Study population
A total of 2,186 T2DM patients were enrolled in the study, 496 patients were excluded from the analysis as survival status (n=460) or diabetes duration (n=36) was not available, thus a total of 1,690 T2DM patients were analyzed. Detailed description of the baseline characteristics is displayed in Table 1.
Table 1. Baseline characteristics of the overall study cohort.
Characteristics
|
Overall cohort (n=1690)
|
Demographics
|
|
Age, years (IQR)
|
63 [54-69]
|
Female, n (%)
|
783 (46)
|
Diabetes duration, years (IQR)
|
10 [5-19]
|
Hypertension, n (%)
|
1135 (67)
|
Dyslipidaemia, n (%)
|
1152 (68)
|
Smoking, n (%)
|
339 (20)
|
BMI, kg/m² (IQR)
|
28.7 [25.4-32.7]
|
Cardiovascular disease
|
|
PCI, n (%)
|
66 (4)
|
PAD, n (%)
|
173 (10)
|
CeVD, n (%)
|
99 (6)
|
CABG, n (%)
|
62 (4)
|
Medications
|
|
Statins, n (%)
|
764 (45)
|
Acetylsalicylic acid, n (%)
|
634 (38)
|
Insulin, n (%)
|
888 (53)
|
Oral antidiabetics, n (%)
|
984 (58)
|
Laboratory parameters
|
|
NT-proBNP, pg/ml (IQR)
|
122 [59-266]
|
Albumin/creatinine ratio, mg/mmol (IQR)
|
0.87 [0.35-2.94]
|
eGFR, ml/min (IQR)
|
72.7 [60.3-85.3]
|
LDL cholesterol, mg/dl (IQR)
|
102 [82-123]
|
HbA1c, % (IQR)
|
7.2 [6.5-8.1]
|
BMI, body mass index; PCI, percutaneous coronary intervention; PAD. peripheral artery disease; CeVD, cerebrovascular disease; CABG, coronary artery bypass graft; NT-proBNP, N-terminal pro-B-type natriuretic peptide; eGFR, estimated glomerular filtration rate; LDL, low-density lipoprotein
Median age of the total study population was 63 years (IQR 54-69), 783 (46%) of the patients were female. CVD was present in 311 (18.4%) patients. According to the ESC/EASD risk model criteria, 25 (1.5%) were classified as moderate, 252 (14.9%) as high and 1125 (66.6%) as very high risk. A total of 288 patients (17.0%) were not classifiable based on the stated ESC/EASD criteria, as 280 patients had diabetes duration less than 10 years with 1 or 2 (but < 3) established CV risk factors and 8 patients presented with a diabetes duration longer than 10 years without any risk factors. Detailed characteristics for the ESC/EASD risk strata are presented in Additional file 1: Table S3. In the overall cohort, 654 patients (39%) had a calculated SCORE risk estimate below 5%, 525 patients (31%) between 5 and 10% and 511 patients (30%) above 10%. The calculated SCORE risk estimates for 10-year fatal CVD increased with ESC/EASD risk category (0% [IQR 0-0] vs 6% [IQR 2-10] vs 8% [IQR 4-12], p<0.001 for the moderate, high and very-high risk category).
Distribution of patients in the ESC/EASD and SCORE risk strata as well as proportion of patients with normal (n=871) and elevated (n=819) NT-proBNP levels at a cut-off 125 pg/ml within these groups are illustrated in Figure 1 (p<0.001 for both models).
Association of NT-proBNP, ESC/EASD risk strata and SCORE with CV and all-cause death at 10 years
During 10 years of follow-up, 448 (26.5%) patients died, CV death accounted to 44.9% (n=201) of all deaths. Kaplan Meier analysis for all risk models stratified into three groups with regards to both endpoints are shown in Figure 2. The ESC/EASD risk model, SCORE and NT-proBNP were all significantly associated with CV death (ESC/EASD: p=0.013; SCORE and NT-proBNP: p<0.001) and all-cause death (ESC/EASD: p=0.022; SCORE and NT-proBNP: p<0.001). Table 2 shows the results of the univariate Cox regression analysis.
Table 2. Association of NT-proBNP, the ESC/EASD and SCORE risk model with outcome in unselected patients with T2DM (n=1690).
|
10-y Cardiovascular death
|
|
10-y All-cause death
|
|
HR [95% CI]
|
P
|
C-index [95% CI]
|
|
HR [95%CI]
|
P
|
C-index [95% CI]
|
ESC/EASD risk model
|
1.65 [1.11 to 2.45]
|
0.013
|
0.53 [0.50 to 0.56]
|
|
1.33 [1.04 to 1.70]
|
0.022
|
0.52 [0.50 to 0.54]
|
SCORE, %
|
1.05 [1.04 to 1.07]
|
<0.001
|
0.64 [0.60 to 0.67]
|
|
1.06 [1.04 to 1.07]
|
<0.001
|
0.66 [0.63 to 0.68]
|
NT-proBNP, pg/ml*
|
5.72 [4.68 to 7.00]
|
<0.001
|
0.80 [0.77 to 0.83]
|
|
3.32 [2.90 to 3.79]
|
<0.001
|
0.73 [0.70 to 0.76]
|
NT-proBNP, tertiles
|
4.11 [3.23 to 5.22]
|
<0.001
|
0.75 [0.73 to 0.78]
|
|
2.33 [2.05 to 2.65]
|
<0.001
|
0.70 [0.67 to 0.72]
|
NT-proBNP, > 125 pg/ml
|
7.15 [4.85 to 10.53]
|
<0.001
|
0.71 [0.68 to 0.74]
|
|
3.14 [2.56 to 3.86]
|
<0.001
|
0.66 [0.63 to 0.68]
|
*Hazard ratio (HR) refers to ln-transformed NT-proBNP per 1-IQR increase
NT-proBNP remained a strong predictor of risk independently from traditional confounders as age, eGFR, sex, hypertension, smoking and albuminuria for both CV death (ln[NT-proBNP]: adjusted HR: 3.92 [2.92-5.27], p<0.001) and all-cause death (ln[NT-proBNP]: adjusted HR: 2.29 [1.89-2.78], p<0.001). Patients with NT-proBNP >125pg/ml had a 7.2-fold and 3.1-fold risk to suffer from death from CV causes or all-causes at 10 years, respectively, compared to individuals with NT-proBNP within the range considered as normal (p<0.001).
Discriminatory performance of NT-proBNP, the ESC/EASD risk strata and SCORE in in unselected T2DM patients
ROC curves for all risk models and the endpoints 10-year CV and all-cause death are shown in Figure 3. The best discriminatory power to predict 10-year CV and all-cause death was observed for NT-proBNP indicated by highest C-indices compared to both the ESC/EASD and SCORE risk model (C-index: CV death: 0.80 vs 0.53 vs 0.64; all-cause death: 0.73 vs 0.52 vs 0.66, p<0.001 for all comparisons).
When NT-proBNP was entered as a categorical variable based on tertiles (1st tertile: 59 [IQR 59 - 59], 2nd tertile: 122 [IQR 90 - 156], 3rd tertile: 376 [IQR 267 - 648] pg/mL) the results remained virtually unchanged (C-index: CV death: 0.75 vs 0.53 vs 0.64, p<0.001 for all comparisons; all-cause death: 0.70 vs 0.52 vs 0.66, p<0.001 for comparison NT-proBNP vs ESC/EASD, p=0.014 for NT-proBNP vs SCORE).
Net individual risk prediction was significantly improved when assessing NT-proBNP confirmed by a significant improvement of the NRI with 81% for CV death and 58% for all-cause death (P<0.001 for both) compared with the ESC/EASD risk model. Comparable results were observed for NT-proBNP compared with the SCORE model with a significant improvement of the NRI with 83% for CV death and 63% for all-cause death (P<0.001 for both).
Predictive performance of NT-proBNP, the ESC/EASD risk strata and SCORE in distinct subgroups
The following subgroups were investigated: Subgroup A including patients without CVD and subgroup B including patients without CVD and aged 40-64 years according to SCORE derivation cohort.
Cox regression analysis for the subgroups regarding 10-year fatal outcome are shown in Additional file 1: Table S4. NT-proBNP and SCORE were equally associated with 10-year CV and all-cause death in subgroup A and B (p<0.001 for all). The ESC/EASD risk model was only associated with 10-year CV death in subgroup B (p=0.050).
Figure 4 displays C-statistics of the risk assessments according to the T2DM population studied, the respective ROC graphics are shown in Additional file 1: Figure S1. NT-proBNP was characterized by robustly highest C-indices across both subgroups comparable to the unselected cohort for both endpoints. The ESC/EASD model was characterized by poor C-indices in all groups. The performance of the SCORE risk prediction model improved with progressing exclusivity of patient criteria, performing best in the cohort closest to its derivation population, i.e. subgroup B. NT-proBNP outperformed SCORE with regards to CV death in subgroup A and was statistically comparable whereas with numerically higher C-indices in subgroup B.
Observed vs predicted risk estimate for 10-year fatal CVD by SCORE
The SCORE risk algorithm underestimated the actual risk of CV death in unselected T2DM patients (observed vs predicted CVD fatal risk: 13% vs 8% [IQR 4-12]; X2 =70.1, p <0.001) and in subgroup A (observed vs predicted CVD fatal risk: 10% vs 8% [IQR 4-12]; X2=19.8, p <0.001). In subgroup B observed risk was 6% compared with a median predicted risk of 4% [IQR 2-8]. Here, goodness-of-fit for SCORE risk estimate was good with a X2 of 4.3 (p = 0.234).
Predictive performance of NT-proBNP, the ESC/EASD risk strata and SCORE for 5-year CV and all-cause hospitalization
Over a follow-up of 5 years, 1053 (62.3%) patients were hospitalized due to any causes and 367 (21.7%) patients due to unplanned CV events. Risk for all-cause hospitalization increased by 7% and for CV hospitalization by 12% per 100pg/ml increase in NT-proBNP (p<0.001 for both). Similarly, the ESC/EASD and SCORE were associated with increased risk for CV and all-cause hospitalizations (p<0.001 for both). Cox regression analysis is presented in Additional file 1: Table S5.
Figure 5 presents cumulative incidence of 5-year hospitalizations for NT-proBNP. In terms of discriminatory accuracy NT-proBNP was superior to the ESC/EASD risk model for both outcomes (C-index: CV hospitalization: 0.74 vs 0.54; all-cause hospitalization: 0.62 vs 0.55; p<0.001 for all comparisons). When comparing NT-proBNP with SCORE, NT-proBNP showed superior discrimination for CV hospitalization (C-index: 0.74 vs 0.62, p<0.001) but not for all-cause hospitalization (C-index: 0.62 vs 0.59, p=0.09).
Outcome specifity of NT-proBNP, the ESC/EASD risk strata and SCORE
As indicated by C-statistics, NT-proBNP yielded better discrimination for CV than for all-cause death (0.80 vs 0.73, p<0.05) and for CV than for all-cause hospitalization (0.74 vs 0.62, p<0.05). No difference for the ESC/EASD risk strata (CV vs all-cause death: 0.53 vs 0.52, p>0.05; CV vs all-cause hospitalization: 0.54 vs 0.55, p>0.05) or the SCORE risk estimation (CV vs all-cause death: 0.64 vs 0.66, p>0.05; CV vs all-cause hospitalization: 0.62 vs 0.59, p>0.05) could be observed.