There is a high cardiac mortality in patients on long-term renal dialysis. No studies have reported long-term outcomes relating to both high-sensitivity cardiac troponin T (hs-cTnT) and high-sensitivity cardiac troponin I (hs-cTnI) in these patients. Patients who underwent long-term dialysis at the Canberra Hospital had blood samples collected for both cardiac and other biomarkers. Samples were stored at −80°C until analysis. Mortality data were collected at 5 years, and univariate and multivariate analyses were performed to identify which biomarkers were predictive of mortality at 5 years. After multivariate analysis, albumin, C-reactive protein (CRP), and hs-cTnT remained independently predictive of all-cause mortality, with hs-cTnT having the highest hazard ratio. If hs-cTnT was excluded from the analysis, then hs-cTnI was independently predictive of mortality. For hs-cTnT, for both genders, the ninety-ninth percentile, derived from a population with subjects with subclinical disease excluded, served as an excellent partition between survivors and nonsurvivors. Receiver-operating characteristic curve analysis for hs-cTnT had area under the curve of 0.798 and for hs-cTnI of 0.774. Kaplan-Meier curves for the aggregation of albumin, CRP, and hs-cTnT showed a stronger predictive power with receiver-operating characteristic area under the curve of 0.805. The addition of echocardiographic data in an analysis of all patients who had an echocardiogram for clinical reasons (n = 105) did not alter the final observations in this subgroup. In conclusion, hs-cTnT retains a superior predictive power in a dialysis-dependent population for identifying those at risk for death and when aggregated with albumin and CRP also has substantial additive value for identifying mortality risk in a renal-dialysis population.
There is a high mortality in patients who underwent long-term renal dialysis and the predominant cause of death is cardiac disease. These patients are frequently asymptomatic and the identification of subjects with a high mortality risk is a potentially useful clinical objective. Shortly after cTn assays became routinely available, it was demonstrated that both cTnT and cTnI identified those patients at increased risk of death. In the present study, we followed a cohort of patients with dialysis-dependent chronic renal failure for >5 years. We report the relative long-term predictive performance for assessing mortality for multiple biomarkers including both cTnT and cTnI measured by high-sensitivity assays.
Methods
This study was approved by the Australian Capital Territory Health Human Research Ethics Committee, and informed written consent was obtained from all participants. We recruited 143 patients who underwent either hemodialysis or peritoneal dialysis through the Canberra Hospital. The age range of our patient population was 15 to 82 years, with a median age of 64 years. The design was at baseline a cross-sectional study with longitudinal follow-up, with all-cause mortality as the primary end point. The details of this study have been described previously. There were no exclusion criteria apart from declining to participate. Only 1 patient from this patient group refused to take part in the study. Over 5 years, only 3 patients were lost to follow-up.
All blood samples were collected immediately predialysis into Vacuette tubes (Greiner Bio-one GmbH, Kremsmunster, Austria) containing potassium EDTA for NT-proBNP analysis and lithium heparin with a separator gel for all other assays. Samples were centrifuged at 3,000 g for 10 minutes and aliquots for each assay stored at −80°C. Before assay, each aliquot was thawed and recentrifuged at 3,000 g for 10 minutes.
Assay details for cTnT, NT-proBNP, and C-reactive protein (CRP) were as described previously. cTnI was measured by a high-sensitivity assay on a Ci16200 analyzer (Abbott Diagnostics, Sydney, Australia). The limit of detection (LoD) was 1.0 ng/L, and the concentration corresponding to the 10% coefficient of variation was 3.8 ng/L. Albumin was measured on an Abbott Ci8200 analyzer with a coefficient of variation of 1.2% at a concentration of 41.4 g/L.
Samples used for analysis were subject to only a single freeze-thaw cycle. The long-term stability of both hs-cTnT and hs-cTnI under these conditions has been documented elsewhere.
The results of echocardiograms performed for clinical reasons on 105 patients were included for a subgroup sensitivity analysis.
Variables were compared between those alive and dead. Variables were included in the multivariate analysis when a significant univariate difference was found (using p <0.01 because of multiple comparisons). The Cox proportional hazards regression model was used to investigate the effect of variables selected because of a univariate or clinically expected relation with survival and to calculate multivariate hazard ratios with 95% confidence intervals. We assessed long-term survival outcomes in a Kaplan-Meier model for relevant variables using the Mantel log-rank test to evaluate differences.
For relevant continuous variables, the clinical utility was assessed by comparing the area under the curve (AUC) of variables related to mortality identified in the multivariate analysis.
As no cut-off points for defining normal values has been universally accepted, we used our previously documented values, namely the ninety-ninth percentiles of an apparently healthy population, and the same population after subjects with subclinical disease were excluded using objective laboratory and clinical criteria. For men, the cTnI ninety-ninth percentile before exclusion was 36.8 ng/L and after exclusion 27.2 ng/L. For women, the corresponding ninety-ninth percentiles were 17.6 and 13.1 ng/L. For cTnT, the respective ninety-ninth percentile values were 44.0 and 18.5 ng/L for men and 35.3 and 21.4 ng/L for women.
A number of sensitivity analyses were performed because of the small sample size and this being a single-center study. First, a simple grading method for combining those biomarker risk factors identified as having an independent association with all-cause mortality was pre-planned before this last analysis. This sensitivity analysis predicted that the presence of increasing numbers of biomarker risk factors should increase the predictive value for all-cause mortality. The grading was into categorical variables represented as 0 for no risk factors, 1 for 1 risk factor, 2 for 2 risk factors, and 3 for 3 risk factors. This categorical variable was to be entered into the same analyses as the identified risk factors. Second, the results of the loss-to-follow-up group were reported as a separate group, and the data reanalyzed with these patients lost to follow-up included in the dead and then in the alive group to assess the impact on the multivariate results. Third, the hs-cTnT and hs-cTnI assay concentrations were entered into the same regression model to identify which was the stronger predictor of death. Finally, all patients who had an echocardiogram performed for clinical reasons were analyzed as a subgroup (n = 105) using exactly the same methods as the whole sample to determine if any standard clinical echocardiographic measurement altered the predictive accuracy of the biomarkers and clinical features in that subgroup. The variables used were left ventricular (LV) wall thickness, LV mass index, LV chamber size, left atrial chamber size, and global LV ejection fraction (EF).
All relevant clinical and laboratory data documented were statistically analyzed with SPSS, version 18 (IBM, Sydney, Australia). Biomarkers with skewed distributions (CRP, NT-pro-BNP, BNP, hs-cTnT, and hs-cTnI) were natural log (log e ) transformed that gave a normal distribution.
Results
Five years after being enrolled in this study, 68 of 143 subjects were still alive (47.6%), 72 were dead (50.3%), and 3 were lost to follow-up (2.1%). Biomarkers and other potential risk factors associated with all-cause mortality are identified in Table 1 . Older patients, having a renal transplant, and length of weekly dialysis were significant predictors of outcome. Albumin, CRP, NT-proBNP, and hs-cTnT and hs-cTnI were all associated with all-cause mortality. Gender was not a significant contributor to mortality.
| Variable ∗ | Alive (n=68) | Dead (n=72) | p value † | Lost to follow-up (n=3) |
|---|---|---|---|---|
| Female/Male § | 22/46 | 27/45 | Ns (χ 2 ) | 2/1 |
| Age in years (SD) ‡ | 53.23 (15.017) | 67.37 (10.3) | <0.001 | 37.67 (29.14) |
| Transplant/no transplant § | 27/41 | 1/71 | <0.001 | 3 unknown |
| Time on dialysis in months at study beginning (SD) ‡ | 41.74 (41.56) | 39.48 (30.06) | ns | 39.00 (56.31) |
| Hours per session (SD) ‡ | 4.34 (0.72) | 4.04 (0.37) | 0.006 | 3.83 (0.29) |
| Frequency of dialysis per week (SD) ‡ | 2.95 (0.28) | 2.96 (0.27) | ns | 3.0 (0) |
| NT-proBNP ng/L (IQR) ¶ | 352 (128-727) | 912 (260-3931) | <0.001 | 283 (86-1434) |
| BNP ng/L (IQR) ¶ | 96.7 (34.6-727) | 140 (72-460) | 0.005 | 224 (134-243) |
| Albumin concentration g/L (SD) ‡ | 43.56 (3.98) | 39.19 (6.45) | <0.001 | 40.33 (9.07) |
| Troponin T μg/L median (IQR) ¶ | 0.018 (0-0.58) | 0.070 (0.034-0.126) | <0.001 | 0 (0-0.040) |
| TnT measurable/total (%) ¶ | 40/68 (58.8%) | 67/72 (93.1%) | <0.001 | 1/3 (33.3%) |
| New hs-TnT ng/L, median (IQR) ¶ | 37.9 (20.47-64.46) | 79.2 (54.20-127.10) | <0.001 | 25.9 (6.38-64.09) |
| Above hs-TnT cutoff ng/L/total measured (%) | 43/66 (65.2%) | 69/70 (99%) | <0.001 | 2/3 (66.7%) |
| New hs-TnI ng/L median (IQR) | 13.06 (7.62-22.79) | 29.66 (17.59-49.26) | <0.001 | 6.11 (1.08-4.05) |
| CRP mg/L median) IQR) ¶ | 4.10 (2.00-9.58) | 11.38 (5.31-26.13) | <0.001 | 2.30 (0.96-21.80) |
| HDL cholesterol mmol/L (SD) | 0.96 (0.29) | 0.97 (0.30) | ns | 0.90 (0.28) |
| Total cholesterol mmol/L (SD) | 4.13 (1.03) | 3.82 (0.88) | 0.063 | 4.08 (0.82) |
| Triglyceride mmol/L (SD) | 1.93 (0.98) | 1.76 (1.09) | ns | 2.29 (1.09) |
| Calcium concentration mmol/L (SD) | 2.39 (0.14) | 2.35 (0.15) | ns | 2.36 (0.28) |
| Phosphate concentration mmol/L | 1.72 (0.60) | 1.57 (050) | ns | 2.05 (0.93) |
| Sodium concentration mmol/L (SD) | 140.2 (3.4) | 139.1 (3.1) | 0.073 | 137 (2.0) |
| Diabetic/not diabetic (N) § | 13/55 | 25/47 | 0.039 (χ 2 ) | 0/0 |
| Hypertension (yes/no) | 27/34 | 28/38 | ns | 0/0 |
| LV diastolic volume cm (SD) | 4.97 (0.66) | 4.93 (0.67) | ns | Not measured |
| LV systolic dimension cm (SD) | 3.11 (0.69) | 3.19 (0.91) | ns | Not measured |
| Septal wall diameter cm (SD) | 1.18 (0.24) | 1.21 (0.20) | ns | Not measured |
| Posterior wall diameter cm (SD) | 1.15 (0.23) | 1.18 (0.20) | ns | Not measured |
| Left atrial diameter cm (SD) | 4.22 (0.74) | 4.29 (0.67) | ns | Not measured |
| Left ventricular ejection fraction % (SD) | 66.6 (9.60) | 61.2 (13.8) | 0.022 | Not measured |
| Left ventricular mass indexed g/m 2 (SD) | 118.5 (39.5) | 129.7 (29.6) | 0.208 | Not measured |
∗ Variables reported as mean, median or number as indicated.
† Differences tested by the ‡ independent t-test; non-parametric and categorical variables as indicated by § Pearson chi-square or ¶ Mann-Whitney U test as indicated.
In the echocardiographic subgroup sensitivity analysis (n = 105, 73.4% of the total population), the only univariate echocardiographic measurement different in the 2 outcome groups was the LVEF. This was significantly lower in the group who died ( Table 1 ).
After a multivariate Cox regression analysis, only age, albumin, log e CRP, and log e hs-cTnT remained independently predictive of all-cause mortality, with log e hs-cTnT having the strongest hazard ratio ( Table 2 ).
| Variable | Significance | Hazard ratio | 95% confidence interval for Hazard Ratio | |
|---|---|---|---|---|
| Hours per dialysis session | 0.026 | 0.486 | 0.257 | 0.919 |
| Age | 0.001 | 1.041 | 1.016 | 1.067 |
| Albumin | <0.001 | 0.886 | 0.834 | 0.941 |
| Log e CRP | <0.001 | 1.658 | 1.253 | 2.194 |
| Log e hsTnT | <0.001 | 3.452 | 2.003 | 5.947 |
Other biomarkers were not identified as independent predictors of all-cause mortality. However, patients in the highest quartile for NT-proBNP appear to have a particularly poor prognosis. Figure 1 shows Kaplan-Meier survival curves for NT-proBNP, with subjects in the highest quartile having only an 18% survival at 5 years.
For hs-cTnT, 142 of 143 patients (99.3%) had results more than the LoD of 5.0 ng/L. For men, 6 of 72 patients (8.3%) had results more than the gender-specific ninety-ninth percentile of 18.5 ng/L for a healthy population with patients with subclinical cardiac disease removed and 10 of 51 (19.6%) women had hs-cTnT more than the gender-specific ninety-ninth percentile of 21.4 ng/L.
For hs-cTnI, 142 of 143 patients (99.3%) had hs-cTnI more than the LoD of 1.0 ng/L. Different patients had the single result less than the LoD. For men, 5 of 72 patients (6.9%) had hs-cTnI more than the gender-specific ninety-ninth percentile of 27.2 ng/L, and 26 of 51 (51.0%) women had hs-cTnI more than the gender-specific ninety-ninth percentile of 13.1 ng/L.
In Figures 2 and 3 , Kaplan-Meier survival curves for women and men for both cTnT and cTnI show a separation between survivors and nonsurvivors which is better delineated when the ninety-ninth percentile after exclusion of subjects with subclinical cardiac disease is used. No women and no men, with results less than these gender-specific cut points died over the 5 years of the study. For cTnI, the picture is less clear-cut. For both men and women, whichever cut point is used, there were a significant number of deaths in subjects with results both more than and less than that concentration.
