J Am Soc Nephrol 14: , 2003

Transcription

1 J Am Soc Nephrol 14: , 2003 Bias in Assessment of Health-Related Quality of Life in a Hemodialysis Population: A Comparison of Self-Administered and Interviewer-Administered Surveys in the HEMO Study MARK UNRUH,* GUOFEN YAN, MILENA RADEVA, RON D. HAYS, ROBERT BENZ, NICOLAOS V. ATHIENITES, JOHN KUSEK, ANDREW S. LEVEY, KLEMENS B. MEYER, and THE HEMO STUDY GROUP *University of Pittsburgh Medical Center, Pittsburgh, Pennsylvania; Cleveland Clinic Data Coordinating Center, Cleveland, Ohio; Department of Health Services, University of California Los Angeles, Los Angeles, California; Lankenau Hospital, Wynnewood, Pennsylvania; New England Medical Center, Boston, Massacusetts; Division of Kidney, Urologic, and Hematologic Disease, National Institute of Diabetes and Digestive and Kidney Diseases, Bethesda, Maryland Abstract. Examined is the relationship of patient-reported health-related quality of life (HRQOL) to the mode of survey administration in the Hemodialysis Study. In addition to selfadministered surveys to assess HRQOL, interviewer-administered surveys were made available to include patients with poor vision, decreased manual dexterity, or strong preference. For examining the predictors of participation by self-administration of the survey, multiple logistic regression was performed. For examining the relationship of HRQOL results to mode of survey administration, adjusted differences between the selfadministered and interviewer-administered groups were obtained from multiple linear regression models accounting for sociodemographic and case-mix factors. A total of 978 of the first 1000 subjects in the Hemodialysis Study completed the survey by interview (n 427) or by self-administration (n 551). The interviewer-administered group was older, was more likely black, had longer duration of ESRD, had a higher prevalence of diabetes, and had more severe comorbidity (all P 0.01). After adjustment for these differences, patients in the interviewer-administered group had higher scores on scales that measured Role-Physical, Role-Emotional, and Effects of Kidney Disease (all P 0.001). Dialysis studies that restrict HRQOL measurement to patients who are able to complete surveys without assistance will not accurately represent the health of the overall hemodialysis population. Clinical studies and clinical practices using HRQOL as an outcome should include interviewer administration or risk a selection bias against subjects with older age, minority status, and higher level of comorbidity. Future investigation should include research of survey modalities with a low response burden such as telephone interview, computer-assisted interview, and proxy administration. Patient reports of health-related quality of life (HRQOL) are recognized as providing important information about the impact of ESRD and its treatment on daily life (1 6). HRQOL questionnaires are used increasingly in studies of ESRD patients (7) and by dialysis providers (8,9). HRQOL surveys may be used in clinical care to screen for potential problems, to prioritize problems, to facilitate communication between health care workers and patients, and to monitor response to treatment (10,11). However, there is little guidance about how best to gather HRQOL information from patients with kidney failure. HRQOL surveys may be self-administered, but other alternatives, such as interviewer administration of surveys, may be Received September 10, Accepted April 22, Correspondence to Dr. Mark Unruh, University of Pittsburgh Medical Center, A909 Scaife Hall, Pittsburgh, PA Phone: ; Fax: ; unruhm@msx.dept-med.pitt.edu / Journal of the American Society of Nephrology Copyright 2003 by the American Society of Nephrology DOI: /01.ASN B1 particularly important in the ESRD population, because many of these patients are advanced in age and have comorbid conditions that preclude traditional questionnaire self-administration. It is precisely these older and sicker patients for whom HRQOL measures may be most needed (12). Although HRQOL assessment has become a commonly reported outcome in ESRD, we are unaware of previous studies examining the influence of mode of survey administration on survey results among this aged and chronically ill population. Therefore, we examined HRQOL results from both intervieweradministered and self-administered questionnaires in the National Institute of Diabetes and Digestive and Kidney Diseases Hemodialysis (HEMO) Study, a randomized clinical trial of the effects of hemodialysis dose and membrane flux. The HEMO Study offered interviewer-administered surveys as an alternative to self-administration for patients who were physically impaired or who expressed a strong preference for interviewer administration. First, we assessed the demographic and clinical predictors of participating by self-administered questionnaire. Second, we examined the extent to which scale scores were related to the mode of survey administration.

2 J Am Soc Nephrol 14: , 2003 Survey Mode and Health-Related Quality of Life among Hemodialysis Patients 2133 Materials and Methods Study Design The HEMO Study was a 15-center, randomized, clinical trial of the effects of hemodialysis dose and membrane flux on mortality and morbidity in patients undergoing chronic dialysis (13,14). The primary end point was all-cause mortality. Patient eligibility criteria have been previously described (15). All subjects who met the study criteria for enrollment at each center were eligible for randomization; there was no sampling strategy within centers. All centers had equal randomization targets of 60 concurrent patients with the exception of one center, where the target was 70. The Institutional Review Board at all of the institutions approved the study protocol, and written consent was obtained from all study participants. At randomization and annually, HEMO Study patients responded to a survey assessing health and satisfaction with care. Enrollment in the HEMO Study began in March 1995 and ended in December We report a cross-sectional analysis of responses to the Kidney Disease Quality of Life Long Form questionnaire (KDQOL-LF) at randomization from the first 1000 patients enrolled. Data Collection Demographic information and clinical history were collected through review of medical records and self-reported questionnaires. Clinical data including BP, dietary intake, laboratory measurements, and anthropometric measures were obtained using standardized protocols (14). The KDQOL-LF was one of several QOL assessments in the HEMO Study. Other HRQOL assessments in the HEMO Study include the Index of Well Being; the Karnofsky Performance Index; and supplemental items regarding sleep disturbances, symptoms during dialysis, and problems with hemodialysis access. Comorbidity was assessed at baseline and annually using the Index of Co-Existing Disease (ICED) (15,16). The ICED comorbidity index scores range from 0 to 3 with a higher score reflecting increased severity of disease. The research coordinators were directed to administer the KDQOL-LF to patients who were unable to self-administer the form as a result of physical impairment such as poor vision or limited manual dexterity or who stated a strong preference for the interview format. Interviewers were directed to read the survey verbatim and to avoid rephrasing questions. The KDQOL-LF survey is a comprehensive assessment of generic and disease-targeted QOL (17) that was slightly modified in the HEMO Study. Although sexual function questions were initially included in the questionnaire, HEMO study coordinators considered them too sensitive for interviewer administration in a hemodialysis unit. These questions were removed from all versions of the questionnaire after the first 374 patients. Furthermore, the scales for work and sleep were not included in this analysis. Although the work items from the KDQOL-LF were included in the survey, these items were not scaled because of floor and ceiling effects. Because most patients did not work at all, little information was gathered from these items. We excluded the KDQOL-LF sleep scale from this analysis examining the relationship of survey scores to mode of administration because there were differences in the response choices between the self-administrated and interviewer-administered questionnaire. Finally, we excluded Quality of Social Interaction from this analysis because it did not meet the minimum requirements for internal consistency reliability for group-level comparisons. KDQOL-LF Assesses Generic and Kidney Disease Targeted HRQOL Domains The SF-36 is the generic core of the KDQOL-LF. The SF-36 has been evaluated extensively in the general population and the ESRD population (8,18,19). The SF-36 questions are grouped into eight scales: Physical Functioning (10 items), Role-Physical (4 items), Bodily Pain (2 items), General Health (5 items), Vitality (4 items), Social Functioning (2 items), Role-Emotional (3 items), and Mental Health (5 items) (20). The range for all scales is from 0 to 100 with higher scores indicating better health. KDQOL-LF includes a Symptoms/Problems scale (34 items) that assesses the extent to which symptoms bother the subject, such as dry itchy skin, thirst and hunger, pain in the joints or back, muscle cramps or soreness, and clotting or other problems with the access site (17). The Effects of Kidney Disease scale (20 items) measures the impact of kidney failure on daily life with questions about restrictions on fluid and dietary intake, work, travel, lifting, and personal appearance. Burden of Kidney Disease (four items) considers the impact of kidney failure on the sense of accomplishment and achievement. Cognitive Function (six items) assesses difficulty with memory and concentration. Social Support (four items) measures one s satisfaction with family and social life. Dialysis Staff Encouragement (six items) measures the extent that dialysis staff encourage patients to be independent and to lead as normal a life as possible. Patient Satisfaction (two items) assesses how well care meets expectations. Data Analysis Demographic and laboratory factors are described as means for continuous variables and frequency for dichotomous variables. The associations between scales and other continuous variables were examined using Pearson correlation coefficient. Differences between the self-administered and interviewer-administered groups were assessed using two sample t statistics for interval or ratio-level variables (e.g., age) and 2 tests for categorical variables (e.g., race). The psychometric performance of each scale in the KDQOL-LF was evaluated using several methods. Descriptive statistics (mean, SD, response rate, and percentage of patients at the floor or ceiling) were calculated for each scale. Internal consistency reliability for scales was estimated using Cronbach coefficient (21). The HEMO Study previously reported the relationship of center, age, race, gender, diabetes, albumin (g/dl), creatinine (mg/dl), and ekt/v with QOL domains (P 0.01) (22). The ICED comorbidity index, duration of dialysis, whether the patient was listed for kidney transplantation, and normalized protein catabolic rate significantly differed between the two groups, and we reasoned that HRQOL should also be related to these variables. We used the entire set of sociodemographic and case-mix covariates as a full model in our regression analyses for each scale. To estimate the likelihood of survey self-administration in the hemodialysis population, we created multiple logistic regression models. All variables that were significant in univariate comparisons between self- and interviewer administration at P 0.20 were put into a backward selection multiple logistic regression model. The criterion for selection to the final model was P 0.10, and final models were examined for convergence and fit. To test the hypothesis that mode of administration is related to KDQOL-LF results, MANOVA was performed. MANOVA is a global test procedure in that it attempts to detect the presence of any differences in the population means among the groups. Two MANOVA procedures were performed by testing for an overall effect of mode of administration in both the generic SF-36 domains and the

3 2134 Journal of the American Society of Nephrology J Am Soc Nephrol 14: , 2003 kidney disease targeted domains. We controlled for our full set of covariates described previously by entering them as covariates in the MANOVA. For examining the associations of scale scores with survey mode of administration, two linear regression models were estimated for each KDQOL-LF scale. The first model contained survey mode only and yielded the unadjusted difference between self-administered and interviewer-administered methods. We examined a second model that included the full set of covariates and yielded adjusted estimates of differences by mode of administration. The signs of the estimated coefficients and the R 2 were examined for the full model of each scale. Although most scales did not exhibit heavy skewness or kurtosis, some scales based on relatively few items did exhibit some deviations from normality as a result of discreteness. However, the methods that we used to relate the mean QOL scores to mode of administration and other factors, including the MANOVA tests, are valid in large samples irrespective of normality as a result of variants of the central limit theorem for linear models. Furthermore, to test for collinearity, we examined the variance inflation factor associated with mode of administration and the degree of correlation between covariates. We examined interaction terms for two reasons. First, to demonstrate the generalizability of our findings, we examined the possibility that the effect of mode of administration on scale scores might vary with race or gender. We did this by including interaction terms in the full model. Second, we examined interactions between the mode of administration and each of the other covariates included in the full model. If a significant interaction persists between the mode of administration and a specific covariate, then we are not adequately accounting for the covariates without including the interaction term. All interaction terms were tested in the full model for each scale. Sensitivity Analyses Because we hypothesized that the interviewed group might be heterogeneous, the mode of administration was reclassified using a three-level variable: self-administration, interview administration as a result of physical impairment, and interview administration as a result of patient preference. The reclassified categorical variable was tested with multiple linear regression in a full set of covariates as described previously. The significance testing of the supplemental model was a pairwise comparison of each of the interviewed groups with the self-administration group. The regression analysis described above that was used to obtain adjusted estimates of the effect of mode of administration on the QOL scale scores treated clinical center as a fixed effect and thus could have underestimated the standard error of the effect of mode of administration if there was heterogeneity in the association of mode of administration with the QOL scales across the centers. To address this issue, we repeated the analyses relating the QOL scales to mode of administration, center, and the specified baseline covariates using a mixed-effects analysis with clinical center and the interaction of center with mode of administration treated as random effects. In addition, to determine whether the scale scores tended to cluster for tests administered by the same interviewers, we examined an additional mixed-effects model in which interviewer was treated as a random effect nested within center. This analysis was restricted to the subset of patients administered the QOL instrument by interviewer at the 13 centers where there was more than one interviewer P 0.05 was considered statistically significant. All significance tests were two-tailed. All analyses were performed with Unix SAS 6.12 (23). Results HEMO Study Patient Characteristics Of the first 1000 patients enrolled in the HEMO Study, 978 completed the KDQOL-LF questionnaire at baseline. The 22 patients who did not respond to the survey did not speak either English or Spanish. Table 1 shows relevant sociodemographic and clinical characteristics of the 978 respondents. The average age was 57.6 yr, and 53.7% were female. Almost two thirds were black, and 5% were Hispanic. A majority had diabetes or hypertension as the cause of ESRD, and approximately one third had severe comorbidity. The average albumin was 3.7 g/dl, and the average hematocrit was 33%. Overall, 44% of study patients were interviewed and the remainder self-administered the KDQOL-LF. Of all those interviewed, 42% were interviewed because of poor vision, 43% because of patient preference, and 15% because of limited manual dexterity. As shown in Table 1, patients who responded to the interviewer-administered questionnaire were on average older; were more likely to be female, black, and to have diabetes as the cause of ESRD; and to have a high burden of comorbid disease. The interviewer-administered participants had significantly lower mean serum albumin and creatinine, and a lower percentage of patients were listed for transplantation. Table 2 displays the independent predictors of survey selfadministration. The likelihood of participating by self-administering a questionnaire becomes significantly lower when the subject was older, had spent more time on dialysis, had a higher ICED score, or had diabetes. Also, both blacks and other ethnic groups were substantially less likely than whites to self-administer a survey. In the adjusted model, gender and higher creatinine were not significant. Survey Administration The HEMO survey was administered to 98% of patients in English and to 2% in Spanish. Excluding the questions regarding sexual function, on average, patients who were able to complete the survey answered 98% of items. The item completion rate ranged from 93% to 99% across the 15 centers; it varied significantly by center but did not significantly vary by patient age, gender, race, presence or absence of diabetes, or cause or duration of renal failure. Furthermore, item completion rate was not significantly different by mode of administration when tested by ANOVA (P 0.68). The three items with the lowest item completion rate of 93% included two questions about attitudes toward employment and an item asking about the number of bad days experienced in a week. Thirteen of 15 centers used more than one interviewer to administer the KDQOL-LF, and the maximum number of interviewers was six at any single center. The identity of the study coordinator was recorded for each administered survey. In general, both modes of administration yielded adequate internal consistency reliability and had high item completion rates. As shown in Tables 3 and 4, the self-administered KDQOL-LF scales had reliabilities that exceeded 0.70 for most scales. Reliability estimates for interviewer-administered

4 J Am Soc Nephrol 14: , 2003 Survey Mode and Health-Related Quality of Life among Hemodialysis Patients 2135 Table 1. Demographic and clinical characteristics of the first 1000 patients enrolled in the HEMO Study a Baseline HEMO Study Patients (n 978) Self-Administered (n 551) Interview-Administered (n 427) P Value Age (yr) 57.6 (13.7) 54.2 (14.1) 61.9 (11.8) Male Duration of dialysis (yr) 4.3 (4.5) 4.4 (4.7) 4.1 (4.3) 0.05 Race (%) black white other/unknown Primary kidney disease (%) glomerulonephritis polycystic kidney disease hypertension diabetes other ICED level Albumin (g/dl) 3.7 (0.4) 3.7 (0.4) 3.6 (0.4) 0.05 Hematocrit 32.6 (4.2) 32.7 (4.3) 32.4 (4.3) 0.99 Body mass index (kg/m 2 ) 25.4 (5.2) 25.1 (5.1) 25.7 (5.4) 0.15 enpcr b 0.97 (0.24) 1.01 (0.25) 0.97 (0.24) 0.01 ekt/v 1.3 (0.2) 1.3 (0.2) 1.3 (0.2) 0.82 Listed for transplantation Creatinine (mg/dl) 10.6 (3.0) 11.2 (3.1) 9.9 (2.7) 0.01 a Continuous variables are presented as means (SD), and dichotomous variables are presented as a percentage. b enpcr, normalized protein catabolic rate. scales exceeded 0.70 for 12 of 15 scales, as shown in Table 3. Furthermore, both Staff Encouragement and General Health were slightly lower than Also, both modes of administration demonstrated item internal consistency and item discriminant validity. The percentage of scaling successes is defined as the percentage of items that correlate 2 SEM higher with their own scale than with competing scales. The percentage of scaling successes were greater than 95% for most scales in both interviewer-administered and self-administered KDQOL-LF surveys (data not shown). Differences in HRQOL by Participant Characteristics Table 5 shows the differences in selected scales by participant characteristics that were independently associated with mode of administration. Older age, higher comorbidity, and diabetes were associated with lower physical functioning. Older age, higher comorbidity, diabetes, and shorter duration of dialysis were associated with lower Role-Physical scores. Older age, higher comorbidity, and diabetes were significantly associated with a significantly higher burden of kidney disease. Differences in Responses to Self-Administered and Interviewer-Administered Questionnaires Because baseline characteristics were associated with both mode of administration and with HRQOL scores, we examined HRQOL scores by mode of administration using a global test. MANOVA showed significant differences in mean scores between interviewer-administered and self-administered surveys. The association of mode of administration with SF-36 scores and with kidney disease targeted domain scores was significant (all P 0.001) after adjusting for demographic and case-mix variables. Because MANOVA showed significant differences in the generic and kidney disease targeted scale scores by mode of administration, we examined the association of mode of administration with individual scale scores. As Table 4 shows, there are differences in SF-36 scores and kidney disease targeted domain scores between the self-administered group and the interviewer-administered group. In the unadjusted analysis (left column), the average scores for Physical Functioning, General Health, Vitality, Burden of Kidney Disease, Cognitive function, Social Support, and Patient Satisfaction were substantially higher (better health) in the self-administered group. The average scores for Role-Physical, Role-Emotional, and Effects of Kidney Disease were substantially lower (worse health) in the self-administered group. After adjusting for demographic and case-mix variables, the differences in scores were substantially diminished in Physical Functioning, General Health, and Vitality. However, many differences remained and some were slightly larger in magnitude after adjustment. The scores that measured Role-Physical, Role-Emotional, Bodily Pain, and Effects of Kidney Disease were significantly lower (worse functioning and well-being) in the self-administered group (all P 0.05). Conversely, the scores

5 2136 Journal of the American Society of Nephrology J Am Soc Nephrol 14: , 2003 Table 2. Independent predictors of survey self-administration a Predictor Adjusted OR for Self-Administration (95% CI) P Respondent age (reference) (ref 18 45) 0.64 ( ) ( ) ( ) Respondent race white 1.00 (reference) black (ref white) 0.60 ( ) other 0.25 ( ) Female (versus male) 0.91 ( ) 0.56 Respondent comorbidity ICED (reference) ICED ( ) ICED ( ) Diabetes (versus nondiabetes) 0.47 ( ) Dialysis exposure (per each 0.95 ( ) year) Creatinine per mg/dl 1.06 ( ) 0.09 a OR, odds ratio; CI, confidence interval. OR and 95% CI for participation by self-administering the KDQOL-LF. The variable for female gender was forced into the model; otherwise, these variables met the P 0.10 selection criteria. Serum albumin, ekt/v, enpcr, and whether listed for transplantation did not meet selection criteria to remain in the model. Table 3. Descriptive statistics and KDQOL-LF score distributions: self-administered sample a Scale n Mean SD % at Floor % at Ceiling SF 36 physical functioning role-physical bodily pain general health vitality social functioning role-emotional mental health Kidney disease Targeted domains symptom/problem effects burden cognitive function social support staff encouragement patient satisfaction a Range for all scales is 0 to 100; on all scales, a higher score indicates better health. for Patient Satisfaction, Social Support, Cognition, and Burden of Kidney Disease were significantly higher (better health) in the self-administered group (all P 0.01). The amount of variance explained by the model (R 2 ) for the SF-36 domains ranged from 0.05 for mental health to 0.27 for Physical Functioning (all P 0.01), and the R 2 for the kidney disease targeted domains ranged from 0.04 for Dialysis Staff Encouragement and 0.11 for Burden of Kidney Disease (all P 0.01 except Dialysis Staff Encouragement P 0.059). The scale scores showed no consistent evidence of statistically signifi-

6 J Am Soc Nephrol 14: , 2003 Survey Mode and Health-Related Quality of Life among Hemodialysis Patients 2137 Table 4. Descriptive statistics and KDQOL-LF score distributions: interviewer-administered sample a Scale n Mean SD % at Floor % at Ceiling SF 36 physical functioning role-physical bodily pain general health vitality social functioning role-emotional mental health Kidney disease Targeted domains symptom/problem effects burden cognitive function social support staff encouragement patient satisfaction a Range for all scales is 0 to 100; on all scales, a higher score indicates better health. Table 5. Adjusted differences between mean scores of self-administered and interviewer-administered surveys Predictor Physical Functioning Role-Physical Burden of Kidney Disease Respondent age b 54.6 a (ref 18 45) Respondent race white b black other Respondent comorbidity ICED b 50.6 a 55.2 b ICED ICED Diabetic status nondiabetic 52.9 b diabetic Dialysis exposure 2 yr yr yr a P b P cant interactions between mode of administration and race, gender, or any of the other covariates. The variance inflation factors for the mode of administration factor in the multiple regression was Variance inflation factors in this range indicate presence of potential confounding and thus indicate that our adjustment for covariates was necessary but are not so large as to invalidate the adjusted estimated effects of mode of administration. In addition, none of the correlations among the

7 2138 Journal of the American Society of Nephrology J Am Soc Nephrol 14: , 2003 covariates was so high (pairwise correlations ranged from 0.36 to 0.37) as to indicate that one or more of the covariates could be dropped as a result of redundancy with other covariates. Whether the reason for interview administration was physical impairment or strong patient preference, the adjusted differences between scores from interview administration and the scores from self-administration were consistent. After adjustment for covariates, scores for Patient Satisfaction, Social Support, Cognition, and Burden of Kidney Disease were significantly higher (better health) in the self-administered survey group than in either of the interviewed groups (P 0.001). The scores measuring Role-Physical, Role-Emotional, Bodily Pain, and Effects of Kidney Disease were significantly higher (better health) in both of the interviewed groups, whether the reason for interview was the physical impairment or patient preference groups (P 0.001). The relationship of mode of administration with HRQOL scores was not found to be associated with the individual interviewers within centers that had more than one interviewer. The variability associated with the individual interviewer nested in center was not significant for any of the scale scores. Because there was still the possibility that the difference between the interviewer- and self-administered tests was a centerrelated phenomenon, we treated the mean difference between the interviewer- and the self-administered tests at each center as a random effect and then considered the inference to the population of all centers. However, there was no evidence of a center variation in the interviewer- versus self-administered comparisons. Discussion To our knowledge, this is the first report to document in detail the mode of administration of HRQOL surveys to ESRD patients and to examine the relationships between mode of administration and responses. These findings have important implications both for the care of dialysis patients and for clinical research because thousands of dialysis patients participate in QOL surveys and increasing numbers of clinical studies include QOL as an outcome. Our findings extend previous studies: we confirm the internal consistency reliability and validity of the KDQOL-LF (17,24) for both modes of administration. In this report, we present strong evidence that exclusive reliance on self-administered HRQOL surveys would result in selection bias, thus distorting HRQOL scores. The common element of selection biases is that the relation between exposure and outcome is different for those who participate and those who theoretically should be eligible for the study but do not participate (25). In studies using only selfadministered surveys, the result is that associations observed in the study represent a mix of forces determining participation, as well as forces determining HRQOL outcomes. The results of the HEMO Study suggest that exclusive reliance on selfadministered questionnaires would result in underrepresentation of the aged, minority groups, and the very sick; these are patients who may benefit most from interventions based on measurements of HRQOL. In addition, the association of mode of administration with HRQOL results suggests either that there are unmeasured differences between patients surveyed by self-administration and those who are interviewed or that the mode of administration itself influences the patients responses. This report shows that after controlling for sociodemographic and clinical factors, the interviewer-administered group reported higher-than-expected HRQOL in certain domains, supporting what previous authors have observed as an interviewer bias with QOL questionnaires. Participants who are interviewed may alter their responses to present themselves in a favorable light. These interviewer-related differences between scores were similar whether patients were interviewed because of strong preference or because of physical disability. Our findings from the HEMO Study demonstrate a selection bias in administering surveys to hemodialysis patients by pencil and paper. Interviews elicited HRQOL information from many patients with visual problems, diminished dexterity, and patient preference, and omission of these patients may limit generalizability of HRQOL results in studies and limit the utility of QOL measures in clinical care. In Table 5, we demonstrate that factors independently associated with mode of administration were also related to HRQOL scores. Indeed, the change in the estimate for mode of administration after adjustment for baseline factors as shown in Table 6 demonstrates this selection bias, which tends to be a larger effect in domains such as physical functioning and vitality that relate to physical well-being. Our findings of a selection bias by selfadministration are consistent with previous observations in hemodialysis patients that suggested that patients who are on dialysis frequently have problems with vision and also with manual dexterity because arm ranges are limited during dialysis (11). Previous studies of ESRD patients that reported only self-administered HRQOL (24,26 30) may have excluded many older patients and those with higher comorbidity and consequently worse QOL. Some studies of ESRD patients that offered only self-administered surveys of HRQOL in ESRD documented attrition rates between 20% and 50% (27,31,32). For measuring and monitor the health and well-being of hemodialysis patients and those with other chronic illness, our results suggest that measurement of HRQOL may require interviewer-administered surveys or other methods with a low respondent burden. However, using interviews to gather patient data also has its drawbacks. Our findings extend previous studies in the general population that have suggested that self-administered surveys yield higher reports of morbidity, disability, and socially undesirable behavior than reported by interviews in person or by telephone (33). In the HEMO Study, scores on the Effects of Kidney Disease, Role-Physical, and Role-Emotional scales were lower (worse functioning and well-being) when the instrument was self-administered. We confirm the observation of an interview bias previously made by investigators who reported a mode of administration effect with the SF-36 by randomly assigning subjects to self-administered and computer-assisted telephone survey (34). In that study of the general population, the Role-Physical scores were 7.5 points lower and Role-Emotional scores were 9.1 points lower among those with

8 J Am Soc Nephrol 14: , 2003 Survey Mode and Health-Related Quality of Life among Hemodialysis Patients 2139 Table 6. Adjusted differences between mean scores of self-administered and interviewer-administered surveys a Instrument Self-Administered Interview Unadjusted Difference Adjusted Difference b SF-36 physical function e 0.5 role-physical d 13.5 e bodily pain c general health c 0.7 vitality d 0.8 social function role-emotional e 15.1 e mental health Kidney disease Targeted domains symptoms/problem effects e 6.0 e burden e 10.0 e cognitive e 8.1 e social support e 10.2 e staff encouragement patient satisfaction e 8.3 e a For both unadjusted and adjusted differences, a positive value reflects a higher self-administered score compared with intervieweradministered score. b Multiple linear regression model adjusted for age, race, gender, ICED comorbidity index, years on dialysis, serum albumin, normalized protein catabolic rate, serum creatinine, equilibrated Kt/V, diabetes, listed for transplantation, and study center. c P d P 0.01 e P self-administration after adjustment. The differences between HRQOL scores in the Role-Emotional and Role-Physical domains in the general population were slightly smaller than what we observed. The effect of mode of administration in the HEMO Study may be augmented by the lack of privacy in hemodialysis units, further encouraging patients who are interviewed by study coordinators to present themselves in a positive manner. Other studies have found that subjects who are interviewed report less morbidity than those who complete questionnaires independently and in private (35). Those investigators also found that the magnitude of bias introduced by mode of administration depends on the phenomenon being measured and demonstrated that mode of administration is particularly important when assessing pain. This interviewer effect may represent a necessary trade-off when measuring QOL in a chronically ill and aged population. Although similar in average age, the first 1000 patients randomized for the HEMO Study differ in some ways from the prevalent U.S. hemodialysis population (36). The HEMO Study includes a higher proportion of blacks (64.7 versus 32.1%), probably because of the geographic distribution of the 15 clinical centers. Furthermore, the HEMO Study includes a higher proportion of women (53.6 versus 45.7%), probably a reflection of lower weight and the ability of these individuals to meet the entry requirement of a delivered ekt/v of 1.3 in 4.5 h. Both race and gender were included in the full regression model that provides an unbiased estimate of the relationship of mode of administration to QOL. Furthermore, we found no evidence of significant interactions between gender and mode of administration or race and mode of administration when tested in the fully adjusted model for each HRQOL scale, so these differences from the U.S. hemodialysis population probably do not affect the generalizability of our findings. The limitations of this study should be considered when interpreting our findings. First, we were unable to examine interviewer reliability, because a single interviewer evaluated each patient. However, a standardized interview protocol for the QOL questionnaire was implemented to limit variability. In addition, we did not find that the individual interviewer contributed significantly to scale score variability in those centers with multiple interviewers. Second, as in any observational study, there may have been unobserved differences between the self-administered and interviewer-administered groups that cannot be adjusted in a model of observed differences. However, our estimates of the administration effect in hemodialysis patients were in similar domains as previous studies in the general population (34). Our findings have important implications for dialysis patient care and future clinical research. We first caution authors and readers about comparing HRQOL results across studies using different modes of administration. Indeed, authors should explicitly include mode of administration in their study methods.

9 2140 Journal of the American Society of Nephrology J Am Soc Nephrol 14: , 2003 Dialysis providers, investigators, and policy makers should ensure that patients with advanced age and higher levels of comorbidity are able to respond to questionnaires by providing interviewer-administered surveys. Future studies of dialysis patients using HRQOL as an outcome should include interview administration or risk a selection bias against participants with higher age, minority status, higher levels of comorbidity, and generally lower HRQOL scores. However, one must recognize that the cost of interviewer administration limits the use of HRQOL surveys in dialysis patient care and in clinical studies. Improving the HRQOL of ESRD patients will require developing better tools for measurement. Acknowledgments The HEMO Study is supported by the National Institute of Diabetes and Digestive and Kidney Diseases via cooperative agreements U01DK 46109, U01DK 46114, U01DK 46126, U01DK 46143, U01DK 49240, U01DK 49241, UO1DK 49242, U01DK 49243, U01DK 49244, U01DK 49249, U01DK 49252, U01DK 49254, U01DK 49259, U01DK 49261, U01DK 49264, and U01DK Dr. Unruh was supported by NIH/T32-DK07777 Training Grant in Epidemiology, Clinical Trials and Outcomes Research. Dr. Hays was supported in part by UCLA/DREW Project EXPORT, National Institutes of Health, National Center on Minority Health & Disparities (P20-MD ), and the UCLA Center for Health Improvement in Minority Elders/Resource Centers for Minority Aging Research, National Institutes of Health, National Institute of Aging (AG ). Some study dialyzers were provided by Baxter Healthcare Corporation (McGaw Park, IL) and Fresenius Medical Care-North America (Lexington, MA). Nutritional supplements were provided by Ross Laboratories (Columbus, OH) and vitamins were provided by R&D Labs (Marina del Rey, CA). We acknowledge Tom Greene, Ph.D., Joseph Cappelleri, Ph.D., and Hong Chang, Ph.D., for invaluable suggestions for data analysis and manuscript preparation. References 1. Rettig RA, Sadler JH: Measuring and improving the health status of end stage renal disease patients. Health Care Financ Rev 18: 77 82, Rettig RA, Sadler JH, Meyer KB, Wasson JH, Parkerson GR Jr, Kantz B, Hays RD, Patrick DL: Assessing health and quality of life outcomes in dialysis: A report on an Institute of Medicine workshop. Semin Nephrol 17: , Schrier RW, Burrows-Hudson S, Diamond L, Lundin AP, Michael M, Patrick DL, Peters TG, Powe NR, Roberts JS, Sadler JH, Siu AL, Lohr KN, Rettig RA: Measuring, managing, and improving quality in the end-stage renal-disease treatment setting: Committee statement. Am J Kidney Dis 24: , Kutner NG: Assessing end-stage renal-disease patients functioning and well-being: Measurement approaches and implications for clinical practice. Am J Kidney Dis 24: , Kimmel PL: Just whose quality of life is it anyway? Controversies and consistencies in measurements of quality of life. Kidney Int 57: S113 S120, Valderrabano F, Jofre R, Lopez-Gomez JM: Quality of life in end-stage renal disease patients. Am J Kidney Dis 38: , Garratt A, Schmidt L, Mackintosh A, Fitzpatrick R: Quality of life measurement: Bibliographic study of patient assessed health outcome measures. Br Med J 324: 1417, Diaz-Buxo JA, Lowrie EG, Lew NL, Zhang HY, Lazarus JM: Quality-of-life evaluation using Short Form 36: Comparison in hemodialysis and peritoneal dialysis patients. Am J Kidney Dis 35: , Chapman MM, Meyer KB: Assessing health status in a dialysis clinic. Am J Health Syst Pharm 52: 15, Higginson IJ, Carr AJ: Measuring quality of life: Using quality of life measures in the clinical setting. Br Med J 322: , Kurtin PS, Davies AR, Meyer KB, DeGiacomo JM, Kantz ME: Patient-based health status measures in outpatient dialysis. Early experiences in developing an outcomes assessment program. Med Care 30[5 Suppl]: MS136 MS149, Addington-Hall J, Kalra L: Who should measure quality of life? Br Med J 322: , Eknoyan G, Levey AS, Beck GJ, Agodoa LY, Daugirdas JT, Kusek JW, Levin NW, Schulman G: The hemodialysis (HEMO) study: Rationale for selection of interventions. Semin Dial 9: 24 33, Eknoyan G, Beck G, Breyer J, Kopple J, Kusek J, Levey A: Design and preliminary results of the Mortality and Morbidity of Hemodialysis Pilot Study. J Am Soc Nephrol 5: 513, Miskulin DC, Athienites NV, Yan G, Martin AA, Ornt DB, Kusek JW, Meyer KB, Levey AS: Comorbidity assessment using the Index of Coexistent Diseases in a multicenter clinical trial. Kidney Int 60: , Athienites NV, Miskulin DC, Fernandez GF, Bunnapradist S, Simon G, Landa M, Schmid C, Greenfield S, Levey AS, Meyer KB: Comorbidity assessment in hemodialysis and peritoneal dialysis using the Index of Coexistent Disease (ICED). Semin Dial 13: , Hays RD, Kallich JD, Mapes DL, Coons SJ, Carter WB: Development of the kidney disease quality of life (KDQOL) instrument. Qual Life Res 3: , Meyer KB, Espindle DM, DeGiacomo JM, Jenuleson CS, Kurtin PS, Davies AR: Monitoring dialysis patients health status. Am J Kidney Dis 24: , DeOreo PB: Hemodialysis patient-assessed functional health status predicts continued survival, hospitalization, and dialysisattendance compliance. Am J Kidney Dis 30: , Ware J, Snow K, Kosinski M, Gandek B: SF-36 Health Survey Manual and Interpretation Guide, Boston, The Health Institute, New England Medical Center, Cronbach L: Coefficient alpha and the internal structure of tests. Psychometrika 16: , Meyer K, Paranandi L, Hays R, Benz B, Athienetes N, Kusek J, Levey A: Clinical correlates of baseline quality of life in the HEMO Study: An interim report. J Am Soc Nephrol 8: 204A, SAS Institute: SAS/STAT User s Guide, Cary, NC, SAS Institute Inc., Korevaar J, Merkus M, Jansen M, Dekker F, Boechoten E, Kediet R: Validation of the KDQOL-SFTM: A dialysis-targeted health measure. Qual Life Res 11: , Rothman K, Greenland S: Modern Epidemiology, Philadelphia, PA, Lippincott Williams & Wilkins, Wu AW, Fink NE, Cagney KA, Bass EB, Rubin HR, Meyer KB, Sadler JH, Powe NR: Developing a Health-Related Quality-of-

10 J Am Soc Nephrol 14: , 2003 Survey Mode and Health-Related Quality of Life among Hemodialysis Patients 2141 Life Measure for End-Stage Renal Disease: The CHOICE Health Experience Questionnaire. Am J Kidney Dis 37: 11 21, Merkus MP, Jager KJ, Dekker FW, Boeschoten EW, Stevens P, Krediet RT: Quality of life in patients on chronic dialysis: Self-assessment 3 months after the start of treatment. The Necosad Study Group. Am J Kidney Dis 29: , Merkus MP, Jager KJ, Dekker FW, De Haan RJ, Boeschoten EW, Krediet RT: Quality of life over time in dialysis: The Netherlands Cooperative Study on the Adequacy of Dialysis. NECOSAD Study Group. Kidney Int 56: , Moreno F, Sanz-Guajardo D, Lopez-Gomez JM, Jofre R, Valderrabano F: Increasing the hematocrit has a beneficial effect on quality of life and is safe in selected hemodialysis patients. Spanish Cooperative Renal Patients Quality of Life Study Group of the Spanish Society of Nephrology. J Am Soc Nephrol 11: , Kalantar-Zadeh K, Kopple JD, Block G, Humphreys MH: Association among SF36 quality of life measures and nutrition, hospitalization, and mortality in hemodialysis. J Am Soc Nephrol 12: , Beusterien KM, Nissenson AR, Port FK, Kelly M, Steinwald B, Ware JE Jr: The effects of recombinant human erythropoietin on functional health and well-being in chronic dialysis patients. J Am Soc Nephrol 7: , Bremer B, McCauley C, Wrona R, Johnson J: Quality of life in end-stage renal disease: A reexamination. Am J Kidney Dis 13: , Bergner M, Bobbitt RA, Carter WB, Gilson BS: The Sickness Impact Profile: Development and final revision of a health status measure. Med Care 19: , McHorney CA, Kosinski M, Ware JE Jr: Comparisons of the costs and quality of norms for the SF-36 health survey collected by mail versus telephone interview: Results from a national survey. Med Care 32: , Grootendorst PV, Feeny DH, Furlong W: Does it matter whom and how you ask? Inter- and intra-rater agreement in the Ontario Health Survey. J Clin Epidemiol 50: , Patient characteristics at the start of ESRD: Data from the HCFA Medical Evidence Form. Am J Kidney Dis 34: S63 S73, 1999