Temporal Relation among Depression Symptoms, Cardiovascular Disease Events, and Mortality in End-Stage Renal Disease: Contribution of Reverse Causality

Study Design and Participants

The Choices for Healthy Outcomes in Caring for ESRD (CHOICE) study is a national prospective cohort study of 1041 incident dialysis patients, aged 19 to 95 yr, from 81 dialysis clinics associated with Dialysis Clinics, Inc. (Nashville, TN; n = 923), New Haven CAPD (New Haven, CT; n = 88), or St. Raphael’s Hospital (New Haven, CT; n = 30) (14–16). Participants were enrolled from October 1995 to June 1998 a median of 45 d after dialysis initiation. To be eligible, patients had to be initiating outpatient hemodialysis or peritoneal dialysis for ESRD, to be 18 yr or older, and to speak English or Spanish. The protocol was approved by Institutional Review Boards, and patients gave written consent to participate. This analysis was limited to 917 (88%) participants with assessments of DS available at enrollment.

Data Collection

Baseline demographics, clinical data (including coexistent medical comorbidity assessed using the Index of Coexistent Disease [ICED] [17]), and antidepressant use were obtained from medical records and questionnaires. Baseline and follow-up assessments of DS (18–20) and laboratory studies, including creatinine, albumin, calcium, phosphorus, total cholesterol, hematocrit, C-reactive protein (CRP), IL-6, and white blood cell (WBC) counts, also were obtained.

Clinical and Laboratory Assessments

Baseline systolic and diastolic BP included the average of all predialysis sessions’ measures during the first 3 mo after enrollment (median 31 readings). Presence of diabetes, CVD (history of myocardial infarction, cardiac revascularization, stroke, carotid endarterectomy, peripheral vascular disease or revascularization, angina, or positive stress test), and heart failure was determined from medical record review.

Nonfasting serum specimens were drawn before dialysis a median of 5 mo from dialysis initiation, shipped overnight on dry ice to a central laboratory, and stored at −80°C. Total cholesterol was measured using colorimetric methods (coefficient of variation [CV] 5.3%), albumin using the Bromocresol Green method (CV 1.1%), creatinine using a modified Jaffe method, hematocrit and WBC count with the AVIA-120 Bayer autoanalyzer (Tarrytown, NY), CRP using a colorimetric competitive ELISA (CV 8.9%), and IL-6 by an ultrasensitive ELISA method (CV 7%). Albumin, calcium, phosphorus, hematocrit, and WBC count were obtained predialysis and were averaged over each 6-mo follow-up period (median three to four measurements per 6 mo).

Assessment of DS

DS were assessed at baseline and at three intervals (6, 12, and 18 mo) using the self-administered five-item mental health subscale of the Medical Outcomes Study Short Form-36 (MHI-5) (16,18–20). MHI-5 items ask respondents how often during the past month they have felt “nervous,” “down in the dumps,” “calm and peaceful,” “downhearted and blue,” and “happy.” Items were scored on a scale of 0 to 100, with greater scores indicating better mental health and well-being (19). MHI-5 items have concurrent validity in identifying DS in ESRD measured using the Beck Depression Symptoms Index (21). DS were assessed in dichotomous (considered present when MHI scores were ≤52) and continuous manners. Scores ≤52 have good specificity and sensitivity for the detection of DS in chronic illness (20,22–25).

Outcome Assessment

Deaths from all causes were ascertained by active surveillance through dialysis clinics and passive surveillance using Centers for Medicare and Medicaid Services data. Of 324 deaths, 126 (59% of 214 in-hospital deaths) terminal hospitalization medical records were available and were reviewed for underlying cause of death using an algorithm modified from the Cardiovascular Health Study (26,27).

A composite CVD outcome was composed of the first fatal (death secondary to coronary artery disease, stroke, and peripheral vascular disease) or nonfatal (abdominal aortic aneurysm, extremity amputation, peripheral arterial bypass, coronary artery bypass grafting, carotid endarterectomy, stroke including intracranial hemorrhage, myocardial infarction, and coronary angioplasty) cardiovascular event during follow-up (28). Events coders were unaware of MHI-5 responses.

Statistical Analyses

We performed bivariate (χ² and t test) analysis to ascertain associations between patient characteristics or date of study enrollment and DS at baseline. In survival analysis, we assessed differences in time to death and CVD events up to 2 yr after enrollment. Individuals were censored as a result of leaving a CHOICE clinic (n = 226 [22%]), transplant (n = 151 [15%]), or end of follow-up (n = 386 [37%]).

To ascertain temporal relations, we assessed (1) the association of presence and duration of DS with outcomes and (2) a possible reverse causal association (medical comorbidity preceding onset of DS). We used Cox proportional hazards models to incorporate DS assessed (1) at baseline only, (2) in a time-varying manner (baseline and follow-up at 6, 12, and 18 mo), or (3) in a time-varying manner with a 6-mo time lag in which only DS that were assessed at least 6 mo before the event were related to outcomes.

In baseline only multivariable analyses, covariates were demographics, comorbidity, modality, antidepressant use, creatinine, CVD risk factors, BP, inflammation, hematocrit, albumin, calcium, and phosphorus. For time-varying models, follow-up measures of BP, hematocrit, creatinine, albumin, calcium, phosphorus, and WBC count also were covariates. We performed stratified analyses and overall multivariable analyses using interaction terms to assess differences in the magnitude or direction of associations according to demographics, diabetes status, modality, comorbid disease status, and smoking status. We performed a sensitivity analysis to assess whether controlling for other aspects of quality of life (physical health and vitality, social function, and role emotional function) would change results.

To investigate further the association of the proximity of measurement and the persistence of DS with outcomes, we categorized patients as having had “distant” DS (at least 6 to 12 mo before event occurrence) or “current” DS (within 6 mo of event occurrence), and we created four mutually exclusive categories for patients with (1) distant and current DS (persistent symptoms), (2) current symptoms only, (3) distant symptoms only, and (4) neither distant nor current symptoms (reference group).

We used ICED scores at baseline and 1 yr of follow-up to determine changes in medical comorbidity during the first year of dialysis and to explore further a possible reverse causal relationship between poor health and DS (29). In the subgroup of participants who had no DS at 1 yr of follow-up and who also had measurement available at 2 yr of follow-up (n = 283), we used generalized estimating equation models for repeated measures to ascertain the relation of a change in ICED score during the first year of dialysis to changes in DS scores at 2 yr of follow-up (16,30–33). All analyses accounted for clustering within clinics using robust variance methods in STATA (release 8.0; StataCorp, College Station, TX) statistical software (34).

Prevalence of DS and Participant Baseline Characteristics

DS were prevalent among 24% of participants at baseline, 23% at 6 mo, 19% at 12 mo, and 21% at 18 mo. The incidence of DS was 14% at 6 mo, 9% at 12 mo, and 13% at 18 mo. Patients who were enrolled within 30 d of dialysis initiation (n = 294) were no more likely than patients who were enrolled later to report DS at baseline (23 versus 25%, respectively; P = 0.64). Black patients, those with less than a high school education, those taking antidepressants, and current smokers were more likely to have prevalent DS. Systolic BP and WBC count were greater among patients with DS versus their counterparts without symptoms (Table 1).

Timing of DS Measurement and Relation with Outcomes

During 1468 person-years of follow-up (mean follow-up 1.5 yr; range 67 d to 2 yr), 189 deaths (51% CVD deaths) and 272 CVD events occurred. Crude cumulative mortality from any cause in Kaplan-Meier plots was no different between patients with and without baseline DS, greater in patients with symptoms assessed in a time-varying manner, and greater in patients with symptoms assessed in a time-varying manner with incorporation of a 6-mo time lag between symptom measurement and mortality (Figure 1).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1.

All-cause mortality (unadjusted) according to presence of depression symptoms (DS) measured at baseline (A), in time-varying 6-mo intervals (B), and time-varying 6-mo intervals with 6-mo time lag between measurement of DS and death (C). (A) Crude cumulative deaths over follow-up by presence of DS. (B) Crude cumulative deaths over follow-up by presence of DS assessed in a time-varying manner. (C) Crude cumulative deaths over follow-up by DS assessed in time-varying manner with 6-mo time lag. All events in the time-lag analysis are classified with respect to whether DS were present 6 mo before that event. A time lag between the measurement of DS at baseline and the occurrence of events after 6 mo of follow-up is represented by a flat line.

Baseline DS scores ≤52 were associated with only death from non-CVD causes after adjustment (Table 2). In contrast, DS scores ≤52 assessed as a time-varying measure showed a 60% to three-fold increased hazard of death from all causes, death from CVD causes, and CVD events when compared with patients without DS in both unadjusted and adjusted analyses. In subgroup analyses of DS scores ≤52 assessed as a time-varying measure, DS seemed more strongly related to all-cause deaths in patients without diabetes and more strongly related to CVD events in white versus black patients. Although stratified analyses could not be performed by dialysis modality (because of few events among peritoneal dialysis patients), analyses that used an interaction term demonstrated that patients who were on peritoneal dialysis and had DS had a hazard of CVD events 2.36 times (95% confidence interval [CI] 1.05 to 5.32) greater than that for similar patients who were on hemodialysis (Table 3). Results were similar when symptom scores were analyzed as a continuous variable (data not shown).

Relation of Proximity of DS Measurement and Duration of DS to Outcomes

Among patients who were depressed at baseline, symptoms persisted in 13% at 6 mo, 12% at 12 mo, and 11% at 18 mo of follow-up. Current symptoms were present in 11% of participants at 6 mo, 7% at 12 mo, and 10% at 18 mo of follow-up. Distant but not current symptoms were present in 11% of participants at 6 mo, 9% at 12 mo, and 8% at 18 mo of follow-up. Persistent and current symptoms were associated with four-fold greater adjusted risks for death from all causes and two-fold greater adjusted risk for CVD events when compared with patients with no symptoms in the previous 12 mo (Table 4).

Investigation of Reverse-Causal Relation between DS and Comorbid Disease

DS scores ≤52 assessed as a time-varying measure with incorporation of a 6-mo time lag were crudely associated with a 50 to 70% increased hazard of all deaths, non-CVD deaths, and CVD events, representing an attenuation of findings in models that did not incorporate a 6-mo time lag. After adjustment, the magnitude of risk was similar and not statistically significant (Table 2).

Of 711 participants with ICED scores at baseline and 1 yr follow-up, 59% experienced no change in ICED score, 17% experienced improvement, and 22% experienced a decline. Of 572 participants with depression scores at baseline and 1 yr, mean (SD) symptom scores at 1 yr of follow-up were 79.5 (13.0) for patients with no depression at baseline or at 1 yr (n = 393), 43.9 (8.8) for no depression at baseline but depression at 1 yr (n = 48), 69.9 (11.5) for depression at baseline but no depression at 1 yr (n = 69), and 38.7 (11.4) for depression at both baseline and 1 yr (n = 62). Patients with DS at either baseline or follow-up were no more likely than the 22% of patients with no symptoms at baseline or 1 yr to experience a decline in ICED score: Patients with no symptoms at baseline but symptoms at 1 yr (23 versus 22%; P = 0.8), patients with symptoms at baseline but no symptoms at 1 yr (17 versus 22% P = 0.4), and patients with symptoms at both baseline and 1 yr (30 versus 22%; P = 0.2). In a subgroup analysis of 155 participants who had no DS at baseline or 1 yr (with ICED scores present at baseline and 1 yr and depression scores also present at baseline, 1 yr, 18 mo, and 2 yr), 21% experienced a decline in ICED during the first year of follow-up. Among patients who experienced a decline in ICED, 25% experienced incident DS at 18 mo or 1 yr versus 19% in patients who did not experience a decline in ICED (P = 0.43, adjusted odds ratio [95% CI] for incident depression 1.39 [0.53 to 3.64]).

Patients with no change or an improvement in their ICED scores during the first year experienced a decline in their DS scores between the first and second years of follow-up after adjustment for age, gender, race, education, and dialysis modality (mean change [95% CI] −2.06 points [−0.33 to −3.79]). Patients with worsening ICED scores during the first year of dialysis did experience greater declines in symptom scores when compared with patients with no change or improved ICED scores during the first year of dialysis, but the results were not statistically significant (difference [95% CI] in change −2.42 points [1.36 to −6.21]).

Sensitivity Analysis on Health-Related Quality of Life

Results of analyses that incorporated measures of physical and mental health–related quality of life as covariates were consistent with main results (data not shown).