Baseline characteristics of participants

A total of 1993 individuals with type 2 diabetes were included in the analysis. The baseline characteristics are shown in Table 1. A total of 24.7% of participants had elevated NT-proBNP. When stratified according to NT-proBNP of <125 pg/ml or ≥125 pg/ml, participants with elevated NT-proBNP were older, were more often male, had a longer duration of diabetes, had a higher WC among female participants, had a higher WHR and had a more adverse lipid profile, with higher LDL-cholesterol, lower HDL-cholesterol, higher SBP, higher urine ACR and lower eGFR. They were also more likely to be receiving lipid-lowering medications, anti-hypertensive medications and oral glucose-lowering drugs, as well as insulin.

Table 1 Clinical characteristics for all participants and stratified by the cut-off of NT-proBNP in the HKDB study Association between NT-proBNP and prevalent diabetes cardiorenal complications

Comparing the prevalence of diabetes-related vascular complications at baseline, participants with elevated NT-proBNP had a frequency of complications at baseline 2–4 times higher compared with those with NT-proBNP<125 pg/ml (Table 1). Using logistic regression, adjusting for sex, age and diabetes duration, elevated NT-proBNP≥125 pg/ml was significantly associated with an increased risk of AF, CHD, CVD and CKD (Model 1 in Table 2). Analysis with NT-proBNP as a continuous trait yielded similar results (ESM Table 2).

Table 2 Association of binary NT-proBNP (≥125 vs <125) with prevalent and incident diabetes cardiorenal complications

In multivariate logistic regression adjusting for covariates at baseline, including sex, age, diabetes duration, smoking status, BMI, WC, SBP, DBP, HbA1c, TG, HDL-cholesterol, LDL-cholesterol and use of medications, elevated NT-proBNP≥125 pg/ml was associated with significantly increased risk of AF (OR 23.0 [95% CI 8.89, 59.4]), CHD (OR 2.35 [1.72, 3.20]), CVD (OR 2.38 [1.78, 3.16]) and CHF (OR 8.49 [4.80, 15.0]) (Model 2 in Table 2). All these associations remained significant after further adjustment for renal dysfunction with inclusion of baseline ACR and eGFR, some with numerically higher OR (Model 3 in Table 2). Elevated NT-proBNP≥125 pg/ml was also associated with increased risk of CKD (OR 3.94 [2.96, 5.24]) and KF (OR 10.7 [3.55, 32.2]) after adjustment for the baseline covariates in Model 2 (Table 2). Further adjustment for baseline ACR and eGFR diminished the association with renal outcomes (Model 3 in Table 2). The associations for all cardiorenal outcomes held true even when we corrected for multiple comparisons.

NT-proBNP and incident cardiovascular complications

As shown in Table 1, participants with NT-proBNP≥125 pg/ml experienced a greater number of incident cardiorenal complications, with 2–6 times higher frequency, over the median 5.1 years of follow-up compared with those with NT-proBNP<125 pg/ml. In analyses adjusting for sex, age and duration of diabetes, elevated NT-proBNP≥125 pg/ml was associated with incident AF (HR 4.64 [95% CI 2.44, 8.85]), CHD (HR 4.21 [2.46, 7.21]), CVD (HR 3.32 [2.20, 5.01]) and CHF (HR 4.18 [2.18, 8.03]) (Model 1 in Table 2). All these associations remained significant after further adjustments for smoking status, BMI, WC, SBP, DBP, HbA1c, TG, HDL-cholesterol, LDL-cholesterol, use of medications, ACR and eGFR at baseline (Model 3 in Table 2). The results remain significant after Bonferroni adjustment, except for CHF with the Model 3 adjustments.

NT-proBNP and incident renal complications

Elevated NT-proBNP≥125 pg/ml was associated with incident CKD (HR 2.51 [1.76, 3.57]), a 40% decrease in eGFR during follow-up (HR 3.13 [2.53, 3.87]) and KF (HR 5.99 [4.11, 8.72]). All these associations remained significant after further adjustments for baseline covariates in Model 2 (Table 2). Further adjustment for baseline ACR and eGFR rendered the association with incident CKD non-significant (HR 1.34 [0.89, 2.02]), although the associations with 40% decrease in eGFR (HR 1.47 [1.15, 1.86]), KF (HR 1.90 [1.20, 2.99]), or a composite renal endpoint of 40% drop in eGFR or KF (HR 1.45 [1.14, 1.84]), remained significant despite reduction in the strength of the association (Model 3 in Table 2).

Clinical utility of NT-proBNP for predicting cardiorenal endpoints in diabetes

The utility of NT-proBNP for predicting different cardiorenal endpoints in diabetes was next examined. Elevated NT-proBNP was found to have good discrimination ability to identify those at risk of future endpoints, with C index that ranged from 0.80 (95% CI 0.74, 0.86) for CVD, to 0.83 (0.76, 0.90) for CHD, to 0.88 (0.81, 0.94) for AF, and was highest for CHF at 0.89 (0.83, 0.95) (Fig. 1). When compared with established risk prediction equations for CHD, CHF and KF, incorporating NT-proBNP enhanced prediction of each complication beyond that provided by clinical risk factors alone, with significant incremental increase in C index (Table 3). Likewise, analysis using NRI indicated a significant contribution of adding NT-proBNP to reclassification (Table 4). IDI and/or relative IDI were significant for incident CHD and CHF (Table 4).

Fig. 1

C index for predicting incident diabetes cardiorenal complications using only the binary NT-proBNP values (≥125 vs <125). 'Composite renal endpoint' was defined as either 40% drop in eGFR or KF. The prediction model includes NT-proBNP only. The C index and SE were estimated based on the risk prediction models computed by Cox regression

Table 3 C index of NT-proBNP in predicting incident cardiorenal complications, over the established clinical risk scoresTable 4 NRI, IDI and relative IDI of NT-proBNP in predicting incident cardiorenal complications, over the JADE risk equations

Elevated NT-proBNP had C index of 0.79 (0.74, 0.85) for incident CKD, 0.81 (0.77, 0.84) for 40% drop in eGFR and 0.88 (0.84, 0.92) for KF (Fig. 1). Compared with an established risk equation for KF generated in a similar population [27], a model combining the risk equation and adding NT-proBNP achieved a C index of 0.94 (0.92, 0.96), with significant increase in C index compared with prediction of clinical risk score alone (Table 3). Analysis using NRI indicated improved prediction of KF after incorporating NT-proBNP, although there was no improvement using either IDI or relative IDI (Table 4).

The actual event rates for incident CHD, CHF and KF, stratified by the quintiles of JADE risk score and NT-proBNP, are shown in ESM Fig. 2 and ESM Table 3. Irrespective of the JADE risk score quintile, the event rate for each outcome was generally higher in participants with NT-proBNP≥125 pg/ml compared with those with NT-proBNP<125 pg/ml.

Sensitivity analyses

Analyses between NT-proBNP and incident complications were repeated using log NT-proBNP as a continuous trait, as well as comparisons of top quintiles of NT-proBNP vs other quintiles, which all yielded consistent and similar significant associations, with numerically higher HR for most complications in comparisons of top quintiles compared with other quintiles (ESM Table 2).

A different cut-off of NT-proBNP of ≥400 pg/ml, which has been adopted in some Asian countries such as Japan, was also examined. A total of 183 participants (9.2%) had NT-proBNP of ≥400 pg/ml. Analysis using this cut-off yielded very similar results to the primary analysis, with significant association with incident CHD, CVD, CHF, as well as CKD and KF, with mostly similar effect size compared with stratification using the cut-off of 125 pg/ml (ESM Table 2).

While it is recommended to compare biomarkers against risk equations most appropriate to the population being studied [3, 7], we have also explored the added benefit of incorporating NT-proBNP, compared with other risk prediction equations developed in non-Asian populations. We used the updated UKPDS risk engine for ischaemic heart disease outcome [29], as well as the RECODe risk equation for myocardial infarction [30]. These were developed in individuals with type 2 diabetes in the UK and USA, respectively. The results were similar to the analysis using the JADE risk equations, whereby incorporation of NT-proBNP improved the prediction of cardiovascular events compared with using the clinical risk factors alone (ESM Table 4).