odds ratio 3.11; P 0.001), patient age (odds ratio 0.96; P 0.001), and respondent age (odds ratio 0.97; P 0.012) remained associated with the intensity of distress after adjusting for study. In order to identify patient and respondent characteristics that were independently associated with the intensity of distress, an ordinal logistic regression model was developed and refined. In this model, after adjusting for study only patient age and a cancer diagnosis were associated with greater distress (patient age odds ratio 0.96, 95% confidence interval: , P 0.003; cancer odds ratio 2.00, 95% confidence interval: , P 0.045). Discussion Research and QI activities that elicit bereaved family members perceptions of a patient s dying experience can provide valuable insights into the adequacy of health care providers efforts to facilitate a good death. However, these respondents are typically still in active phases of mourning several months after a loved one s death. 11,12,21 Therefore, IRBs may express concerns about the distress that they may cause. 22,23 The results of the current study provide important data that should help investigators to address these concerns in at least three ways. First, the results of this study suggest that data collection from family members does pose a small risk of distress. This risk should be considered in the ethical review of planned studies just as the risks of experimental procedures or medications are considered. 24 In addition, it seems reasonable to include a description of these risks in the study s informed consent process. Second, it should be noted that only a minority of respondents reported any distress, and those who did generally reported distress that was mild or moderate. This suggests that these studies may be much less distressing to study subjects than researchers and IRBs might expect. These findings should ameliorate concerns that many providers may have about collecting data from bereaved family members. 23 Furthermore, it seems likely that these small

5 Vol. 28 No. 6 December 2004 Risk of Distress in Research 597 risks are outweighed by the benefits that family members may experience when they have the opportunity to discuss these issues. Third, and most important, this study found that the likelihood of distress is not uniform, and that the likelihood and intensity of distress are associated with patient and respondent characteristics and study characteristics. This finding is important because it suggests that the probability of distress may be very low in certain populations (i.e., older family members, family members of older patients, and family members of patients who die of non-cancer illnesses). One result the finding of a higher incidence of distress among women is consistent with the results of a previous study, 16 which suggests that these findings may be generalizable to other populations. Studies that recruit respondent/patient pairs with these characteristics, particularly when interviews are used, appear to pose a low risk of distress and should not raise concerns. Although the associations observed here with age, diagnosis, and study are strong, the reasons for these associations are not at all obvious. It is possible that younger patients and those with non-cancer diagnoses underwent more sudden periods of decline. Their deaths may have been less expected, and thus family members may not have had sufficient time for grieving by the time these data were collected. The reasons for the higher incidence of distress in Study 1 are equally unclear. It is possible that mail surveys allow more time to reflect and consequently are more likely to cause distress. It is also possible that these respondents were more comfortable in admitting distress anonymously. Conversely, respondents who participated in interviews may have underreported their distress in order to avoid appearing critical of the interviewer. Finally, it is also possible that the interviewer s presence was therapeutic and ameliorated the distress caused by the questions in the other studies (Studies 2 4). Although these explanations are plausible, further research is needed to explore the mechanisms that underlie the associations reported here. When studies or QI activities sample populations in which the likelihood of distress is higher, it may be useful to include procedures for minimizing distress, such as additional pilot testing of questionnaires to identify and perhaps modify particularly distressing questions. However, each of these studies used this process, and still caused distress for a minority of respondents. It may also be useful to include procedures for managing distress if occurs. For instance, investigators could identify a counselor who can provide crisis support and arrange a referral if desired. Investigators may also want to give consideration to the method of contact chosen for the study. However, the data are not sufficient to warrant a requirement for any of these procedures, and investigators should continue to use their best clinical judgment. This study has two limitations. First, it assessed distress in 4 studies in selected populations of patients, and therefore these findings may not be generalizable to other studies in other populations. Second, this study used a single self-rating of distress, rather than validated instruments to assess emotional states such as anxiety. Nevertheless, these findings provide valuable insights into the risk of distress in this population, as measured by a single item with adequate face validity. Research and QI activities that collect data from family members have the potential to advance our understanding of patients and families perceptions of care and provide opportunities to improve care. Therefore, it would not be appropriate to restrict these sorts of studies, or to prohibit them, because of concerns about distress that they may cause for respondents. Instead, as with any research, a study s risks must be evaluated carefully. These risks must then be balanced against any potential benefits and the importance of the knowledge to be gained. 25,26 Furthermore, investigators have an obligation to reduce these risks as much as possible and to ameliorate harms, such as distress, when they occur. 26 Studies that adhere to these ethical principles have a valuable contribution to make to improving end-of-life care. Acknowledgments Dr. Casarett is supported by an Advanced Research Career Development Award in Health Services Research from the Department of Veterans Affairs, by a Paul Beeson Physician Faculty Scholars Award, and by grants from the

6 598 Takesaka et al. Vol. 28 No. 6 December 2004 Greenwall Foundation and the Hartford Foundation. References 1. Cohen LM, McCue JD, Germain M, Kjellstrand CM. Dialysis discontinuation: a good death? Arch Intern Med 1995;155(1): Payne SA, Langley-Evans A, Hillier R. Perceptions of a good death: a comparative study of the views of hospice staff and patients. Pall Med 1996; 10(4): Groth-Juncker A, McCusker J. Where do elderly patients prefer to die? Place of death and patient characteristics of 100 elderly patients under the care of a home health care team. J Am Geriatr Soc 1983;31(8): Steinhauser K, Cipp EC, McNeilly M, et al. In search of a good death: observations of patients, families, and providers. Ann Intern Med 2000;132: Hanson LC, Henderson M, Menon M. As individual as death itself: a focus group study of terminal care in nursing homes. J Pall Med 2002;5(1): Patrick DL, Engelberg RA, Curtis JR. Evaluating the quality of dying and death. J Pain Symptom Manage 2001;22: Singer PA, Martin DK, Kelner M. Quality endof-life care. Patients perspectives. JAMA 1999;281: Addington-Hall J, McPherson C. After-death interviews with surrogates/bereaved family members: some issues of validity. J Pain Symptom Manage 2001;22: Teno JM, Claridge B, Casey V, Edgman-Levitan S, Fowler J. Validation of toolkit after-death bereaved family member interview. J Pain Symptom Manage 2001;22: Lynn J, Teno JM, Phillips RS, et al. Perceptions by family members of the dying experience of older and seriously ill patients. Ann Intern Med 1997; 126(2): Parkes CM. Bereavement: Studies of grief in adult life. London: Tavistock, Casarett D, Kutner J, Abrahm J. Life after death: A practical approach to grief and bereavement. American College of Physicians Consensus Paper. Ann Intern Med 2001;138: Cook AS, Bosley G. The experience of participating in bereavement research: stressful or therapeutic? Death Studies 19(2): Riches G, Dawson P. Making stories and taking stories: methodological reflections on researching grief and marital tension following the death of a child. Brit J Guidance Counselling 1996;24(3): Hadjistavropoulos T, Smythe WE. Elements of risk in qualitative research. Ethics Behav 2001; 11(2): Dyregrov K. Bereaved parents experience of research participation. Soc Sci Med 2004;58(2): Casarett D, Crowley R, Hirschman K. Surveys to assess satisfaction with end of life care: does timing matter? J Pain Symptom Manage 2003;25: Casarett D, Takesaka J, Karlawish J, Hirschman K, Clark C. How should clinicians discuss hospice as an option for patient care in the final stage of dementia? Caregivers attitudes toward hospice and information needs. Alzheimer Dis Assoc Disord 2002;16: Mickey J, Greenland S. A study of the impact of confounder-selection criteria on effect estimation. Am J Epid 1989;129: Hosmer DW, Lemeshow S. Applied logistic regression. New York: John Wiley and Sons, Lindemann E. Symptomatology and management of acute grief. Am J Psychiatr 1944;101: Robley LR. The ethics of qualitative nursing research. J Professional Nursing 1995;11(1): Casarett D, Karlawish J, Hirschman K. Are hospices ready to participate in palliative care research? Results of a nationwide survey. J Pall Med 2001;5: The Nuremberg Code. Reprinted in Brody, B. The ethics of biomedical research. An international perspective. New York: Oxford University Press, 1998: Declaration of Helsinki. Available at net/e/policy/17-c_e.html. Accessed January 12, Department of Health and Human Services. Protection of Human Subjects. Title 45 Part 46: Revised. Code of Federal Regulation June 18.