Multiple measurements of depression predict mortality in a longitudinal study of chronic hemodialysis outpatients

Kidney International, Vol. 57 (2000), pp. 2093–2098

Multiple measurements of depression predict mortality in a longitudinal study of chronic hemodialysis outpatients PAUL L. KIMMEL, ROLF A. PETERSON, KAREN L. WEIHS, SAMUEL J. SIMMENS, SYLVAN ALLEYNE, ILLUMINADO CRUZ, and JUDITH H. VEIS Departments of Medicine and Psychiatry and Behavioral Sciences, School of Medicine, and Department of Epidemiology and Biostatistics, School of Public Health, George Washington University Medical Center; Department of Psychology, George Washington University; Departments of Human Development and Psychoeducational Studies and Medicine, Howard University Medical Center; and Department of Medicine, Washington Hospital Center, Washington, D.C., USA

Multiple measurements of depression predict mortality in a longitudinal study of chronic hemodialysis outpatients. Background. The medical risk factors associated with increased mortality in hemodialysis (HD) patients are well known, but the psychosocial factors that may affect outcome have not been clearly defined. One key psychosocial factor, depression, has been considered a predictor of mortality, but previous studies have provided equivocal results regarding the association. We sought to determine whether depressive affect is associated with mortality in a longitudinal study of end-stage renal disease (ESRD) patients treated with HD, using multiple assessments over time. Methods. Two hundred ninety-five outpatients with ESRD treated with HD were recruited from three outpatient dialysis units in Washington D.C. to participate in a prospective cohort study with longitudinal follow-up. Patients were assessed every six months for up to two years using the Beck Depression Inventory (BDI), age, serum albumin concentration, Kt/V, and protein catabolic rate (PCR). A severity index, previously demonstrated to be a mortality marker, was used to grade medical comorbidity. The type of dialyzer with which the patient was treated was noted. Patient mortality status was tracked for a minimum of 20 and a maximum of 60 months after the first interview. Cox proportional hazards models, treating depression scores as time-varying covariates in a univariable analysis, and controlling for age, medical comorbidity, albumin concentration, and dialyzer type and site in multivariable models, were used to assess the relative mortality risk. Results. The mean (⫾ SD) age of our population at initial interview was 54.6 ⫾ 14.1 years. The mean PCR was 1.06 ⫾ 0.27 g/kg/day, and the mean Kt/V was 1.2 ⫾ 0.4 at baseline, suggesting that the patients were well nourished and dialyzed comparably to contemporary U.S. patients. The patients’ mean BDI at enrollment was 11.4 ⫾ 8.1, in the range of mild depression. Patients’ baseline level of depression was not a significant predictor of mortality at 38.6 months of follow-up. In contrast,

Key words: chronic renal disease, end-stage renal disease, hemodialysis, depression, Beck Depression Inventory, mortality and dialysis. Received for publication September 9, 1999 and in revised form November 23, 1999 Accepted for publication December 18, 1999

 2000 by the International Society of Nephrology

when depression was treated as a time-varying covariate based on periodic follow-up assessments, the level of depressive affect was significantly associated with mortality in both single variable and multivariable analyses. Conclusions. Higher levels of depressive affect in ESRD patients treated with HD are associated with increased mortality. The effects of depression on patient survival are of the same order of magnitude as medical risk factors. Our findings using both controls for factors possibly confounded with depressive affect in patients with ESRD and time-varying covariate analyses may explain the inconsistent results of previous studies of depression and mortality in ESRD patients. Timevarying analyses in longitudinal studies may add power to defining and sensitivity to establishing the association of psychosocial factors and survival in ESRD patients. The mechanism underlying the relationship of depression and survival and the effect of interventions to improve depression in HD outpatients and general medical inpatients should be studied.

The prevalence of depression and its role in mediating survival of patients with end-stage renal disease (ESRD) has been highly controversial [1]. Depression has been shown to be associated with excess mortality in patients hospitalized with myocardial infarction [2, 3] and with other medical illnesses [4, 5]. Many of the symptoms used to diagnose depression mimic those of medical illnesses, including uremia [1]. Unipolar depression has been projected to be a leading future cause of morbidity and mortality and the second most common debilitating disease worldwide over the next 20 years, surpassed only by the morbidity of cardiovascular disease in the number of years of productive life diminished by disability [6, 7]. While several studies of ESRD patients have shown an association of depression with mortality [1, 8–13], others have not [14–16]. Older studies of the association of depression and mortality in ESRD patients were, however, often poorly controlled for the medical factors that might have affected mortality or been confounded with depression,

2093

2094

Kimmel et al: Depression and survival in HD patients

such as the symptoms of comorbid renal and systemic diseases. Previous studies also varied widely in their observation periods. In addition, studies have not assessed ESRD patients’ depressive affect in a longitudinal fashion. In a previous study, including a small number of patients treated with both hemodialysis (HD) and chronic peritoneal dialysis, we showed that depressive indices predicted survival of patients at one year, but by two years, medical markers were more powerful predictors of patient survival [13]. We recently studied outcomes prospectively in a larger number of patients treated with HD for ESRD, and reported that psychosocial factors measured at baseline, such as perception of the effects of illness and level of social support, were associated with differential survival. However, we were unable to delineate an expected association of extent of depressive affect and mortality in this population [16]. Recently, studies have demonstrated the power of longitudinal observations in explaining survival in ESRD patients treated with HD, using time-varying covariate survival analyses [17]. We therefore assessed the association of higher levels of depressive affect, based on repeated assessments, with mortality in our study population of urban patients with ESRD treated with HD, while controlling for the variation in various potential confounders. METHODS Our patient recruitment techniques and study population have been extensively described [1, 16, 18, 19]. Recruitment began September 1, 1992, and concluded March 31, 1996, and the observation period ended December 31, 1997. George Washington University Medical Center’s Ambulatory Dialysis Unit (GWUMC), Howard University Medical Center’s Dialysis Unit (HUMC), and the Washington Veterans Affairs Medical Center Dialysis Unit (VAMC), all in Washington D.C. (USA), were the study sites. The population of the three units was primarily composed of urban, African American patients. All patients enrolled in chronic ESRD HD programs at the three units and currently receiving dialysis on an outpatient basis were eligible for the study, with the exception of HIV-infected patients, patients who had a psychiatric diagnosis of psychosis, and patients who scored less than 23 on a mini-mental status exam [20]. Up to 10% of the patient population was known to have HIV infection [21]. Informed consent was obtained from all patients prior to enrollment. The study was approved by the institutional review boards of the three medical centers. Disease severity (age and the presence of comorbid conditions, including diabetes mellitus) was quantitated by the ESRD severity coefficient [13, 16, 18, 19, 22], previously validated in a large sample of ESRD patients [22]. The mean of three sequential monthly serum albumin concentrations after the patients’ enrollment was

determined [16]. The dialyzer used to treat each patient at study entry was noted and categorized as (1) unmodified cellulose or (2) modified cellulose or synthetic, as previously described [16, 23]. The Karnofsky Performance Status Scale score was determined as previously described [18]. The level of depressive symptoms was assessed using the Beck Depression Inventory (BDI) [24] in patients at study entry. Thereafter, patients were evaluated every six months for up to two years using the BDI. Questionnaires were administered in an interview format by trained personnel. The BDI is a well-validated measure of depression, correlating with diagnostic criteria for depression [24, 25]. It has been used frequently to assess depression in patients with ESRD [1, 12, 13, 16, 18, 19, 26]. A second measure, the Cognitive Depression Index (CDI), a subscale of the BDI, is composed of 15 of its 21 items [1, 13, 18, 19, 27] and focuses on thoughts and feelings related to the diagnosis of depression, such as guilt, disappointment, and failure, excluding its somatic items [1, 13, 16, 27]. This scale was used to reduce the possible confound between symptoms of medical illness and the somatic components of depression measured in the BDI [1, 24, 27]. The CDI has previously been highly correlated with the BDI in ESRD patients [1, 13, 18, 19], but was unassociated with measures of severity of illness, nutrition, renal function, and delivery of dialysis [27]. The number of assessment periods per patient varied according to when each patient entered the study and whether each patient died or left the HD center during their first three years of follow-up. The longest followup for any surviving patient was five years from first interview to the end of the study, and the shortest was 20 months. The survival time for each patient was determined by the number of days between the initial study evaluation and the end of the study observation period or date of death [16]. Survival status was confirmed using the Health Care Finance Administration database, obtained through ESRD Network 5 (Richmond, VA, USA) for each patient enrolled in the study [16]. Cox proportional hazards regression was performed to predict the mortality hazard associated with baseline level of depressive affect, controlling for the effects of variation in patient age, severity coefficient, level of serum albumin concentration, dialyzer type, and site, as previously described [16, 28]. Survival analyses were then conducted using Cox’s proportional hazards regression models, which allow the incorporation of time varying covariates [29, 30]. Survival statistics were calculated with Proc PHREG from SAS 6.12 (SAS Institute Inc., Cary, NC, USA) [16, 30] using the “exact” method for tied follow-up times [30]. BDI and CDI scores were treated as time-varying covariates, incorporating up to six patient assessments [29]. Each

2095

Kimmel et al: Depression and survival in HD patients

patient contributed a separate observation spanning the time from each BDI assessment until the next assessment or death or the end of the study. The time-varying form of the Cox regression model produces relative risk estimates of mortality similar in meaning to that of the standard Cox regression model [30]. Time-varying models have at least two advantages over standard survival models. First, time-varying models essentially predict from each assessment point, rather than assuming that prediction from baseline is sufficient. Second, because more data points are considered, statistical power may be greater. The relative risk reported is related to the risk set and level of the covariate for the exact time that each estimate is made, not just to baseline levels. All other covariates (age, severity coefficient and serum albumin concentration, dialyzer type, and site) were fixed, based on their assessed values at the start of study enrollment. For each of the two depression measures, two Cox regression models were fit. The first model used only the depression measure as a predictor. The second added covariates, which had previously been shown to predict mortality in this [1, 16, 23] and other populations [17, 23, 31]. In order to enhance interpretability, all continuous variables used as predictors—BDI and CDI scores, age, severity coefficient, and albumin—were standardized to a standard deviation of 1.0, allowing a comparison of the effects of changes in levels of risk factors across different parameters in a population, as previously described [16]. Cumulative survival estimates were plotted against time for three levels of depression: BDI levels less than 10 (low or nonexistent symptoms), 10 to 15 (mild levels), and 16 or greater (moderate to severe levels) [25]. One-way analysis of variance was used to assess differences between means at different time periods after study enrollment. The ␣ level of tests of survival was 0.05. Data are presented as mean ⫾ SD. RESULTS The total sample surveyed comprised 295 subjects, a 60.8% enrollment rate [16]. Less than 10% of HIVuninfected, recruited patients were excluded because of failure to pass the mental status exam or presence of psychosis. The median follow-up time from the first interview to patients’ death or the end of the observation period was 38.6 months. More than 92% of our patient sample was comprised of African Americans. Twentyeight percent of the patients were female, and 42.3% had diabetes mellitus (Table 1). The mean age of our patient population was 54.6 ⫾ 14.1 years. Patients had been treated with HD for mean and median times of 30.2 ⫾ 47.5 and 12.4 months, respectively, at the study entry. The patients’ mean protein catabolic rate (PCR) was 1.06 ⫾ 0.27 g/kg/day. The mean Kt/V was 1.2 ⫾ 0.4. The mean serum albumin concentration was 3.82 ⫾ 0.49

Table 1. Clinical characteristics of the patient population (N ⫽ 295)

Age years Diabetes % Karnofsky score Albumin g/dL Kt/V PCR g/kg/day BDI

Mean

SD

Range

54.6 42.3 73.7 3.82 1.2 1.06 11.4

14.1

19–84

17.2 0.49 0.4 0.27 8.1

10–100 2.3–5.65 0.36–1.96 0.49–2.37 0–47

Abbreviations are: PCR, protein catabolic rate; BDI, Beck Depression Inventory.

g/dL, and the mean arm circumference, performed in 85% of enrolled patients, was 30.2 ⫾ 5.9 cm, suggesting that the patients were representative of the contemporary African American population of the U.S. ESRD program, and were well dialyzed and in an adequate nutritional state compared with contemporary populations [31–33]. The mean baseline BDI score for the entire population was 11.4 ⫾ 8.1, in the range of mild depression [25]. Depression scores at baseline assessments were not significantly correlated with the patients’ age, severity coefficient, or serum albumin levels. One hundred fourteen of the 295 patients died during the study follow-up period. The crude mortality rate was 0.13 deaths per person year, a rate similar to that found in other studies of the ESRD population [31]. All patients received a psychosocial assessment between one and six times (mean ⫽ 2.9 ⫾ 1.7 assessments). The longest period between first and last assessments for a patient was 34 months. There was no significant difference between the mean baseline BDI scores of patients who died and those who survived during the observation period. The mean BDI score of patients who survived to have three assessments (N ⫽ 122) decreased from 11.4 ⫾ 9.0 at baseline to 10.0 ⫾ 9.2 at the second assessment and 9.2 ⫾ 8.3 at the third assessment, approximately one year after the initial study entry (P ⫽ 0.004). As in our previous report, the baseline level of depressive affect was not significantly associated with mortality at a longer mean follow-up of 38.6 months (BDI, relative risk ⫽ 1.09, 95% CI, 0.92 to 1.29; CDI, relative risk ⫽ 1.09, 95% CI, 0.91 to 1.31; Table 2) [16]. For the next step in the survival analysis, survival was predicted using each of the two depression measures in separate models as a time-varying covariate. The BDI score predicted survival with an estimated relative risk of 1.24 (95% CI, 1.05 to 1.46, P ⫽ 0.01), indicating that an 8.1 point (1 SD) increment in BDI scores was associated with a 24% higher risk of death. A similar, but smaller effect was noted when CDI scores were assessed (relative risk ⫽ 1.18, 95% CI, 1.00 to 1.38, P ⫽ 0.05). The analyses were repeated with the control variables added. Both BDI and CDI scores remained significant

2096

Kimmel et al: Depression and survival in HD patients

Table 2. Depressive affect as a predictor of mortality in ESRD patients treated with hemodialysis RR (CI) Baseline Beck Depression Baseline Cognitive Depression Time-varying Beck Depression Time-varying Cognitive Depression Adjusted Time-varying Beck Depression Adjusted Time-varying Cognitive Depression

1.09 1.09 1.24 1.18 1.32

(0.92, (0.91, (1.05, (1.00, (1.13,

1.29) 1.31) 1.46) 1.38) 1.55)

1.23 (1.05, 1.43)

P value NS NS 0.01 0.05 0.0006 0.009

Data are adjusted and unadjusted for age, severity, serum albumin concentration, dialyzer type, and site. All relative risks have been standardized to a standard deviation of 1.0 (based on the SD’s of the 295 patients). Analyses for Timevarying Beck and Cognitive Depression scores are univariate. All adjusted analyses are controlled for age, severity, serum albumin concentration, dialyzer type and site. CI is confidence interval.

predictors, with an estimated 32 and 23% higher risk of death being associated with a one standard deviation increment in BDI and CDI scores, respectively (95% CI, 1.13 to 1.55, P ⫽ 0.0006, and 1.05 to 1.43, P ⫽ 0.009, respectively). Cumulative survival estimates were plotted against time for the three clinically delineated levels of depression: BDI levels less than 10 (low or nonexistent symptoms), 10 to 15 (mild levels), and 16 or greater (moderate to severe levels) [25] using BDI and CDI measurements pooled for each patient from each observation period (Fig. 1). Patients with mild symptoms or greater levels of depressive affect were at greater mortality risk than patients with low or nonexistent symptoms, although there was no difference in risk between patients with mild or more severe levels of depressive affect. DISCUSSION Previous studies of the association of depressive affect and survival have yielded contradictory results. Early studies may have assessed highly selected populations, dialyzed less intensively than current guidelines mandate, with dialysis membranes that did not provide optimal clearance and biocompatibility profiles [8–11]. In our previous analysis of this population, using a shorter, 26-month average follow-up period, we were unable to show an association of depressive affect at baseline with survival, when age, severity of comorbid illness, serum albumin concentration (as a nutritional marker), dialyzer type, and site were used as control variables [16]. Likewise, two other recent studies of ESRD patients using relatively long-term follow-ups provided similar results [14, 15]. Devins et al (four year follow-up) [14] and Christensen et al (average 44 month average follow-up) [15] both failed to find a relationship between depression and mortality. These recent studies, using a wide range of medical and psychosocial control variables (but not dialytic parameters) and sophisticated survival analyses, suggested baseline level of depression was not associated

Fig. 1. Association between the level of depressive affect and survival in 295 patients with end-stage renal disease treated with hemodialysis followed and assessed longitudinally. Cumulative survival estimates were plotted against time for three levels of depression: BDI levels less than 10 (low or nonexistent symptoms, thin line), 10 to 15 (mild levels, medium line), and 16 or greater (moderate to severe levels, broad line). For the time-varying proportional hazards modeling, each patient contributed a separate observation spanning the time from each BDI assessment until the next assessment or death or the end of the study. The cumulative survival estimates plotted are based on the same pooling of observation periods. Patients with depression had greater mortality risk than patients without symptoms of depressive affect, although there was no difference in risk between patients with mild or more severe symptom levels.

with mortality, although Christensen et al found perception of social support predicted survival, as we did [16]. On the other hand, the deleterious effects of depression, measured at the outset, have been often found in older studies, usually without fully controlling for variations in medical comorbidity, nutrition, or the effects of dialytic interventions [8–12]. Foster, Cohn, and McKegney showed that HD patients who exhibited less evidence of general psychopathology during initial interviews had improved survival over a two-year observation period [8]. Wai et al found that at an 18-month followup of home dialysis patients, younger age, lower depression and stress scores, and higher serum albumin concentrations were associated with improved patient survival [9]. Likewise, Burton et al studied home dialysis patients starting ESRD therapy, measuring psychological, physical, social function and support, personality profiles, and economic well-being parameters [10]. After a two-year period, the mean depression score was lower in the group

Kimmel et al: Depression and survival in HD patients

of survivors compared with subjects who died shortly after the initial interview. Ziarnik, Freeman, and Sherrard also noted an association of depression with poorer one-year survival, but not longer survival [11]. Shulman, Price, and Spinelli demonstrated that the BDI score and age, which were unrelated at original evaluation, were the significant predictors of mortality in their ESRD population [12]. Survival differences between ESRD patients with high and low levels of depression were maximal over the first two years of observation [12]. We demonstrated similar findings in an earlier study of a different population of patients from a similar urban catchment group [13]. These latter studies suggested that depression might mediate poor survival for HD patients, but over shorter observation periods. These data, plus our initial visual inspection of the survival curves, lead us to perform a time-varying covariate analysis of the effects of depression on HD patients’ survival. Timevarying covariate analyses provide an integrated, longitudinal assessment of the effect of a given change in a parameter over time on a specified outcome. Recent analyses have demonstrated the power of time-varying analyses to delineate important associations of change in serum albumin over time with survival in HD patients [17]. The current analysis provides a more realistic model, since it uses multiple and recent levels of depressive symptoms, rather than a single baseline value to test associations with survival over varying periods of time. The magnitude and effects of unipolar depression may be experienced with intensity during a relatively short period. Alternatively, patients with a unipolar depressive illness may experience high levels of depression with relatively few changes in magnitude. Time-varying analyses can distinguish between these two patterns. In addition, this type of analysis is especially important with a parameter likely to be variable, in contrast to other mortality markers in HD patients, such as age, gender, and ethnic background, which are perforce immutable, and it takes advantage of the longitudinal design of our study with multiple measurements of depressive affect. Studies have demonstrated that depression may predict survival in patients hospitalized with a myocardial infarction [2, 3] or with a general medical problem [4, 5]. In a patient with a chronic illness, such as ESRD, treated as an outpatient, such signal events as hospitalizations or myocardial infarctions are not useful for planning preventive interventions. High levels of depressive affect in chronically ill outpatients may develop at any time. Depression may be a reaction to a particular stressor, such as loss of a loved one, or a medical setback, or may develop gradually as a function of highly negative perceptions of the effects of the illness on functioning and quality of life. Our findings from this and previous studies suggest that when the patient experiences high levels of depression, even after an initial assessment, a period of time exists during which depression may have deleterious mortality consequences.

2097

Time-varying analyses allow for multiple time point assessments of level of depression and can capture both stable levels of depression and dynamic changes in depression. Patients may adapt to a stressor or their affect may improve, resulting in increased survival, while other patients whose mood has remained negative over a long period of time or worsened over a short period of time may succumb to a combination of the psychologic and medical disorders. A recent study by Friend et al, also using a time-dependent approach, demonstrated that depression predicts changes in serum albumin levels in patients with ESRD, but the converse does not occur [34]. Such data not only underscore the importance of controlling analyses for medical factors, but provide possible mechanisms for mediating the adverse effects of depressive affect on patient outcomes, such as effects on nutrition and the immune system, as we previously hypothesized [1]. Alternatively, depressive affect may be a marker for unmeasured mortality risk factors, such as maladaptive health behaviors [1, 5, 16]. However, our findings suggest that the symptom of depressive affect, rather than solely a marker event, may be the locus for intervention in an outpatient setting. Such notions would suggest that surveillance of ESRD patients for depression would be useful on an ongoing basis. Although the observed magnitude of the association between depression scores and survival is modest, these levels are of the same order of magnitude as effects of medical predictors of outcome delineated in other studies [16, 31]. If the observed effect size were confirmed in a clinical trial, the public health consequences could be substantial. For example, a treatment capable of producing a 8.1-point drop in BDI scores among moderately or severely depressed patients (an achievable effect) [35] could lead to an estimated 32% increase in survival time among HD patients. Our results do not necessarily demonstrate causality. It is possible that medically related deterioration among some patients leads to greater depressive symptoms prior to death. Notions of causality can only be tested in a trial using a clinical intervention. The results, however, are particularly applicable to a group of patients, such as older African Americans, who may be specifically at risk for developing depressive symptoms [36, 37] and who are overrepresented in the U.S. ESRD program [31]. The findings become trenchant when considered in the light that depression is a problem of the older population [36] and is relatively common in the ESRD population [37], and that the mean age of the US dialysis population is increasing at a rapid rate [31]. Our findings must therefore be replicated in more diverse ethnic populations within the ESRD program. The findings, however, have important implications for treatment of all HD patients: A high level of self-reported depression may indicate “a window of opportunity” during which psychopharmacological or cognitive therapy might enhance patient sur-

2098

Kimmel et al: Depression and survival in HD patients

vival. Such interventions might be short term, perhaps only involving counseling in contrast to medication. However, the introduction of new, potent antidepressants may provide enhancements of quality of life and survival in patients treated for ESRD with HD [1]. Such notions should be assessed in rigorously controlled, prospective clinical trials. ACKNOWLEDGMENTS This work was supported by National Institutes of Health grant 1-RO1 DK45578. We are grateful to David Reiss, M.D., and Juan P. Bosch, M.D., for preliminary discussions regarding these findings. We thank the nursing, social work, and dietary staff of the outpatient HD units of the GWUMC, the HUMC, and the VAMC for cooperating in this study. We are indebted to the hard-working research staff and volunteers who collected and entered these data and to Ms. Nancy Armistead and the staff of ESRD Network 5, who provided data on vital statistics of patients enrolled in the study. Most of all, we thank our patients who continue to teach us about chronic renal disease, for volunteering to participate in this study. Reprint requests to Paul L. Kimmel, M.D., Division of Renal Diseases and Hypertension, Department of Medicine, George Washington University Medical Center, 2150 Pennsylvania Avenue, N.W., Washington, D.C. 20037. E-mail: [email protected]

REFERENCES 1. Kimmel PL, Weihs KL, Peterson RA: Survival in hemodialysis patients: The role of depression. J Am Soc Nephrol 4:12–27, 1993 2. Ruberman W, Weinblatt E, Goldberg JD, Chaudhary BS: Psychosocial influences on mortality after myocardial infarction. N Engl J Med 311:552–559, 1984 3. Frasure-Smith N, Lesperance F, Talajic M: Depression and 18month survival after myocardial infarction. Circulation 91:999–1005, 1996 4. Herrmann C, Brand-Driehorst S, Kaminsky B, Leibing E, Staats H, Ruger U: Diagnostic groups and depressed mood as predictors of 22-month mortality in medical inpatients. Psychosom Med 60:570– 577, 1998 5. Covinsky KE, Kahana E, Chin MH, Palmer RM, Fortinsky RH, Landefeld CS: Depressive symptoms and 3-year mortality in older hospitalized patients. Ann Intern Med 130:563–569, 1999 6. Murray CJ, Lopez AD: Global mortality, disability and the contribution of risk factors: Global Burden of Disease Study. Lancet 349:1436–1442, 1997 7. Murray CJ, Lopez AD: Alternative projections of mortality and disability by cause 1990-2020: Global Burden of Disease Study. Lancet 349:1498–1504, 1997 8. Foster FG, Cohn GL, McKegney PF: Psychobiologic factors and individual survival on chronic renal hemodialysis: Two year followup: Part I. Psychosom Med 35:64–82, 1973 9. Wai L, Burton H, Richmond J, Lindsay RM: Influence of psychosocial factors on survival of home dialysis patients. Lancet 2:1155– 1156, 1981 10. Burton HJ, Kline SA, Lindsay RM, Heidenheim AP: The relationship of depression to survival in chronic renal failure. Psychosom Med 48:261–269, 1986 11. Ziarnik JP, Freeman CW, Sherrard DJ, Calsyn DA: Psychological correlates of survival on renal dialysis. J Nerv Ment Dis 164:210– 213, 1977 12. Shulman R, Price JDE, Spinelli J: Biopsychosocial aspects of long-term survival on end-stage renal failure therapy. Psychol Med 19:945–954, 1989 13. Peterson RA, Kimmel PL, Sacks CR, Mesquita ML, Simmens SJ, Reiss D: Depression, perception of illness and mortality in patients with end-stage renal disease. Int J Psychiatry Med 21:343– 354, 1991

14. Devins GM, Mann J, Mandin H, Paul LC, Hons RB, Burgess ED, Taub K, Schorr S, Letourneau PK, Buckle S: Psychosocial predictors of survival in ESRD. J Nerv Ment Dis 178:127–133, 1990 15. Christensen AJ, Wiebe JS, Smith TW, Turner CW: Predictors of survival among hemodialysis patients: Effect of perceived family support. Health Psychol 13:521–525, 1994 16. Kimmel PL, Peterson RA, Weihs KL, Simmens SJ, Alleyne S, Cruz I, Veis JH: Psychosocial factors, behavioral compliance and survival in urban hemodialysis patients. Kidney Int 54:245–254, 1998 17. Culp K, Flanigan M, Lowrie EG, Lew NL: Modeling mortality risk in hemodialysis patients using laboratory values as time dependent covariates. Am J Kidney Dis 28:741–746, 1996 18. Kimmel PL, Peterson RA, Weihs KL, Simmens SJ, Boyle DH, Verme D, Umana WO, Veis JH, Alleyne S, Cruz I: Behavioral compliance with dialysis prescription in hemodialysis patients. J Am Soc Nephrol 5:1826–1834, 1995 19. Kimmel PL, Peterson RA, Weihs KL, Simmens SJ, Boyle DH, Umana WO, Alleyne S, Cruz I, Veis JH: Psychological functioning and behavioral compliance in patients beginning hemodialysis. J Am Soc Nephrol 7:2152–2159, 1996 20. Folstein MF, Folstein SE, McHugh PR: Mini-mental state: A practical method for grading the cognitive states of patients for the clinician. J Psychiatr Res 12:189–198, 1975 21. Vijayvargiya R, Bosch JP: Dialysis and transplantation in patients with HIV infection, in Renal and Urologic Aspects of HIV Infection, edited by Kimmel PL, Berns JS, Stein JH, New York, Churchill Livingstone, 1995, pp 253–277 22. Plough AL, Shwartz M, Salem SR, Weller JM, Ferguson CW: Severity analysis in end stage renal disease: A risk group approach. Am Soc Artif Intern Org J 8:33–40, 1985 23. Hakim RM, Held PJ, Stannard DC, Wolfe RA, Port FK, Daugirdas JT, Agodoa L: Effect of the dialysis membrane on mortality of chronic hemodialysis patients. Kidney Int 50:566–570, 1996 24. Beck AT, Steer RA, Garbin MG: Psychometric properties of the Beck Depression Inventory: Twenty-five years of evaluation. Clin Psychol Rev 8:77–100, 1988 25. Stehouer RS: Beck Depression Inventory, in Test Critiques Compendium, edited by Keyser DJ, Sweetland RC, Kansas City, Test Corporation of America 1987, pp 24–28 26. Craven JL, Rodin GM, Littlefield C: The Beck Depression Inventory as a screen device for major depression in renal dialysis patients. Int J Psychiatry Med 18:365–374, 1988 27. Sacks CR, Peterson RA, Kimmel PL: Perception of illness and depression in chronic renal disease. Am J Kidney Dis 15:31–39, 1990 28. Lee EL: Statistical Methods for Survival Data Analysis (2nd ed). New York, Wiley, 1992, pp 250–263 29. Allison PD: Survival Analysis Using the SAS System: A Practical Guide. Cary, SAS Institute, Inc., 1995, pp 138–154 30. SAS Institute Inc. SAS/STAT Software: Changes and Enhancements Through Release 6 11. Cary, SAS Institute, Inc., 1996, pp 807–884 31. United States Renal Data System: USRDS 1998 Annual Report. National Institute of Diabetes and Digestive and Kidney Diseases, National Institutes of Health, Bethesda, 1998 32. Nelson EE, Changgi DH, Pesce AL, Peterson DW, Singh S, Pollack VE: Anthropometric norms for the dialysis population. Am J Kidney Dis 16:32–37, 1990 33. Consensus Development Conference Panel: Morbidity and mortality of renal dialysis: A NIH Consensus Conference statement. Ann Intern Med 121:62–70, 1994 34. Friend R, Hatchett L, Wadhwa NK, Suh H: Serum albumin and depression in end-stage renal disease. Adv Perit Dial 13:155–157, 1997 35. Reynolds CF III, Frank E, Perel JM: Nortriptyline and interpersonal psychotherapy as maintenance therapies for recurrent depression. JAMA 281:39–45, 1999 36. Blazer DG, Landerman LR, Hays JC, Simonsick EM, Saunders WB: Symptoms of depression among community-dwelling elderly African-American and White older adults. Psychol Med 28:1311– 1320, 1998 37. Kimmel PL, Thamer M, Richard CM, Ray NF: Psychiatric illness in patients with end-stage renal disease. Am J Med 105:214–221, 1998