Predictive Properties of Biomarkers GDF-15, NTproBNP, and hs-TnT for Morbidity and Mortality in Patients With Type 2 Diabetes With Nephropathy

The epidemic of type 2 diabetes (T2D) is a major cause of the increased prevalence of chronic kidney disease (CKD), end-stage renal disease (ESRD), and cardiovascular complications (CVCs). Identification and stratification of patients with increased risk may help to tailor therapy and alleviate this burden for individual patients as well as health care systems. The onset and development of diabetic kidney disease is typically characterized by subsequent transitions from normoalbuminuria to microalbuminuria and to macroalbuminuria. The progression of disease is increasingly more difficult to halt in patients with microalbuminuria and macroalbuminuria, respectively (1), and therefore the early identification of patients at risk for ESRD and CVCs is important. However, once macroalbuminuria and renal impairment develop in patients with T2D, the validity of current biomarkers in predicting further progression of renal complications and CVCs is largely unknown.

Previously, we identified growth differentiation factor-15 (GDF-15), also known as macrophage inhibitory cytokine-1, as a clinically valuable marker for predicting progression to microalbuminuria and macroalbuminuria in both T2D and hypertension patients (2), suggesting that GDF-15 may be a valuable marker for individual risk stratification in patients with normoalbuminuria and microalbuminuria. Although GDF-15 can be upregulated in diverse tissues in response to damage, circulating levels of GDF-15 appear to reflect primarily damage to the vascular endothelium (3–5). In addition to GDF-15, we later demonstrated in the same cohort of patients that the cardiac biomarker high-sensitivity troponin T (hs-TnT) also predicted the worsening of albuminuria in both T2D and hypertension (6). This may seem surprising, as hs-TnT was considered a specific marker for cardiac stress including myocardial infarction and heart failure. However, recent data suggest that increased levels of hs-TnT and also N-terminal prohormone of B-type natriuretic peptide (NTproBNP) may indicate a state of microvascular disease, reflecting the development not only of cardiac pathology but also of diabetic nephropathy and diabetic retinopathy (7).

Although we substantiated the predictive properties of GDF-15 and hs-TnT in patients with normoalbuminuria and microalbuminuria, it is unclear whether these biomarkers may also predict further disease progression in patients with T2D who have macroalbuminuria. Evidence for the prognostic properties of GDF-15 in advanced stages of CKD may be found in a study by Breit et al. (8), demonstrating that GDF-15 is an independent serum marker of mortality in patients with ESRD (8). However, this study was performed in patients with incident and prevalent dialysis where GDF-15 most likely reflected protein-energy wasting. In addition to GDF-15, NTproBNP and hs-TnT have been shown in previous studies to enhance the risk prediction of CVCs in patients with T2D (9). Furthermore, in elderly patients without a history of heart disease, increased levels of NTproBNP and hs-TnT were indicative of an accelerated decline of renal function and higher incidences of CKD (10). However, these studies were performed in patients with differences in dialysis treatment and varying levels of renal impairment. As renal impairment or efficacy of replacement therapy may lead to elevated biomarker levels, these may be important confounding factors in these studies.

To investigate whether GDF-15, NTproBNP, and hs-TnT levels predict renal and cardiovascular disease progression in patients with T2D who have nephropathy irrespective of renal function, we determined the baseline serum level in patients with T2D with similarly reduced renal function and macroalbuminuria who had previously been enrolled in the sulodexide macroalbuminuria (Sun-MACRO) trial (11). In a post hoc analysis, we evaluated whether baseline measurements of GDF-15, NTproBNP, and hs-TnT improved a multiparameter prediction of mortality and morbidity end points, both renal and cardiovascular.

The current study is a post hoc biomarker study to investigate whether baseline serum levels of GDF-15, NTproBNP, and hs-TnT predict renal and cardiovascular morbidity and mortality in patients with type 2 diabetes who had advanced diabetic nephropathy. For this, we investigated a cohort of patients with type 2 diabetes who had macroalbuminuria who had previously been enrolled and followed in the Sun-MACRO study (Clinical trial reg. no. NCT00130312, ClinicalTrials.gov). Eight hundred sixty-one patients were included in the current study based on availability of both baseline serum samples and follow-up data. In the Sun-MACRO study, drug (sulodexide) treatment had no effects on primary or secondary end points, and therefore both placebo- and sulodexide-treated patients were included in the present analysis. To ascertain that potential effects of treatment were excluded, all statistical analyses were adjusted for treatment assignment. The current study was conducted according to the guidelines and regulations of the institutional review board of the University Medical Center of Groningen.

The Sun-MACRO trial was a prospective, randomized, controlled trial that has been described in detail previously (11–13). Briefly, the Sun-MACRO trial aimed at demonstrating that the drug sulodexide offered additional benefit in preventing or ameliorating advanced diabetic nephropathy. Sulodexide belongs to a class of drugs called glucosamine glycan and has been shown to reduce albuminuria in patients with type 1 and 2 diabetes through different modes of action (14–20). The primary outcome was a doubling of baseline serum creatinine level (50% loss of kidney function) or ESRD. Patients aged ≥18 years with a diagnosis of T2D and marked proteinuria (urinary protein excretion ≥0.9 g/24 h or >0.9 mg/g creatinine) and serum creatinine >1.3 mg/dL (>1.0 mg/dL in females) were included. Patients with type 1 diabetes, patients with known additional nondiabetic renal disease, and patients with the need for long-term immunosuppressive therapy were excluded. Eligible patients were randomly assigned to treatment with sulodexide 200 mg/day or placebo. After randomization, patients were seen every 3 months and were followed until the occurrence of a renal event, which was defined as a confirmed doubling of serum creatinine levels, a sustained serum creatinine concentration of >6.0 mg/dL, or ESRD, defined as long-term dialysis or renal transplantation. The Sun-MACRO trial was terminated early (2008) based on an interim analysis showing that there was no effect on the surrogate (proteinuria) and no effect on the hard outcome. At the time of termination, safety data were collected up until 30 days after study drug discontinuation, and there was no suggestion for benefit or harm when the cardiovascular or mortality end point was analyzed. Moreover, serious adverse events were similar between the placebo and sulodexide groups. Possible explanations for the failure of the drug may include the inability of sulodexide to ameliorate early but not late kidney disease, poor oral absorption of sulodexide from the gastrointestinal tract, and differences between glycosaminoglycan formulations between the drug used in previous studies and the sulodexide used in the Sun-MACRO study (11). The Sun-MACRO study was conducted according to the Declaration of Helsinki, the institutional review boards of each center approved the trial, and all patients provided written informed consent.

Baseline serum samples were stored at −80°C. GDF-15 level was measured with a novel precommercial assay based on the Eclia (electro-chemiluminescence immunoassay) principle (Roche Diagnostics, Mannheim, Germany) as described previously (2). The detection limit of the assay was 200 pg/mL with an intraindividual coefficient of variation of 6.7–9.2%. The measurement of hs-TnT and NT-proBNP used commercial kits (Roche Diagnostics) as described previously (6,21).

Renal and cardiovascular composite end points of the current study were identical to those of the Sun-MACRO study with one addition. Considering the early termination of the Sun-MACRO study, a sustained 40% decline in estimated glomerular filtration rate (eGFR) was included as an additional renal end point. The validity of GFR decline as an end point for clinical trials in CKD was recently reviewed, and an eGFR decline of 30% was proposed as an alternative surrogate end point in trials of CKD along with stronger evidence for a 40% eGFR decline (22).

Analyses were performed with STATA version 13.1 (StataCorp, College Station, TX). Histograms and normality plots were used for evaluating the normality of data. For data with a nonlinear distribution, log transformation was performed. Graphical methods were used to confirm the normalization of the distribution after transformation. Baseline differences in GDF-15, NTproBNP, and hs-TnT tertiles were analyzed by one-way ANOVA. Data are expressed as the mean ± SD, unless otherwise stated. P values were two tailed, and values <0.05 were considered statistically significant. Kaplan-Meier plots showing the probability of reaching a renal or cardiovascular end point were constructed in Stata. A log-rank test with trend analysis across the three tertiles was performed to test for increased risk in subsequent tertiles.

Cox proportional hazards regression models were used to predict renal and cardiovascular end points. Base models were obtained by backward negative selection of established renal and cardiovascular risk markers, including serum albumin-to-creatinine ratio (ACR), hemoglobin, hemoglobin A_1c (HbA_1c), diastolic and systolic blood pressure, total cholesterol, and smoking. Factors with P > 0.05 were not included in the base model. Age, sex, and treatment were subsequently included in the model. As age and sex are included in the estimate equations for eGFR, serum creatinine level was used as an indicator of renal function. To exclude the potential effects of sulodexide treatment on risk prediction, treatment was included in all risk prediction analyses. The model for renal events included serum creatinine, hemoglobin, ACR, age, and sex. For cardiovascular events, the model included diastolic blood pressure, ACR, total cholesterol, age, and sex. Subsequently, baseline properties of GDF-15, hs-TnT, and NTproBNP (above or below the cut point) were entered into the model to examine whether the inclusion of biomarkers in the model improved risk prediction. Discriminative abilities of the models were estimated as C-statistics for Cox regression models (23). Differences in C-statistics were tested with the χ² test.

Receiver operating characteristic (ROC) curves were constructed for the three biomarkers in relation to renal and cardiovascular end points. Subsequently, optimal cut points for the three biomarkers were estimated using the CUTPT plugin for Stata (written by Philip Clayton, ANZDATA Registry, Adelaide, Australia) using the Liu method (24). Cut points were separately calculated for renal and cardiovascular end points. Patients were marked HIGH (1) if baseline biomarker levels were above the cut point and marked LOW (0) when they were equal to or below the cut point.

Patient characteristics (n = 861) at baseline are shown in Table 1. The mean age of all patients was 64 ± 9 years. The average BMI was 32 kg/m², indicating obesity. The mean HbA_1c level was elevated at 8 ± 1.6% (64 ± 17 mmol/mol), confirming T2D. ACRs were consistent with macroalbuminuria. eGFR was on average 33.1 mL/min/1.73 m², indicating moderate to severe renal impairment (CKD class 3B). In the current cohort, median GDF-15 serum concentration was 3,228 pg/mL (interquartile range [IQR] 2,345–4,310 pg/mL). The median NTproBNP concentration was 380 ng/L (IQR 155–989 ng/L), and the median hs-TnT concentration was 30 ng/L (IQR 20–47 ng/L).

Patient characteristics were subsequently subdivided by tertiles of GDF-15, NTproBNP, and hs-TnT (Table 1). Although all patients were characterized by macroalbuminuria and renal impairment, small but significant differences were present in serum creatinine and eGFR values between tertiles. Furthermore, hemoglobin, ACR, GDF-15, NTproBNP, and hs-TnT values were significantly different between tertiles of all three biomarkers. Age was higher in increasing tertiles of GDF-15 and NTproBNP, whereas the percentage of men differed in tertiles of NTproBNP and hs-TnT. BMI was lower in increasing tertiles of GDF-15. Systolic blood pressure was higher in the middle and highest tertiles of NTproBNP, whereas diastolic blood pressure was lower in increasing tertiles of GDF-15 and NTproBNP. Total cholesterol and HDL cholesterol levels were different in tertiles of hs-TnT. Glucose levels were lower in increasing GDF-15, and HbA_1c level was lower in increasing tertiles of both GDF-15 and NTproBNP. Finally, uric acid was different in tertiles of GDF-15 and hs-TnT.

GDF-15, NTproBNP, and hs-TnT serum concentrations were not normally distributed and were therefore log₂ transformed (as shown for GDF-15 in Fig. 1A and B). Similarly, ACR was log₁₀ transformed. Correlations of log-transformed GDF-15 concentrations with eGFR and log-transformed ACR are shown in Fig. 1C and D. Although statistically significant, weak correlations were found between baseline biomarker levels and eGFR and ACR. Higher GDF-15 serum concentrations associated with lower eGFR and higher ACR (R² = 0.07, P < 0.001 and R² = 0.05, P < 0.001, respectively). Similarly, NTproBNP and hs-TnT had a negative correlation with eGFR (R² = 0.06, P < 0.001 and R² = 0.07, P < 0.001, respectively), whereas NTproBNP and hs-TnT had a positive correlation with ACR (R² = 0.04, P < 0.001 and R² = 0.03, P < 0.001, respectively). Furthermore, a weak correlation was found between GDF-15 and BMI (R² = 0.02, P < 0.001) but not for NTproBNP and hs-TnT.

Median follow-up was 9 months (IQR 4–17 months). During follow-up, 60 patients (7.7%) reached a renal end point and 107 patients (12.4%) reached the secondary CVC end point. Kaplan-Meier curves were constructed (Fig. 2) to plot the proportion of patients with a renal complication or CVC end point over time by tertiles of GDF-15, NTproBNP, and hs-TnT. The probability of reaching an end point was significantly higher with increasing tertiles of GDF-15 (Fig. 2A and B), NTproBNP (Fig. 2C and D), and hs-TnT (Fig. 2E and F) for both renal and CVC end points (GDF-15: P < 0.001 and P = 0.022, respectively; NTproBNP: P = 0.008 and P < 0.001, respectively; hs-TnT: P < 0.001 and P < 0.001, respectively).

Optimal cut points for the three biomarkers were calculated separately for renal and cardiovascular end points. Estimated optimal cut points for renal end points were 3,420 pg/mL for GDF-15, 973 ng/L for NTproBNP, and 40.5 ng/L for hs-TnT. For cardiovascular end points, estimated optimal cut points were 3,420 pg/mL for GDF-15, 407 ng/L for NTproBNP, and 30.6 ng/L for hs-TnT.

A Cox regression analysis was performed including baseline measurements for both renal complication and CVC end points (Table 2). In the clinical reference model consisting of ACR, hemoglobin, and serum creatinine, all factors significantly contributed to the prediction of renal events in the follow-up period (hazard ratio [HR] 7.03 [95% CI 2.79–17.7], P < 0.001; HR 0.77 [95% CI 0.64–0.93], P = 0.007; HR 1.71 [95% CI 1.03–2.83], P = 0.037, respectively). Inclusion of age, sex, and treatment in the model did not affect the significance of ACR, hemoglobin, and serum creatinine on predicting renal events. Age also predicted renal events in the model (HR 0.96 [95% CI 0.93–1.00], P = 0.033), whereas sex and treatment had no significant effects.

On top of the adjusted model, GDF-15, NTproBNP, and hs-TnT all significantly predicted renal events in the model (HR 1.83 [95% CI 1.02–3.28], P = 0.041; HR 2.34 [95% CI 1.31–4.20], P = 0.004; HR 2.09 [95% CI 1.16–3.8], P = 0.014, respectively).

In the clinical reference model for CVC (Table 2) consisting of diastolic blood pressure, ACR, and total cholesterol, all factors significantly predicted CVC events in the follow-up period (HR 0.97 [95% CI 0.95–0.99], P = 0.003; HR 2.14 [95% CI 1.16–3.93], P = 0.014; HR 1.00 [95% CI 1.00–1.01], P = 0.021, respectively). Subsequent adjustment did not affect the significance of diastolic blood pressure, ACR, and total cholesterol and demonstrated no significant effects of age, sex, and treatment.

Addition of GDF-15 or hs-TnT to the adjusted model did not significantly predict CVC end points (HR 1.19 [95% CI 0.77–1.86], P = 0.432; HR 1.25 [95% CI 0.92–2.30], P = 0.101, respectively). In contrast, NTproBNP significantly predicted CVC events in the model (HR 3.45 [95% CI 2.07–5.76], P < 0.001).

As NTproBNP associated with both renal and cardiovascular risk, we aimed at demonstrating the relevance of NTproBNP on improving prediction of renal and cardiovascular events. For this, we calculated the incidence of renal complication and CVC events for patients who were above and below the cut point for NTproBNP. The incidence of renal events in patients above the cut point was 12.7%, a 2.6-fold increase over the 4.9% in the patients below the cut point. Likewise, the incidence of CVC events was similarly increased in patients above the cut point (23.4% above vs. 8.7% below the cut point). Above the NTproBNP cut point, 3.9% of patients experienced both a renal and a cardiovascular event, whereas this was 0.47% in patients below the cut point, an 8.3-fold increase.

To further characterize the predictive properties of GDF-15, NTproBNP, and hs-TnT, ROC curves were constructed and the C-statistics were calculated (Table 3). The clinical reference model for renal end point prediction, consisting of ACR, serum creatinine, and hemoglobin, already provided a fair accuracy with a C-statistic of 0.729 (95% CI 0.653–0.806). The inclusion of age, sex, and treatment in the model did not significantly increase the C-statistic. In addition, none of the individual biomarkers GDF-15, hs-TnT, and NTproBNP improved the C-statistic significantly. However, the combination of NTproBNP and hs-TnT improved the C-statistic significantly (C-statistic 0.793, P = 0.04), whereas other combinations did not significantly improve the C-statistic.

In the CVC model, the accuracy of the adjusted model was low (C-statistic 0.658 [95% CI 0.592–0.724]). The addition of NTproBNP significantly improved the C-statistic to 0.722 (95% CI 0.665–0.778, P = 0.018). None of the possible combinations of biomarkers further improved the C-statistic significantly.

To the best of our knowledge, this study is the first to report on the predictive properties of a panel of biomarkers of both renal and cardiovascular events in a cohort of T2D patients with nephropathy. Our post hoc analysis of patients previously enrolled in the Sun-MACRO trial included 861 patients with T2D. GDF-15, NTproBNP, and hs-TnT were independently associated with renal end points as demonstrated by Cox regression, whereas a combination of NTproBNP and hs-TnT improved the C-statistic. Furthermore, NTproBNP was independently associated with cardiovascular end points with a significant increase in the C-statistic.

Only a limited portion of the literature is available on the predictive properties of NTproBNP and hs-TnT in renal disease progression, and, to the best of our knowledge, none of these studies was performed exclusively in patients with diabetes who had impaired renal function. Previously, in a large cohort of elderly (mainly nondiabetic) patients in the Cardiovascular Health Study (25), both NTproBNP and hs-TnT were associated with rapid renal function decline and onset of CKD. Furthermore, BNP and NTproBNP were found to predict the progression of nondiabetic CKD in patients with mild to moderate nondiabetic kidney disease (26). Finally, a combination of hs-TnT and NTproBNP improved the prediction of microvascular events in the kidney but not in the retina in T2D patients with an eGFR of >60 mL/min/1.73 m² (7). Of importance is that these studies were performed in patients with varying levels of renal impairment, ranging from normal kidney function to patients with an eGFR <30 mL/min/1.73 m². As levels of NTproBNP are strongly increased in patients with reduced renal function (this study and Spanaus et al. [26]), the results of these previous studies could primarily indicate that the severity of renal disease at baseline predicts the risk of renal disease progression.

However, differences in renal function at baseline do not account for the association between biomarkers and renal risk in the current study, as all patients were characterized by impaired renal function. Furthermore, renal function was only marginally different among tertiles, and all three biomarkers showed just a weak correlation with eGFR. Moreover, we found that the association between NTproBNP/hs-TnT and renal risk was still significant even in statistical analyses that included baseline ACR and serum creatinine levels. Therefore, the current study indicates that these biomarkers are associated with renal disease progression irrespective of renal function at baseline. Nevertheless, the increased biomarker levels in patients with impaired renal function may hamper the setting of cut points for NTproBNP and hs-TnT in studies that do include patients with both normal and impaired renal function. In such studies, NTproBNP and hs-TnT may be used as continuous variables rather than as a binary parameter in combination with markers reflecting renal function at baseline. In addition to the association with renal events, we found that NTproBNP was associated with cardiovascular events. Although the predictive properties of NTproBNP for cardiovascular disease have been well established in patients with type 2 diabetes (27–30), the current study demonstrates that even in T2D patients with nephropathy and elevated NTproBNP levels a higher baseline NTproBNP is still associated with increased cardiovascular risk.

As NTproBNP was associated with both renal and cardiovascular risk, this biomarker may reflect a common pathophysiological mechanism in both renal and cardiovascular disease. Recent data suggest that increased levels of hs-TnT and NTproBNP reflect microvascular disease, contributing to the development of both cardiac pathology and diabetic nephropathy (7). If NTproBNP levels indeed reflect a common pathway for both renal and cardiovascular risk, the incidence of renal and cardiovascular events should be coupled and enriched in the subgroup of patients with NTproBNP levels above the cut point. In accord with this assumption, patients with high NTproBNP values had a higher incidence of experiencing both a renal and a cardiovascular event (8.3-fold) than expected by the increase in renal and cardiovascular events individually (6.8-fold). However, the limited number of patients experiencing both a renal and a cardiovascular event (n = 12) did not allow for a definite statistical analysis.

The predictive properties of GDF-15 for renal and cardiovascular events were rather modest. GDF-15 was associated with renal events only in the Cox regression analysis, but GDF-15 was not associated with cardiovascular events. Furthermore, GDF-15 did not improve C-statistics of both renal and cardiovascular risk. This is a surprising finding, as several studies have shown an association between GDF-15 and renal and cardiovascular risk. The Women’s Health Study (31), a prospective nested case-control study conducted in elderly patients without cardiovascular disease, showed that in aged women with eGFR <60 mL/min the highest concentration of GDF-15 increased the risk of combined cardiovascular death approximately threefold compared with women with the lowest concentrations. Moreover, an increased plasma level of GDF-15 predicted myocardial infarction in patients with type 1 diabetes independent of known cardiovascular biomarkers (32). Also, higher plasma levels of GDF-15 in patients were associated with an abnormal vascular response and a decline in renal function and predicted an increased risk of CV mortality in patients with early-stage diabetic renal disease independent of GFR changes (33). Further support for GDF-15 as a risk marker in T2D originates from our previous study (2) showing that plasma levels of GDF-15 predict the transition from microalbuminuria to macroalbuminuria. Taken together, our findings suggest a possible pathophysiological role for GDF-15 in the progression of diabetic kidney disease; however, the clinical relevance of GDF-15 as a biomarker for both renal and cardiovascular disease progression in a cohort of T2D patients with nephropathy could not be confirmed.

In addition to GDF-15, Cox regression and C-statistic analyses did not fully match for NTproBNP and hs-TnT, as both biomarkers were individually associated with renal risk in the Cox regression, whereas only their combination significantly improved the C-statistic for renal risk. Although these statistical analyses are distinct and have different objectives, a major difference between the two analyses is the inclusion of time to event in the Cox regression analysis. This may suggest that the investigated biomarkers primarily predict the rate of renal disease progression rather than the overall incidence of renal events in the follow-up period. Furthermore, we cannot exclude that the relatively short follow-up period of the current study may have contributed to the apparent difference in outcome between the Cox regression and C-statistic analyses, especially as the C-statistic is considered a relatively insensitive tool because a very strong independent association between a new risk marker and the outcome is required to result in a significant change in the C-statistic (34).

A further limitation may be that the current study was a post hoc analysis of patients previously enrolled in a negative clinical trial. Although secondary analyses of clinical trials may be liable to misinterpretations, the current study used the original cohort of patients for a biomarker study at baseline. Treatment was included in all of our statistical analyses, and we found no associations between treatment and renal or cardiovascular risk. Therefore, our findings are unlikely to be influenced by the intervention. Strengths of the current study include the large, homogeneous study cohort in which all patients were characterized by a well-defined phenotype of nephropathy and as a result a relatively high occurrence of renal and CVC events.

In conclusion, in the current study, GDF-15, hs-TnT, and NTproBNP were associated with renal risk whereas NTproBNP predicted cardiovascular risk independent of renal function. Although the current study did not improve renal risk prediction beyond currently established risk markers, our findings warrant a large prospective study to confirm that these biomarkers can indeed predict renal risk in T2D patients with nephropathy.

Acknowledgments. The authors thank Dr. Dietmar Zdunek and Ben Aalderink, Roche Diagnostics, for continuous support and providing the assays for the measurement of GDF-15, NTproBNP, and hs-TnT.

Funding. The Sun-MACRO trial was funded by Keryx Biopharmaceuticals.

Duality of Interest. H.J.H. is a consultant for and received honoraria from AbbVie, Astellas, AstraZeneca, Boehringer Ingelheim, Fresenius, Janssen, and Merck and has a policy of honoraria going to his employer. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. A.B. and S.P.H.L. conducted the analysis and wrote the manuscript. H.J.H., M.J.P., and H.B. reviewed and edited the manuscript. R.H.H. developed and formulated the research questions, conducted the analysis, wrote the manuscript, contributed to discussion, and reviewed and edited the manuscript. L.E.D. developed and formulated the research questions, contributed to the acquisition of data, contributed to discussion, and reviewed and edited the manuscript. A.B. and L.E.D. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.