Growth Differentiation Factor 15 is Associated with Lack of Ventricular Recovery and Mortality following Transcatheter Aortic Valve Replacement

Abstract

Background:

Recent data suggest that circulating biomarkers may predict outcome in patients undergoing Transcatheter Aortic Valve Replacement (TAVR). We examined the association between inflammatory, myocardial and renal biomarkers and their role in ventricular recovery and outcome following TAVR.

Methods and Results:

A total of 112 subjects undergoing TAVR were included in the prospective registry. Plasma levels of B-type natriuretic peptide (BNP), high sensitivity troponin I (hs-TnI), C-reactive protein (CRP), growth differentiation factor 15 (GDF-15), galectin-3 (GAL-3) and cystatin-C (Cys-C) were assessed prior to TAVR, and in 100 sex-matched healthy controls. Among echocardiographic parameters, we measured global longitudinal strain (GLS), indexed left ventricular mass (LVMI), and indexed left atrial volume (LAVI). The TAVR group included 59% male, with an average age of 84 years, and 1-year mortality of 18%. Among biomarkers, we found GDF-15 and CRP to be strongly associated with all-cause mortality (p<0.001). Inclusion of GDF-15 and CRP to STS score significantly improved C-index (0.65 to 0.79, p<0.05), and provided a category-free net reclassification improvement of 106% at 2 years (p=0.01). Among survivors, functional recovery in GLS (>15% improvement) and LVMI (> 20% decrease) at 1 year occurred in 48% and 22%, respectively. On multivariate logistic regression, lower baseline GDF-15 was associated with improved GLS at 1 year (HR=0.29, p<0.001). Furthermore, improvement in GLS at 1 month correlated with lower overall mortality (HR=0.45, p=0.03).

Conclusion:

Elevated GDF-15 correlates with lack of reverse remodeling and increased mortality following TAVR, and improves risk prediction of mortality when added to the STS score.

Introduction

With the aging population, severe calcific aortic stenosis (AS) is becoming more prevalent.¹ The emergence of transcatheter aortic valve replacement (TAVR) as an effective therapy for calcific AS has now greatly expanded the patient population able to receive this intervention.² The mortality associated with TAVR remains however, significant exceeding 20% at 1 year.³ This has led several groups to investigate whether imaging or circulating biomarkers can improve risk stratification.^4–8 Lindman et al. have found that growth-differentiating factor 15 (GDF-15), ST2, and N-terminal prohormone of brain natriuretic peptide (NT-proBNP) are incremental to STS score for predicting outcome following TAVR.⁶ Sinning et al. also showed that GDF-15 improved reclassification of patients when added to EUROSCORE.⁸ In recent years, other cardiac biomarker have also been associated with risk in heart failure but have not been extensively investigated in patients undergoing TAVR including galectin-3, an emerging marker of fibrosis, cystatin-C, a marker of renal function and C-reactive protein, an established marker of inflammation.^9–11 Among imaging markers, global longitudinal strain is a recently identified strong prognostic marker in patients with AS.^{12, 13}

In this prospective single center registry study, we sought to determine the relationship between a selected group of biomarkers and outcome as well as recovery of ventricular function in patients undergoing TAVR. The selected biomarkers encompass different key processes such as cardiac remodeling (left ventricular mass and left atrial size), ventricular function (global longitudinal strain), ventricular wall stress (BNP), myocardial injury (hs-troponin-I), inflammation (CRP), cardiac repair or oxidative stress (GDF-15), cardiac fibrosis (GAL-3) and renal function (Cystatin-C). The first objective of our study was to compare the level of biomarkers between patients undergoing TAVR and age and sex matched healthy controls. Our second objective was to identify independent physiological correlates for the different circulating biomarkers in patients undergoing TAVR. Finally, our third objective was to explore which biomarker or combination of biomarkers were more strongly associated with mortality and recovery of ventricular function in patients undergoing TAVR.

Methods

The data, analytic methods, and study materials will not be made available to other researchers for purposes of reproducing the results or replicating the procedure.

Study population and design

Adult patients with severe AS undergoing TAVR at Stanford University Medical Center were included. The protocol was approved by the Institutional Review Board of Stanford University, and written consent was obtained from all subjects. Please refer to Supplemental Methods for further details. In the setting of LV systolic dysfunction and low-flow, low-gradient AS, the severity of AS was confirmed by low-dose dobutamine stress echocardiography.¹⁴ Patients with prior valve replacement surgery, known active infection or cancer, on immunosuppressive therapy, or end-stage renal disease on dialysis were excluded. Healthy sex-matched controls were recruited from a previously described population based cohort for comparison of baseline biomarker profiles.¹⁵ The healthy controls were not followed for clinical outcomes in this study.

Biomarkers Analysis

At baseline, we measured B-type natriuretic peptide (BNP), high sensitivity troponin I (hs-TnI), high sensitivity C-reactive protein (CRP), and galectin-3 (GAL-3) using Abbott architect instrument (ci4100). Roche Biochemistry analyzer was used to measure Cystatin C; ELISA assay (R&D) was used for GDF-15. All samples analyzed were collected in the fasting state and frozen at −80ºC at Stanford Human Immune Monitoring Core Laboratory. All biomarker analyses were performed at Abbott Core Laboratory, which was blinded to the clinical characteristics of the study.

Echocardiographic Assessment

Echocardiography was performed using commercially available echocardiographic systems (Sonos 7500, iE33, and EPIQ 7C; Philips Medical Imaging, Eindhoven, the Netherlands), and analyzed by Stanford Biomarker and Phenotypic Core Laboratory following standard protocols.¹⁴ Further details are included in the Supplemental Methods. We chose 15% change in GLS as meaningful change according to the recent recommendation of the ASE and task force of strain imaging.¹⁶ We chose a higher value of 20% for changes in LV mass and LA volume recognizing the greater variability in serial imaging of wall thickness and cardiac dimensions.¹⁷

Statistical Analysis

Please refer to the Supplemental Methods for further details. We normalized GDF-15 and BNP values using logarithmic base 10 transformation for analysis. One-way ANOVA or Kruskal-Wallis test was used for comparison of biomarkers among three healthy groups (young, middle, old) and the TAVR group, and post-hoc analysis was performed with Turkey-HSD multiple comparison tests or Games-Howell, as appropriate. Repeated ANOVA was used to compare echocardiographic data of LVMI, GLS, and LAVI at the three time points (baseline, 1-month and 1-year).

Univariate and multivariate analyses were performed to determine the associates of each biomarker using covariates as age, sex, BMI and hemodynamic and echocardiographic parameters. Parameters with p<0.10 were selected for creating a multivariate logistic regression model using an indicator for improvement in LV function as the binary outcome separately for GLS, LVMI and LAVI. Similarly, univariate and multivariate Cox-regression tests were performed for outcome analysis. The log-rank test was performed to compare survival and shown in Kaplan-Meier plots. We also performed a partial correlation analysis to assess the correlation between each pair of biomarkers while adjusting for the other measured biomarkers.

survival, survC1, and survIDINRI packages in R were used for survival analysis and calculation of C-index, integrated discrimination improvement (IDI) and net reclassification improvement (NRI) based on Uno et al.¹⁸ The IDI is independent of category and considers separately the actual change in calculated risk for each individual. NRI illustrates the percentage of a population that is accurately reclassified, providing clinically meaningful results rather than only statistically significant differences between χ2 and c-statistic values.

Results

Baseline Characteristics

One hundred and twelve consecutive patients with severe AS who underwent TAVR and met inclusion criteria were enrolled in this study. One hundred subjects were recruited to the healthy aging cohort. Further, we divided the healthy aging group to younger (< 60 years, n= 34), middle (60–75 years, n= 32) and an older age group (age > 75 years, n= 32).

The mean age of the TAVR cohort was 84 years and 59% were men (Table 1). The older healthy aging cohort was matched to the TAVR cohort for age and sex. Table 1 and 2 summarize the clinical and echocardiographic characteristics of enrolled patients. Transfemoral, transaortic, and transapical approaches were used in 92 (82%), 15 (13%), and 5 (5%) patients, respectively. Baseline echocardiographic examination was performed in all patients and repeated in 77 patients at 1-year after TAVR. Twenty patients (18%) died within the first year. Median follow up was 679 days (Q1-Q3 346–964 days) following TAVR.

Table 1.

Baseline clinical characteristics

	Healthy Subjects			TAVR
	Younger (n=34)	Middle (n=34)	Older (n=32)	Patients at baseline (n=112)
Age (years)	50.7±11.4	66.1±3.2	82.8±5.4	84.0±8.5
Male sex, n (%)	17 (50)	17 (50)	15 (47)	66 (59)
BSA (m2)	1.84±0.23	1.78±0.20	1.76±0.20	1.85±0.23
BMI (kg/m2)	25.1±3.3	23.4±3.1	24.9±4.5	26.6±5.0
STS-score	-	-	-	7.9±4.2
Frailty ≥ II, n (%)	-	-	-	67 (60)
NYHA functional class ≥ III, n (%)	-	-	-	59 (53)
Comorbidity
Hypertension, n (%)	3 (9)	1 (3)	16 (50)	94 (84)*
Dyslipidemia, n (%)	5 (15)	13 (38)	13 (41)	78 (70)*
Diabetes mellitus, n (%)	0	0	2 (6)	36 (32)*
History of coronary artery disease*, n (%)	-	-	-	62 (55)
CKD grade ≥ III, n (%)	-	-	-	88 (79)

Includes history of myocardial infarction in 28 patients (23%), bypass surgery in 38 patients (31%), and percutaneous coronary intervention in 36 patients (30%).

Table 2.

Echocardiographic parameters at baseline and 1-year follow-up

		Subgroup with 1 year Follow-up
	Baseline (n=112)	Baseline (n=77)	1-year (n=77)	P value*
Heart rate (bpm)	73 ± 13	73 ± 14	72 ± 13	0.92
Systolic blood pressure (mmHg)	125 ± 17	125 ± 17	135 ± 18	<0.001
Diastolic blood pressure (mmHg)	69 ±12	70 ± 12	67 ± 11	0.16
Left ventricular internal diameter (cm)	4.7± 0.9	4.7 ± 0.9	4.8 ± 0.8	0.22
Interventricular septum (cm)	1.2 ± 0.2	1.2 ± 0.2	1.1 ± 0.2	<0.001
Posterior wall (cm)	1.2 ± 0.2	1.2 ± 0.2	1.1 ± 0.2	<0.001
LV mass index (g/m2)	120 ± 46	116 ± 44	102 ± 38	<0.001
LVEF (%)	55 ± 13	55 ± 13	60 ± 9	<0.001
GLS (%)	−13.0 ± 3.2	−13.1 ± 3.3	−15.1 ± 2.6	<0.001
e’ (cm/s)	5.2 ±1.4	5.3 ± 1.6	5.7 ± 1.8	0.01
a’ (cm/s)	6.2 ± 2.6	6.5 ± 2.8	6.6 ± 2.4	0.85
E/e’	25.0 ± 11.0	24.2 ± 11.4	20.7 ± 10.2	0.004
LAVI (ml/m2)	48.2 ± 14.8	48.2 ± 15.9	46.6 ± 17.6	0.22
AS severity
Aortic valve area (cm2)	0.62 ± 0.19	0.62 ± 0.17	1.38 ± 0.43	<0.001
Aortic valve area index (cm2/m2)	0.34 ± 0.09	0.34 ± 0.08	0.75 ± 0.23	<0.001
Mean trans-aortic pressure gradient (mmHg)	50.3 ± 15.7	51.2 ± 14.2	10.6 ± 5.4	<0.001

^*p values are between echocardiograms at Baseline (n=76) and 1-year (n=76).

EF, ejection fraction; GLS, global longitudinal strain; LAVI, left atrial volume index; LV, left ventricular.

Reverse remodeling and functional recovery after TAVR

Echocardiographic parameters at baseline and 1-year after TAVR are shown in Table 2. Taken as a cohort, the mean LV function parameters such as LVMI, GLS improved at 1-year, while LAVI did not change significantly. Among patients who completed 1-year follow-up after TAVR, LVMI and GLS changed significantly (117±44 vs.100±36 g/m², p<0.001 for LVMI and −13.1±3.3 vs. −15.1±2.6%, p<0.001 for GLS). Supplemental Figure 1 shows the change in (A) LVMI, (B) GLS, and (C) LAVI at 1-month and 1-year after TAVR in patients. 23% of patients presented with improved LVMI (decreased ≥ relative change of 20%) and 77% patients remained stable (−20% < relative change < 20%) at 1-year, while 47% presented with increased GLS (increased in absolute value ≥ relative change of 15%) and 53% remained stable (−15% < relative change < 15%).

Biomarker profiles at baseline

Compared to age and sex matched controls, patients with TAVR had significantly higher levels of six biomarkers (Figure 1). Among biomarkers, there was an age-related progressive increase in the biomarker levels, more prominent for GDF-15, BNP and Cys-C (Figure 1). In our healthy cohort, CRP did not increase in the different aging groups.

Figure 1. Distribution of biomarker profiles in healthy aging cohorts and patients with TAVR.

There is significant differences in baseline biomarker profiles with age. One-way ANOVA was performed to compare (A) Younger (Age<60; mean 50.7 yrs), (B) Middle (60≤Age<75; mean 66.1 years), (C) Older (75≤Age; mean 82.8 years) healthy cohorts, and (D) the TAVR cohort (mean 83.7 years) (Table 1).

*, p<0.05 vs. Younger; #, p<0.05 vs. Middle; +, p<0.05 vs. Older. Box-whisker plot description

The correlations between baseline characteristics, echo parameters, and biomarkers are represented in Figure 2A (Supplemental Table 1). The different biomarkers in the TAVR group were not independent of each other. BNP and GDF-15 correlated with each other (r=0.53, p<0.0001), and GDF-15 also correlated with GAL-3 and CRP (r=0.36, p=0.005; r=0.26, p<0.0001, respectively). A partial correlation analysis showed GDF-15 to be correlated to GAL-3, CRP, BNP, and Cys-C, when adjusted for other biomarkers, highlighting the connectivity of GDF-15 among common biomarkers (Figure 2B).

Figure 2. Correlation matrix of baseline cardiac function and biomarkers and Partial correlation diagram of biomarkers.

(A) A correlation matrix was created based on dissimilarity index using R ggplot (color gradient is based on 1- r value). There was significant correlation between baseline ventricular function and BNP. (List of r and p values of correlations is in Supplemental Table 1). (B) Partial correlation network analysis. Lines connecting variables represent statistically significant relationships (p < 0.05). Partial correlation analysis showed GDF-15 to significantly correlated with GAL-3, CRP, BNP and Cys-C when adjusted for the other biomarkers within the figure. A full line represents a positive relationship while a dashed line represents an inverse relationship. Thicker lines indicate stronger correlation as measured by r value. BNP - brain natriuretic peptide, hsTnI – highly sensitive troponin I, GDF-15 - growth differentiation factor 15, CRP - C-reactive protein, GAL-3 - galectin-3, Cys-C – cystatin-C.

Supplemental Table 2 summarizes the different physiological correlates of biomarker levels. LVMI was associated with BNP, GDF-15 and hs-TnI; GLS was associated with BNP. No strong ventricular physiologic correlates were found for CRP, GAL-3 or Cys-C.

BNP associated significantly with GLS, and LVMI (r=0.46 and r=0.37, respectively. Both p<0.0001, Supplemental Table 1). LVMI was associated with both hs-TnI (r=0.26, p<0.01) and GDF-15 (r=0.23, p=0.01). On multivariate analysis, age, male sex, systolic wall stress, and LVMI were each independently associated with GDF-15 levels (R²=0.2, Supplemental Table 2). No echocardiographic parameters were significantly associated with GAL-3 on multivariate analysis.

Association of biomarkers with all cause mortality

Of the biomarkers assayed, GDF-15 and CRP emerged as significantly associated with all cause mortality on both univariate multivariate analyses (Table 3). Furthermore, survival analysis based on previously defined cut-offs for GDF-15 (2000 pg/mL) and CRP (2mg/L), ^{19, 20} showed significantly worse survivals in subgroups with elevated biomarker levels (Figure 3; GDF-15, log-rank p<0.0001; CRP, log-rank p=0.01). Although not found to be independently correlated in multivariate analysis, BNP and hsTnI also showed significant difference in survival based on pre-defined cutoff values of 250 pg/mL and 6 ng/L, respectively (Figure 3).^{21, 22} GAL3 and Cys-C trended towards significance (p values 0.06, 0.07 respectively). However, when utilizing median values for the biomarkers, Cys-C also showed significant difference in mortality (Supplemental Figure 2).

Table 3.

Association with all-cause mortality (Cox-regression Analysis)

Variable	HR	95% CI	p
Association of mortality with the baseline parameters/biomarkers
Univariate
Age	2.29	1.24–4.24	0.02
Male sex	1.56	0.99–2.46	0.05
STS score	1.54	1.17–2.01	<0.01
CRP	1.45	1.17–1.82	0.001
Log BNP	1.84	1.23–2.75	0.003
Log GDF-15	2.36	1.61–3.44	<0.001
Cys-C	1.30	0.93–1.82	0.12
Multivariate
Age	2.10	1.07–4.08	0.03
Log GDF-15	2.03	1.40–3.00	<0.001
CRP	1.28	1.00–1.63	0.05
Association of mortality with the change in cardiac function at 1 month
Univariate
Relative change in GLS	0.51	0.27–0.96	0.04
Relative change in LVMI	0.74	0.47–1.17	0.20
Relative change in LAVI	1.24	0.77–1.96	0.39
Multivariate
Age	1.71	0.92–3.18	0.09
Relative change in GLS	0.45	0.24–0.86	0.02
Male sex	1.83	1.05–3.20	0.03

Univariate analysis was performed using variables as age, male sex, BMI, STS score, frailty, 6 biomarkers, baseline LVMI, absolute GLS, and LAVI. Only variables with p<0.15 are shown in the table and were included in the multivariate analysis.

For the 1 month echocardiographic analysis, age, sex, baseline GLS, and the change in GLS at 1 month (p<0.15) were included in multivariate analysis. Values (in HR, 95%CI) were standardized by SD.

Figure 3. Biomarkers predict worse 2-year mortality after TAVR.

Kaplan-Meier plots show significantly worse survival in groups with higher GDF-15, BNP, CRP and Cys-C. Groups were divided based on pre-defined cutoff values for each biomarker. Log-rank test was performed for p-values.

When stratified by STS scores (intermediate risk (<8) and high risk (≥8)), the elevation in GDF-15 and CRP was associated with significant increase in mortality at 1 year in both subgroups (Figure 4A). The C-statistics for STS score alone in risk prediction of mortality improved significantly when GDF-15 and CRP were added to the risk model (0.65 to 0.79, p<0.05). The integrated discrimination improvement (IDI) and category-free net reclassification improvement (NRI) indices were also calculated for the censored survival data at 2 years using models with and without inclusion of CRP and GDF-15 to STS score. There was a significant improvement in both indices – IDI 0.19 (95% CI 0.09–0.29, p<0.001), and category-free NRI(>0) 1.06 (95% CI 0.64–1.44, p<0.001). Overall, 79% of the event group, and 26% of the non-event group were reclassified with the addition of biomarkers. In contrast, inclusion of frailty to STS score did not result in significant improvement in prediction for risk of mortality (IDI 0.01, p=0.219; NRI 0.258, p=0.169). However, when frailty was used with GDF-15, they performed similarly compared to the model including GDF-15 and CRP (Figure 4). We also calculated the STS-TAVR score for each subject, which is a recently developed score to predict in-hospital mortality based on the STS/TVT Registry (http://tools.acc.org/tavrrisk).²³ We found that GDF-15 and CRP significantly improved the C-statistics of STS-TAVR score in predicting 1 year survival (0.43 to 0.79, p<0.0001, Supplemental Figure 3). Analysis of censored data using the log-rank method in these subgroups also showed that elevated biomarkers correlated with worse outcome in both the intermediate and high STS score subgroups, while frailty was not predictive of worse outcome in the intermediate STS score subgroup (Supplemental Figure 4).

Figure 4. Addition of GDF-15 and CRP to STS score results in integrated discrimination improvement and net reclassification improvement.

(A) Elevated biomarkers (GDF-15>2000, CRP>2) is associated with increased all-cause mortality in both intermediate-risk and high-risk groups by STS score at 1 year. High GDF-15 (left), and elevated GDF-15 and CRP (right) are associated with significant increase in all-cause mortality. (B) Presence of frailty was associated with increased mortality at 1 year in the high risk group, however, no correlation was found in the intermediate risk group. Frailty did not result in significant IDI and NRI. When combined with GDF-15, frailty is associated with increased mortality. (C) The empirical distribution function of the paired difference between the risk scores (on the probability scale) estimated at time = 2 years using models with and without the inclusion of CRP and GDF-15 are shown. The added value of variables (i.e. IDI) is proportional to the shaded area. The vertical difference at s = 0 (between the two black dots; where s scales the graphic between −1 and +1) is ½NRI(>0), and the horizontal difference (between the two gray dots) equals the median risk-score difference. y-axis, cumulative probability; x-axis, s = difference between two model risk scores.

Association of reverse remodeling with all cause mortality

Baseline echocardiographic parameters were not associated with all cause mortality. However, improvement in GLS at 1 month associated with better outcome in a multivariate regression analysis (Table 3, Chi-square p=0.01).

Association between baseline biomarkers and functional recovery at 1 year

We found biomarkers and baseline cardiac function to independently correlate with improvement in GLS, LVMI and LAVI at 1 year (Table 4). Specifically, age, baseline GLS, logGDF-15 and Cys-C together predicted improvement in GLS with AUC 0.90 (p<0.0001), and baseline LVMI and GAL-3 predicted improvement in LVMI (AUC 0.79, p<0.0001) on multivariate analysis. Lower level of GDF-15 was the only significant predictor of improvement in LAVI (AUC 0.75, p=0.003).

Table 4.

Biomarkers predict change in LA and LV indices at 1 year.

	LVMI improvement	GLS improvement	LAVI Reverse remodeling
AUC, p value	0.79, <0.0001	0.90, <0.0001	0.75, 0.003
Factors, OR [95%CI]	Baseline LVMI, 4.09 [2.18–8.77] GAL-3, 1.44 [1.00–2.31]	Baseline GLS, 0.08 [0.02–0.25] Log GDF-15, 0.29 [0.09–0.85] Cystatin C, 2.16 [0.92–5.10] Age, 3.26 [1.31–7.50]	Log GDF-15, 0.37 [0.18–0.78]

Variables were retained in the model if p<0.1.

GLS, longitudinal strain; LAVI, left atrial volume index; LVMI, left ventricular mass index; OR, odds ratio Covariate included age, sex, STS score, baseline of considered parameter, and the 6 biomarkers (BNP, GDF-15, CRP, hs-TnI, GAL-3, and cystatin C). The values of OR and 95%CI were standardized by SD.

Discussion

The main finding of our study is that GDF-15 is a centrally connected biomarker in patients undergoing TAVR and is associated with all-cause mortality as well as lack of recovery in ventricular function. While baseline echocardiographic parameters were not as strongly related to outcome as circulating biomarkers, lack of improvement at 1 month in GLS emerged as a stronger correlate of outcome.

GDF-15 is a stress-induced cytokine of the transforming growth factor-β(TGF-β) superfamily, that increases during tissue injury and inflammatory states. It is highly expressed in cardiomyocytes, macrophages, endothelial cells and vascular smooth muscle cells.^{24, 25} We found GDF-15 to be the most connected biomarker among the different biomarkers assayed as well as with ventricular function, potentially highlighting its central role in the ventricular remodeling process. The central role of GDF-15 is also highlighted by its independent correlation with survival as well as reverse ventricular remodeling.

Previous studies have found an age-specific increase in BNP, and TnI in healthy subjects ^26–28 Here, we report that GDF-15 levels also increase with healthy aging, in addition to other cardiac biomarkers. When compared to the healthy aging cohort, all biomarkers were significantly elevated in the TAVR patients. Furthermore, as presented in our correlation matrix and partial correlation analyses, we found that the circulating and imaging biomarkers assayed were not independent of each other. BNP was closely related with baseline GLS and LVMI while GDF-15 and hs-TnI were correlated with LVMI. Few studies have analyzed these correlations in patients with pressure-overloaded ventricles. Kuznetzova et al. found a relationship between LVH and hs-TnI in the general population.²⁹ Lancellotti et al. found ST-2, a marker of cardiac stretch to correlate with LAVI and AVA in patients with AS.³⁰

Consistent with previous studies, we have also observed an overall improvement in LV function 1 month following TAVR, and this improvement extended out to 1 year from TAVR (Supplemental Figure 1). The response was however variable with some patients showing greater improvement than others. Furthermore, we found baseline biomarkers to have predictive capacity for cardiac recovery. Baseline ventricular function and GDF-15 were strong independent predictors of reverse cardiac remodeling following TAVR (AUC>0.80). Despite strong correlation of BNP with LV functional recovery following TAVR on univariate analysis, this association was no longer significant when the baseline function was included in the multivariate model. Among baseline parameters, GDF-15 was the only parameter associated with all-cause mortality as well as functional recovery at 1 year in survivors. GDF-15 also significantly improved the risk model when added to STS score. Although frailty has been associated with worse 1 year outcome in TAVR patients in a prior study,³¹ we found that frailty alone did not perform as well as the biomarkers and did not significantly improve net reclassification when added to STS in our cohort. The difference in the risk classification was especially pronounced in subjects with STS<8 in our cohort. Given the rapidly increasing number of patients with low to intermediate STS scores that are receiving TAVR, the clinical relevance of these findings is significant.

Among echocardiographic parameters, we found lack of improvement in GLS at 1 month to be predictive of worse outcome. Despite the association of baseline GLS and LVMI with cardiac recovery, these echo parameters did not predict all-cause mortality in multivariable models. These findings are consistent with recent reports showing biomarkers and early ventricular recovery following TAVR to predict improved survival.^{6, 7,8, 32}

To the best of our knowledge, this is the first study to investigate the association between selected cardiovascular and inflammatory biomarker panels and reverse remodeling in patients undergoing TAVR. We also present a unique comparison of biomarker profiles in healthy cohort with TAVR group. The strong predictive capacity of GDF-15 and CRP together for adverse outcome reiterates the role of inflammation in aortic stenosis induced ventricular remodeling. In terms of clinical implications, this study suggests that the current risk prediction model for patients undergoing TAVR leaves significant margin for improvement and that it would be possible to improve risk stratification by inclusion of limited number of biomarkers.

There are limitations to this study. First, the study is small in size and without a validation cohort, and hence the findings from this study are exploratory and need further studies for validation. However, our findings do validate recent studies highlighting the role of GDF-15. Future studies would include whether a combined clinical and biomarker risk score can prospectively improve the net reclassification of patient risk groups and outcomes. Measuring the dynamic changes in biomarker following TAVR may also answer the question of whether the TAVR procedure is restorative of the inflammatory process that occurs in the ventricle exposed to the large afterloads.

Conclusion

We found circulating biomarkers to have superior prognostic properties than baseline echocardiographic parameters for all-cause mortality in patients undergoing TAVR. Higher level of GDF-15 predicted both poor ventricular recovery and worse survival following TAVR, and improved risk prediction when added to STS score. Overall, our data suggest that an inflammatory process is important in predicting ventricular recovery and outcome in patients receiving TAVR. These markers should be validated in larger cohorts.