Association of NTproBNP and cTnI with outpatient sudden cardiac death in hemodialysis patients: the Choices for Healthy Outcomes in Caring for ESRD (CHOICE) study

The Choices for Healthy Outcomes in Caring for ESRD (CHOICE) Study is a longitudinal, prospective cohort study of 1041 patients starting dialysis that were recruited from 81 dialysis centers in 19 states. Eligibility criteria included dialysis initiation in preceding 3 months, age 18 years or older, ability to speak English or Spanish, and ability to consent. Each participant completed informed consent. The participants were enrolled from October 1995 to June 1998 and were on average 45 days post-dialysis initiation. A specimen bank was established for all Dialysis Clinic, Inc. (DCI) participants. Our analysis included 503 hemodialysis participants with available stored specimens. The Johns Hopkins Medicine Institutional Review Board and the clinical centers’ review boards approved the study and all participants provided informed consent.

We collected blood samples prior to routine outpatient dialysis session, centrifuged within 30–45 min of collection, and sent them overnight on ice to the central laboratory. We divided each sample into multiple vials and stored at −80 °C till they were thawed and aliquoted for this study. We measured NT-pro-BNP using a one-step sandwich chemiluminescent immunoassay also based on LOCI* technology. We measured cTnI by homogeneous, sandwich chemiluminescent immunoassay based on LOCI* technology. Both NTproBNP and cTnI were measured on the Dimension Vista System at the University of Maryland School of Medicine, Baltimore, Maryland. The coefficient of variation (CV) for NT-pro-BNP was 5.0 % at 107.2 pg/mL and 1.6 % at 275 pg/mL and 1.2 % at 3,313 pg/mL. The reliability correlation coefficient for NT-pro-BNP was 0.998. The CV for cTnI was 9.1 % at 0.090 ng/mL, 5.9 % at 1.08 ng/mL and 1.6 % at 4.87 ng/mL. The reliability correlation coefficient for cTnI in a 5 % sample of masked duplicate specimens was 0.969.

The primary outcome for this study was outpatient SCD defined as an out-of-hospital deaths including deaths that occurred in an emergency department or one in which the patient was reported to be “dead on arrival” with the following codes noted in the National Death Index death certificate data: ICD-9 390–398, 402 or 404–429; and ICD-10, I00-I09, I11, I13 and I20–I51, as previously described [6]. We excluded deaths with hyperkalemia, sepsis or malignancy listed as a contributing cause of death or if the death occurred while under hospice care.

We collected data on participants’ age, sex, race and body mass index (BMI). We adjudicated baseline comorbidities including prevalent cardiovascular disease and left ventricular hypertrophy (as assessed by electrocardiograms) by abstraction of dialysis unit records, hospital discharge summaries, medication lists, consultation notes, diagnostic imaging, and cardiac imaging reports and Index of Coexistent Disease (ICED) scoring. For this index, comorbid conditions, including diabetes mellitus, ischemic heart disease, congestive heart failure, hypertension, peripheral vascular disease, and other conditions, were determined present by two trained nurses, and then, the severity of each comorbid condition for each patient was calculated using Index of Disease Severity (IDS) and Index of Physical Impairment (IPI) (30). These two scores were then used to calculate the ICED score, which serves as a validated medical record-derived index that captures both presence and severity of comorbid conditions^19,20,30. ICED scores range from 0 to 3, with 3 as the highest severity level. We abstracted antihypertensive medication use at baseline from patients’ charts and obtained routine laboratory data from medical records. We measured serum albumin (CV, 1.9 %) in the same specimen as NTproBNP and cTnI at University of Minnesota, Minneapolis, Minnesota. We also used data on C-reactive protein, interleukin-6 and p-selectin that had been previously measured for research [10, 11]. Other laboratory data measured in routine clinical care included hemoglobin, calcium, phosphate, blood urea nitrogen, creatinine, glucose, potassium and bicarbonate. These laboratory data were obtained from the same time as cardiac biomarker assessment.

We compared the baseline characteristics of the participants across categories of NTproBNP and cTnI using chi-squared test for categorical variables and t-tests for continuous variables. Missing data for variables were as follows: educational status (2.8 %), smoking history (2.8 %), BMI (5.6 %) systolic blood pressure (3.8 %), serum potassium, calcium and phosphate (6.8 %) and serum bicarbonate (12 %). Missing data values were imputed with 10 data replicates using multiple imputation by the chained equations method implemented by the ice program in Stata. We modeled NTproBNP and cTnI as continuous variables after natural log transformation and as categorical variables. We categorized NTproBNP as tertiles and cTnI as a low, mid and high category. The low cTnI group included those with cTnI below detection limit; (<0.015 ng/mL; n = 336). Remaining patients are divided into two groups at the median with the mid category referring to group with detectable values but below median (<0.040 ng/mL; n = 85) and high category referring to the group with values at or above median (≥0.040 ng/mL; n = 82). We visualized the association between NTproBNP and cTnI and outcomes by calculating incidence rates for SCD adjusted for age, sex and race using a Poisson regression model with biomarkers modeled as linear spline with 2 knots corresponding to the categories. We used Cox proportional hazards regression to analyze the association between the biomarkers and SCD. We assessed proportional hazards assumptions graphically and by tests of Schoenfeld residuals. We used hazard ratios (HRs) to quantify the associations of biomarkers with SCD after adjustment for a priori defined confounders, including demographic characteristics [age, sex, race (white or other)] and clinical factors [smoking status (ever versus never), ICED score, diabetes, cardiovascular disease, left ventricular hypertrophy, congestive heart failure, BMI and systolic blood pressure], laboratory tests (hemoglobin, serum albumin, serum potassium, serum bicarbonate, serum corrected calcium and serum phosphate) and β-blocker use. We conducted sensitivity analyses further adjusting for biomarkers previously associated with SCD in CHOICE study (i.e., CRP, IL-6, and p-selectin) and accounting for the competing risk of death from other causes using competing-risks regression based on Fine and Gray’s proportional subhazards model. In exploratory analyses we further determined SCD risk association within subgroups by testing for interactions with age, sex, race, cardiovascular disease, diabetes, serum albumin, serum potassium and serum bicarbonate, with continuous variables categorized above or below median. We evaluated risk prediction by NTproBNP and cTnI over 5-years by calculating C statistic and net reclassification improvement (NRI) [12, 13]. We assessed model calibration using modified Hosmer–Lemeshow statistic [14, 15]. We performed all statistical analyses using Stata software, version 12.1 (Stata Corp.). We defined statistical significance as p <0.05 using two-tailed tests.

The final study sample comprised of 503 hemodialysis patients. Compared to the overall cohort, the included patients were less likely to be White and had higher urea creatinine, potassium, calcium, phosphate and albumin and lower hemoglobin and Kt/Vurea (Additional file 1: Table S1). The average age of the participants was 58 years and 54 % were men. At baseline, 57 % of the participants had diabetes, 56 % had cardiovascular disease, and 24 % had a history of myocardial infarction. Table 1 summarizes the baseline characteristics of the participants according to categories of NTproBNP and cTnI. Patients with higher NTproBNP and cTnI were older, more likely to have a history of cardiovascular disease, and have elevated IL-6 levels. In addition, patients with higher NTproBNP were more likely to be White, had lower body BMI and higher CRP whereas those with higher cTnI were more likely to be male.

The 75 SCD events occurred over 1,814 person-year of follow-up (median 3.5 years) with an incidence rate of 41.4 SCD events over 1000 person-years. The causes of death in these 75 patients, cross-tabulated by the causes listed in the National Death Index and Center for Medicare and Medicaid Services (CMS) death notification form (Form 2746) are presented in Additional file 1: Table S2. Fig. 1 presents the incidence rate of SCD adjusted for age, sex and race, demonstrating a linear increase in SCD incidence rate with NTproBNP and a relatively flat association with cTnI. In unadjusted models and minimally adjusted models, both NTproBNP and cTnI were associated with risk of SCD (Table 2). After adjustment for demographics, clinical factors, comorbidities, laboratory tests and β-blocker use, NTproBNP continued to have a statistically significant association [HR, 95 % confidence interval (CI)] with SCD [1.27 (1.13–1.43); p < 0.001]] whereas as the association between cTnI and SCD was not statistically significant [1.17 (0.98–1.40); p = 0.08]. Similar results were noted in the categorical analysis (Table 2). In the fully adjusted models, compared to the lowest tertile of NTproBNP, those in the highest tertile had a more than 3-fold higher risk of SCD [3.03 (1.56–5.89); p = 0.001]. Compared to the low cTnI category (undetectable cTnI), those in the mid (detectable and below median) and high cTnI category (detectable and above median) had a non-significant trend towards increased risk of SCD in the fully adjusted models.

Further adjustment for markers of inflammation (CRP, IL6 and p-selectin) did not significantly change the magnitude or the direction of association (data not presented). Analyses using competing risks models showed results similar to the primary analysis (Additional file 1: Table S3).

The subgroup analyses should be interpreted with caution due to small sample size and multiple comparisons. There were no significant interactions in the models for NTproBNP (Fig. 2a) but there was suggestion of effect modification by cTnI and sex (p-interaction =0.02), baseline cardiovascular disease (p-interaction =0.01), serum albumin (p-interaction =0.008), serum potassium (p-interaction = 0.01) and bicarbonate (p-interaction = 0.02).

Compared to the fully adjusted model, the improvement in 3-year and 5-year risk prediction was greater with NTproBNP than cTnI (Table 3). Addition of both NTproBNP and cTnI to the model did not lead to further improvement in risk prediction.