Development of a biomarker panel to predict cardiac resynchronization therapy response: Results from the SMART-AV trial

Abstract

Background: Predicting a favorable cardiac resynchronization therapy (CRT) response holds great clinical importance.

Objective: The purpose of this study was to examine proteins from broad biological pathways and develop a prediction tool for response to CRT.

Methods: Plasma was collected from patients before CRT (SMART-AV [SmartDelay Determined AV Optimization: A Comparison to Other AV Delay Methods Used in Cardiac Resynchronization Therapy] trial). A CRT response was prespecified as a ≥15-mL reduction in left ventricular end-systolic volume at 6 months, which resulted in a binary CRT response (responders 52%, nonresponders 48%; n = 758).

Results: Candidate proteins (n = 74) were evaluated from the inflammatory, signaling, and structural domains, which yielded 12 candidate biomarkers, but only a subset of these demonstrated predictive value for CRT response: soluble suppressor of tumorgenicity-2, soluble tumor necrosis factor receptor-II, matrix metalloproteinase-2, and C-reactive protein. These biomarkers were used in a composite categorical scoring algorithm (Biomarker CRT Score), which identified patients with a high/low probability of a response to CRT (P <.001) when adjusted for a number of clinical covariates. For example, a Biomarker CRT Score of 0 yielded 5 times higher odds of a response to CRT compared to a Biomarker CRT Score of 4 (P <.001). The Biomarker CRT Score demonstrated additive predictive value when considered against a composite of clinical variables.

Conclusion: These unique findings demonstrate that developing a biomarker panel for predicting individual response to CRT is feasible and holds potential for point-of-care testing and integration into evaluation algorithms for patients presenting for CRT.

Trial registration: ClinicalTrials.gov NCT00677014.

Keywords: Biomarker; Cardiac resynchronization therapy response; Clinical algorithm; Heart failure; Multiplex assay; Prediction modeling; Scoring algorithm.

Copyright © 2018 Heart Rhythm Society. Published by Elsevier Inc. All rights reserved.

Figures

Figure 1.

(A) Sample sizes and distribution for biomarker measurements and primary/secondary end-points utilized in the present study from the SMART-AV trial. (B) Biomarker selection algorithm for predicting response to CRT whereby 12 candidate biomarkers were moved forward for evaluation.

Figure 2.

(A) Histograms for the 12 candidate biomarkers measured in the entire SMART-AV CRT sample set (n=758) with a vertical line representing the mean of the SMART-AV sample. The referent normal values (“Ref Norm”; age-matched control subjects (n=26)) vertical line has been superimposed as a frame of reference. (B) Distribution of the 12 candidate biomarkers as a function of 1st and 3rd quartile and the median value (box and whisker plots), dichotomized for the pre-specified response to CRT. NT-proBNP: 1pmol/L = 0.445 ng/mL

Figure 3.

(A) The final 4 biomarkers selected were used to develop a categorical Biomarker CRT Score. The distribution of the different permutations for elevated biomarker clusters is presented and was used in the composite score algorithm. (B) Distribution of response to CRT as a function of the Biomarker CRT Score, whereby the overall positive response to CRT was ~52%. A Biomarker CRT Score of 0 identified patients with a much higher likelihood of a favorable response to CRT and a score of 4 conferred a poor likelihood of response to CRT (Chi-Square analysis, p<0.001). As shown by the vertical dashed line, the Biomarker CRT Score identified a subset of patients with a very low probability for a response to CRT. The sample sizes for these Biomarker CRT Score quartiles are shown.

Figure 4.

(A) Distribution of Biomarker CRT Score as a function of changes in secondary response variables: LV end-diastolic volume (EDV), LV ejection fraction (EF), NYHA classification, Six Minute Walk test (6 MW), and Quality of Life (QOL) questionnaire. With a higher Biomarker CRT Score, LV functional indices and NYHA worsened at the end of the follow-up period in those patients with a high CRT response score vs those with a low Biomarker CRT Score (all p<0.001 as indicated). However, 6 MW and QOL score were not related to Biomarker CRT Score (p=0.965, p=0.174, respectively). (B) A Composite Freedom from Death and Heart Failure Hospitalization over the course of the observation period (6 months) was constructed as a function of Biomarker CRT Score. As the Biomarker CRT Score increased, the composite mortality/morbidity outcome worsened (log-rank test; p<0.001). The table shows the values computed at each time point and the relative risk as a function of Biomarker CRT Score. (C) Comparison of Odds of CRT Response between Biomarker CRT Scores, adjusted for confounding variables described in the text. A high Biomarker CRT Score resulted in low odds for a positive response to CRT. For example, a Biomarker CRT Score of 4 yielded a 5 times lower odds for a positive response to CRT compared to a Biomarker CRT Score of 0 (p<0.001).

Figure 5.

(A) Distribution of Biomarker CRT Score as a function of a composite MADITCRT Score. With higher MADIT-CRT Scores (>5), a subset of patients could still be identified with a high probability of a favorable response to CRT using the Biomarker CRT Score (bracket indicates p<0.05). Thus, despite classifying patients using a previously established CRT clinical scoring algorithm, a subset of patients could be identified over and above this clinical scoring algorithm. (B) Using an adaptive design and recursive partitioning (Supplemental Methods) demonstrated that the Biomarker CRT Score identified a subset of patients with a low likelihood of response to CRT, which would not have been realized by a clinical scoring algorithm alone.

Figure 6.

A proposed integration of the Biomarker CRT Score into a clinical HFrEF algorithm. The left sequence is the current standard of care recommendations and the right sequence identifies how the Biomarker CRT Score in combination with a clinical CRT score (MADIT-CRT) would identify unique cohorts of patients with a low and high likelihood of a favorable CRT response. In patients with a high Biomarker CRT Score, a low probability of a CRT response exists and alternative/advanced HF treatment may be appropriate. A low Biomarker CRT Score coupled with a high MADIT-CRT score would confer a high probability of a CRT response. A moderate Biomarker CRT Score coupled with a low MADIT-CRT would also suggest a low likelihood of a favorable CRT response. This schematic is derived from results shown in Figure 5B.