The median age of the patients was 67 [60-73] years, and 80% of them were male. The median QRS duration was 163 [141-184] msec, and 82% of the patients had left bundle branch block (left bundle branch block) morphology in the electrocardiogram (Table 1). After 6 months of CRT, the left ventricular ejection fraction increased significantly 28% (23-33) vs 37% (31-41); P<.0001, while the left ventricular end-diastolic volume and left ventricular end-systolic volume decreased (left ventricular end-diastolic volume: 303 [251-351] mL vs 259 [203-332] mL; P<.0001; left ventricular end-systolic volume: 303 [154-276] vs 167 [116-242] mL; P<.0001).

We investigated the change of the studied biomarkers during the 6 months of the follow-up period. We measured statistically significant lower plasma levels of pentraxin-3 (4.92 [3.29-6.84] ng/mL vs 3.13 [2.39-4.48] ng/mL; P<.0001), NT-proBNP (2612 [1377-5124] pg/mL vs 1626 [725-3300] pg/mL; P<.001), HGF (1379 [1029-1863] pg/mL vs 1083 [862-1328] pg/mL; P<.001) and carbohydrate antigen-125 (22.75 [11.05-69.90] U/L vs 18.30 [9.15-34.35]; P < .001) after 6 months of follow-up. Plasma levels of fractalkine and matrix metalloproteinase-9 remained statistically unaltered.

A total of 70 patients (51%) were considered as responders, and 66 patients (49%) were nonresponders according to our criteria (> 15% relative decrease in left ventricular end-systolic volume). N-terminal pro-B-type natriuretic peptide levels were significantly higher in nonresponders both at baseline (P=.0002) and at 6 months (P < .0001). We observed a significant decrease in plasma NT-proBNP levels only in responders (P=.0001). Pentraxin-3 and carbohydrate antigen-125 levels were considerably higher in nonresponders at baseline (pentraxin-3: P=.0005; carbohydrate antigen-125: P=.001) and 6 months (pentraxin-3: P=.02; carbohydrate antigen-125: P=.002). After CRT, plasma levels decreased both in responders (pentraxin-3: P=.0001; carbohydrate antigen-125: P=.01) and nonresponders (pentraxin-3: P < .0001; carbohydrate antigen-125: P=.02). Patients lacking a reverse remodelling response to CRT had significantly higher baseline HGF (P=.001) and fractalkine (P=.004) levels. Hepatocyte growth factor levels were significantly reduced following CRT in both groups (responders, P=.003; nonresponders, P=.01), while plasma fractalkine levels did not change over time. We found no significant differences in matrix metalloproteinase-9 levels between the groups, although we observed a significant rise in responders after 6 months (P=.003). The aforementioned results are detailed in Table 2.

We investigated 5-year all-cause mortality by univariate Cox regression and 6-month reverse remodelling using univariate logistic regression analysis (Table 1). During the 5 years of follow-up, 58 patients (43%) died. All-cause mortality was significantly related to the absence of left bundle branch block (P < .0001), or beta-blocker therapy (P=.003) and increasing NT-proBNP levels (P < .0001), while increasing age showed a marginally significant association (P=.07). Of the novel biomarkers, increasing levels of fractalkine (P=.01), HGF (P < .0001) and matrix metalloproteinase-9 (P=.04) predicted 5-year mortality.

A total of 70 patients (51%) experienced reverse remodelling at 6 months based on our echocardiographic criterion. Increasing age (P=.004), NYHA functional class III-IV (P=.01) and increasing levels of NT-proBNP (P=.004) were predictive of the absence of 6-month reverse remodelling. A marginally significant relationship was seen between ischemic etiology of HF (P=.069), mineralocorticoid receptor inhibitor therapy (P=.08) and a lack of response. Increasing plasma levels of fractalkine (P=.04), pentraxin-3 (P=.01) and HGF (P=.01) predicted the lack of response to CRT.

To further assess the impact of novel biomarkers on mortality, we set up a basic multivariable Cox regression model with all the variables previously shown to be relevant by the univariate analysis. The baseline multivariable model included age, left bundle branch block, beta-blocker therapy and NT-proBNP. In the next step, we entered the novel biomarkers separately into the baseline model in a forward stepwise manner. Only HGF predicted mortality in the multivariable model (HR, 1.35; 95%CI, 1.11-1.64; P=.003, per 1 standard deviation increase) (Table 3).

We applied the same approach to investigate reverse remodelling. We included all the relevant factors to the basic multivariable model: age, ischemic etiology of HF, NYHA functional class III-IV, mineralocorticoid receptor inhibitor therapy and NT-proBNP. We entered the novel biomarkers separately into the baseline model in a forward stepwise manner. Similarly to the mortality prediction, HGF was the only independent predictor of reverse remodelling (OR, 1.83; 95%CI, 1.10-3.04; P=.01, per 1 standard deviation increase) (Table 3).

The area under the curves were assessed by the receiver operating characteristics analysis to set up the best fitting cutoff values. Our aim was to determine a clinically relevant cut-off point that had relatively high sensitivity and the utmost specificity.

In this way, we found plasma HGF levels above 1236 pg/mL to be an optimal cutoff value (area under the curve=0.71; 95%CI, 0.63-0.80; P < .0001; sensitivity: 81% [69-90]; specificity: 54% [45-65]) (Figure 1). Hepatocyte growth factor levels exceeding 1236 pg/mL were associated with an increased risk of long-term mortality (HR, 3.48; 95%CI, 1.80-6.73; P < .0001) (Figure 1).

Figure 1.

Impact of baseline HGF levels on 5-year mortality. A: comparison of the Kaplan-Meier survival curves by log-rank test. The survival of patients with baseline plasma HGF levels below and above 1236 pg/mL was compared. B: receiver operating characteristics analysis was performed to determine the optimal cutoff point for baseline plasma HGF levels. AUC, area under the curve; HGF, hepatocyte growth factor; HR, hazard ratio; NPV, negative predictive value; PPV, positive predictive value.

(0.11MB).

When HGF was added to the baseline multivariable model of mortality (NRI=0.69; 95%CI, 0.39-0.99; P < .0001; IDI=0.06; 95%CI, 0.02-0.11) and reverse remodelling (NRI=0.39; 95%CI, 0.07-0.71; P=.01; IDI=0.03; 95%CI, 0.00-0.06) respectively, we observed a significant improvement in reclassification and discrimination (Figure 2). Adding the HGF biomarker to the baseline model increased the discrimination slope of the prediction models (from 0.12 to 0.15 for reverse remodelling; from 0.16 to 0.23 for mortality). The predicted and observed risks, assessed by the Hosmer-Lemeshow test, did not differ significantly throughout the analyses, which indicated good calibration and confirmed the validity of the results (Table 4). The decrease in Brier score and increase in Nagelkerke's R2 suggested an improved predictive capacity.

Figure 2.

Improved discrimination capacity of the prediction models. The improved discrimination capacity is visualized by means of a box and whiskers diagram. The discrimination slope is defined as the difference in the mean probabilities of events minus nonevents (upper panel, 6-month reverse remodelling; lower panel, 5-year mortality). Adding the HGF biomarker to the basic model increases the discrimination slope (red line) of the prediction models. The difference in the discrimination slopes is the IDI itself. HGF, hepatocyte growth factor; IDI, integrated discrimination improvement.

(0.26MB).

Chronic HF is a complex clinical syndrome affected by various pathophysiological processes. The exact role of inflammation, fibrosis, and neurohormonal activation in the development and progression of HF is under extensive research; nevertheless, all of them are under consideration as integrant components of the pathomechanism. As a consequence, numerous biomarkers have been studied, which reflects the activation of these processes and may enable diagnosis, therapy guidance, and prognosis assessment.14

Despite all the advances in the complex treatment of HF seen in the previous decades, morbidity and mortality are still high, and they exert a significant burden on health care worldwide.15 To address these problems, refinements in patient selection have been required, but approximately one-third of patients still do not respond to CRT.2 Thus, biomarkers predicting response or mortality may therefore be helpful in identifying high-risk patients in need of special care.

In our study, we investigated a set of novel HF biomarkers associated with different pathways of HF progression. These biomarkers have been shown to correlate with prognosis and HF severity in non-CRT populations.4–7,16 Levels of fractalkine, a mediator of inflammation, are elevated in HF, and this is independently associated with increased mortality.7 Pentraxin-3 is an important component of humoral innate immunity and has been associated with clinical outcomes and disease severity in HF patients with reduced and preserved ejection fraction.6 The tumor marker carbohydrate antigen-125 is used for screening, diagnosis and monitoring the treatment of ovarian cancer, and elevated levels were found in various types of malignancies, ascites, pleural effusion, and other types of peripheral edema. Furthermore, it reflects the hemodynamic status of both sides of the heart and correlates with fluid overload and clinical outcomes in HF patients.4,17 Matrix metalloproteinase-9 plays a leading role in myocardial remodelling by degrading extracellular matrix proteins. Along with other parameters, it was found to have predictive value in ischemic and dilated cardiomyopathy.5,8

Hepatocyte growth factor, a classic growth factor produced by a wide range of cell types, exerts angiogenetic, antifibrotic and antiapoptotic effects3; therefore, it could be considered a multipotent cardioprotective agent. This protective role was justified in a wide range of animal models using different methods of gene transfer.3,18 Elevated levels of HGF were found in acute and chronic HF patients, and surprisingly, this was a predictor of worse clinical outcomes.16,19 Taking these findings into consideration, highly elevated HGF levels may reflect the failing of the protective pathways of the cardiovascular system, therefore identifying patients in the most unstable clinical condition and at the highest risk.

A considerable favorable effect of CRT on outcomes was observed during follow-up, but clinical outcomes were slightly worse than those reported in the literature. This finding can be explained by the fact that in the Hungarian population overall, cardiovascular mortality is higher and life expectancy is significantly shorter than in high-income countries.20

In this study, pentraxin-3 predicted reverse remodelling and matrix metalloproteinase-9 predicted long-term mortality, while fractalkine was a predictor of both endpoints in univariate models. Despite these promising results, all those previous biomarkers lost their predictive values in the multivariate models. Hepatocyte growth factor was the only marker that was able to predict both the absence of reverse remodelling and 5-year mortality independently of other clinical covariates.

Furthermore, we analyzed the predictive value of HGF by reclassification analyses, which showed statistically significant reclassification improvement and discrimination development in the case of both endpoints, demonstrating that measuring HGF may be beneficial in the risk assessment of HF patients undergoing CRT. This result parallels the findings of Richter et al.,21 who found HGF to be the strongest predictor of cardiovascular mortality in advanced HF patients among several novel HF biomarkers. A possible explanation could be that HGF is a growth factor that acts directly on the level of genes and is triggered by nonspecific cardiovascular distress. Hence, HGF reveals a more comprehensive picture of the complex HF pathophysiology than the other biomarkers, which instead reflect a particular pathway.

Based on our results, HGF may be helpful in identifying CRT patients with the highest risk of poor clinical outcomes. Patients with inadequate response to CRT need special care and early consideration of heart transplantation or mechanical circulatory support. Due to the relevant predictive value of HGF, it may be useful for the timely identification of potential nonresponders or it may be used for guiding further device therapy in the absence of clinical response to CRT.

Notable limitations are the single-center design and the relatively small sample size. We considered all-cause mortality as the primary endpoint; cardiovascular death was not investigated separately due to the relatively low event numbers. Routine laboratory measurement of HGF is not widespread to date, which might limit the direct adaptation of our findings into real-world clinical practice. Left ventricular end-systolic volumes were measured using the Teicholz method, and not with the Simpson method, which would have been more accurate for calculating LV volumes in this particular patient cohort. Taking all these factors into consideration, our results should be interpreted as preliminary hypothesis-generating findings. Their validation is clearly necessary with larger, prospective studies.