Study Sample

The study sample consisted of 202 consecutive patients with New York Heart Association class III to IV heart failure despite medical therapy, LVEF ≤35%, and LBBB (QRS duration [QRSd] ≥120 ms; delayed intrinsicoid deflection in leads I, V₅, V₆ ≥50 ms; broad slurred R waves in I, aVL,V₅, and V₆; rS or QS waves in V₁ through V₂; ST-T wave vectors opposite the major QRS vector). Patients with right bundle branch block (RBBB), right ventricular (RV) pacing, or myocardial infarction ≤3 months previously were excluded. Simultaneous biventricular pacing was applied without exception for the first 6 months.

Echocardiography and ECG

Standard supine 12-lead ECGs (filter range, 0.15 to 100 Hz; AC filter, 60 Hz, 25 mm/s, 10 mm/mV) were obtained at baseline and before discharge. Echocardiograms were performed at baseline and after 6 months of CRT. Patients were imaged in the left lateral decubitus position with a commercially available system (Vingmed System Seven, General Electric–Vingmed, Milwaukee, Wis) with a 3.5-MHz transducer at a depth of 16 cm in conventional parasternal and apical views. Standard 2-dimensional and color Doppler data triggered to the QRS complex were saved in cine-loop format. LV end-systolic volume (ESV) and end-diastolic volume (EDV) were measured at the apical 2- and 4-chamber views; LVEF was calculated with the use of the biplane Simpson method. Interobserver and intraobserver reproducibility for LV volume measures were 90% and 96%, respectively.³

ECG Analysis of Ventricular Activation

ECGs were analyzed blinded to echocardiographic results. All measurements were made with the use of digital calipers at 200% magnification calibrated for paper speed 25 mm/s. Normal frontal plane axis was +90°↔<−30°, left axis deviation (LADEV) ≥−30°↔<−60°, right axis deviation (RADEV) ≥−90°↔>+90°. Right superior axis (RSA) was >180°↔−90°; right inferior axis (RIA) +90°↔180°. “Incomplete” right superior axis (IRSA) was extreme LADEV (−60°↔−90°) with rS or QS complexes in the inferior leads and equiphasic QR in I (see below for QRS hieroglyphic framework). Axis quadrant shift was displacement of LBBB frontal plane axis by ≥1 quadrant during CRT (eg, normal→LADEV, normal→RSA).

QRS Hieroglyphic Framework for Ventricular Activation Pattern Comparisons

The QRS complex in each lead before and after CRT was deconstructed into 4 possible waveform elements (R, S, Q, QS). Absolute amplitudes (mV) and durations (ms) of all elements of each QRS complex were used to characterize specific activation patterns (eg, R and S present, R>S amplitude, pattern=Rs). Ventricular activation in each lead was characterized by 9 possible patterns or QRS hieroglyphs, as follows: (1) R; (2) RS (R=S and both >1 mm, or both <1 mm; equiphasic); (3) Rs (R >s); (4) rS (S >r); (5) QS; (6) qR (q <R); (7) QR (Q=R>1 mm, or both <1 mm; equiphasic); (8) Qr (Q >r); and (9) QRS (all 3 waveforms present).

Characterization of Ventricular Activation During LBBB

Typical LBBB activation is registered as right (R)→left (L) (frontal plane), anterior (A)→posterior (P) (horizontal plane), and variable axis. This yields a QRS hieroglyphic signature with dominant positive forces in I, aVL (R, Rs); negative forces in aVR (QS); variable forces in II, III, AVF (R, Rs, rS, QS); dominant negative forces in V₁ through V₂ (QS, rS); transition V₃ through V₅ (rS→Rs, R); and dominant positive forces in V₅ through V₆ (R, Rs).

QRS Notching

LBBB is characterized by sequential ventricular activation (RV→LV)^1,4,5 and registered as fragmented QRS complexes with RSR′ configuration (“notch”). QRS notching was defined as ≥1 notch in the R or S wave present in ≥2 adjacent anterior, lateral, or inferior leads.⁶

RV and LV Activation Time

Because multiple notches in the R and S wave during LBBB may occur because of myocardial scar,⁶ the first notch was assumed to indicate the transition between RV and LV depolarization. Notching in the first 40 ms of the S wave in V₁ through V₂ was excluded because this indicates scar in the QRS score. RV activation time (RVAT) was measured as time (ms) between QRS onset and first notch in any of ≥2 adjacent leads. LV activation time (LVAT) was QRSd−RVAT (ms).

For modeling purposes, the longest LVAT (LVAT_max) recorded in any lead and region was used. A numeric relationship between QRSd and LVAT_max was derived with the use of linear regression (LVAT_max [ms]=−35.839+0.763×QRSd [ms]+ 0.000619×QRSd [ms] ∧2), permitting estimation of LVAT_max in the absence of QRS notching (40 patients).

Quantification of LV Scar

Myocardial scar has been linked to reduced CRT response.³ ECG quantification of LV scar volume was calculated with the use of the Selvester QRS score for LBBB.^7,8 The effects of scar on LBBB surface registration translate as specific QRS hieroglyphic signatures, manifest as unopposed rightward electric forces by infarct region: qR, QR, rS in I, aVL (anterior-superior); qR, QR, rS in I, V₅ through V₆ (apical); qR, QR, rS in II, aVF (inferior); QS→rS, RS, or Rs in V₁ through V₂ (anterior-septal). Notching on the ascending limb of the S wave in V₁ through V₂ indicates posterolateral infarct.

Evidence for Ventricular Activation Fusion During Biventricular Pacing

Experimental models of LBBB demonstrate maximum improvement in LV pump function occurs when intra-LV electric asynchrony is minimized by wavefront fusion.¹ Wavefront opposition and reversal during biventricular pacing should yield ECG evidence of fusion, as follows: (1) expected change in frontal plane electric axis: normal or LADEV→RADEV; (2) expected changes in QRS hieroglyphic signatures: rightward forces emerge in leads with dominant leftward forces (R in I, aVL→qR, QR, QS); anterior forces emerge in leads with dominant posterior forces (eg, QS in V₁→rS, RS, Rs, R; QS or rS in V₂→RS, Rs, R; rS or RS V₃→Rs or R); (3) an alternate way of characterizing evidence for ventricular fusion is using regional or global measures of CHange in R wave amplitude in the expected direction in Lateral (I, aVL) and Precordial leads (V1-V6) (CHRLP) before and after CRT.

Although LV pacing yields specific ventricular activation patterns generally in accordance with stimulation site, nearly identical ECG activation sequences may be registered at widely spaced sites,⁹ and electric resynchronization during CRT correlates unpredictably with stimulation site because of conduction blocks.⁵ For these reasons, LV lead position was not relevant to this analysis.

End Points

The primary end point was LV reverse volumetric remodeling, defined as ≥10% reduction in ESV at 6 months.² Supplemental reverse remodeling end points, chosen to provide additional insights into the predictive model, were ≥10% reduction in EDV (mechanistically similar measure of reverse remodeling) and ≥30% reduction in ESV (“superresponse”²).

Statistical Analysis

For descriptive summaries, categorical variables are presented as n (%) and continuous variables as median (25th, 75th percentile).

Indices of change in R amplitude were generated by calculating, for each patient, the change in R amplitude for each lead from baseline to post-CRT, as a proportion of the baseline value. The postimplantation value was used in cases in which an R was absent at baseline and present postimplantation; a value of 1 was used in cases in which an R was present at baseline and absent postimplantation; and a value of 0 was used in cases in which both were absent. Change values were set to positive if the change was in the expected direction (decrease in I, aVL, and V₄ through V₆; increase in V₁ through V₃) and were negative otherwise. Leads II, III, and aVF had no expected direction and retained their calculated sign. These change values were then averaged across relevant leads for each patient. Four summaries were created: (1) mean of I and aVL; (2) mean of V₁ and V₂; (3) mean of all leads except aVR; and (4) mean of I, aVL, and V₁ through V₆ (CHRLP).

Candidate predictors for multivariable modeling were selected with the use of the aforementioned framework, including variables that characterize LBBB (LVAT, LV scar volume, QRS hieroglyphs) and expected evidence of ventricular fusion (axis change, change in R amplitudes, and changes in QRS hieroglyphic signatures). In cases in which several related measures were available (eg, regional LVAT versus LVAT_max), univariate tests were used to determine which measure had the strongest relationship with the end point. Continuous variables were checked for linearity of their relationship with the outcome, and revised versions (usually truncations) were created where necessary; categorical variables related to each other were checked to determine whether a combination of the variables would be a better predictor than the individual variables.

After this screening process, 13 variables were selected as candidates for the primary end point logistic regression model (6 baseline, 7 post-CRT). Baseline variables were as follows: QRSd, LVAT_max, QRS score, QS hieroglyph in aVR, R hieroglyph in V₆, and R >5 mV in V₅ or V₆. Post-CRT variables were as follows: LADEV→RADEV, R→Qr, qR, QR, or QS in I and aVL, new R waves >5 mV in both V₁ and V₂, and mean change in R amplitude in the expected direction in I and aVL, in V₁ and V₂, and in all leads except aVR. For the 2 supplemental models, the same set of predictors was used, with minor modifications.

For each end point, all candidate predictors were entered into a logistic regression model, and stepwise selection was then used to identify the set of significant independent predictors. Relative risk is expressed as odds ratio (OR) (95% confidence interval [CI]). Predicted probabilities of the primary end point were generated from the final logistic regression model. In cases in which model variables are not specifically displayed in a plot, median or mode values were used in generating predictions (QRS points=6, change in R amplitude=4.5, LADEV→RADEV=no). For characterizing tiers of response, patients were divided into low, middle, and high responders, with predicted probability of response=0% to 49%, 50% to 74%, and 75% to 100%, respectively, and patterns among their predictor values were examined.

QRS Hieroglyphs by Pivotal Leads: Baseline and After CRT

The preliminary screening process for candidate predictors of ≥10% reduction in ESV indicated that the expected changes in local and regional QRS hieroglyphic signatures were most pronounced in I, aVL, and V₁ and V₂, designated pivotal leads. During LBBB, a dominant R→L, A→P activation pattern was observed in ≥95% of patients (Table 2). Less than 5% of patients had L→R and/or P→A activation because of dominant Q or S waves in I and aVL and R waves in V₁ through V₂, caused by scar.

During CRT, rightward forces emerged in I and aVL and anterior forces in leads V₁ through V₂. These effects were greatest in I and V₁ and were observed in 74% and 53% of patients, respectively. Reciprocally, evidence of persistent leftward and posterior activation was recorded in 26% (I) and 47% (V₁).

Evidence for Ventricular Activation Fusion During CRT

A rightward axis emerged in 67% of patients. However, a RSA was observed in only 58% and RIA in 9%. In contrast, 20% had LADEV or normal axis during CRT, and 12% had IRSA. Therefore, nearly one third of patients did not have evidence of ventricular fusion by frontal plane axis during CRT. Similarly, although ≈90% had an axis quadrant shift, a RSA or RIA was achieved in only three fourths of these patients. The remaining one fourth had primarily a leftward (LADEV or IRSA) axis quadrant shift (Table 3).

Evidence for ventricular fusion with the use of QRS hieroglyphs was recorded in all 4 pivotal leads, with greatest frequency in I and V₁. Evidence of wavefront collision (Q and S emergence, R regression) was observed in I for 71% of patients. However, total (QS) or near-total (Qr) reversal of activation in I was observed in only 55%. Similarly, total reversal of activation in V₁ (dominant R emergence) was observed in only 50%. Evidence for activation reversal was ≈50% less evident in aVL and V₂.

A ≥50% reduction in R-wave amplitude in I and aVL, reflecting L↔R wavefront fusion in the horizontal plane, was observed in 60% of patients. Reciprocally, a ≥50% increase in R-wave amplitude in V₁ and V₂, due to A↔P wavefront fusion, was observed in only 62%.

Predictors of ≥10% Reduction in ESV at 6 Months

A multivariable model identified 2 baseline and 2 post-CRT independent predictors of ≥10% reduction in ESV at 6 months (Table 4). Increasing LVAT_max was associated with a greater probability of ≥10% reduction in ESV (OR [CI]=1.30 [1.11, 1.52] per 10-ms increase) up to 125 ms; for longer LVAT_max, there was no further increase in probability of response (Figure 1). Patients with values in the lowest quartile (≤80 ms) had a 51% response rate compared with 73% response in patients with values of ≥125 ms. Although QRSd was weakly associated with reverse remodeling probability in preliminary univariate comparisons, this relationship was replaced by LVAT_max in the multivariable model.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1. Probability of reverse remodeling by baseline and post-CRT predictors. Top left, Increasing probability of response with greater LVAT_max. Line represents predicted probability of response for each value of LVAT_max. Diamonds represent actual response rates among LVAT_max quartiles. Top right, Increasing probability of response with greater LVAT_max and post-CRT evidence of ventricular fusion (P→A activation reversal). Bottom left, Increasing probability of response with greater LVAT_max and post-CRT evidence of ventricular fusion (L→R activation reversal). Bottom right, Increases in QRS score reduce response probability for any value of LVAT_max.

Increasing QRS scores were negatively associated with reverse remodeling (OR [CI]=0.49 [0.27, 0.88] for each 1-point increase from 0 to 4; 0.92 [0.83, 1.01] for each 1-point increase >4). Patients with QRS scores in the lowest quartile (0 to 3) had a response rate of 78% compared with patients with scores in the highest quartile (>9), whose response rate was 45%.

After CRT, increasing R amplitudes in V₁ through V₂, indicating ventricular fusion, were associated with increased probability of reverse remodeling. This effect was not observed until the mean change in R amplitude was ≥4.5 times the baseline value (OR [CI]=2.76 [1.01, 7.51] for each 1× increase ≥4.5×). Although only 25 patients fell in this upper range, the response rate was 84% versus 60% in patients with mean change V₁ through V₂ R <4.5 times baseline. A second measure of ventricular fusion, LADEV→RADEV, was associated with increased probability of reverse remodeling (OR [CI]=2.00 [0.99, 4.02]). Patients showing this axis quadrant shift had a response rate of 70% compared with 60% in patients without.

Visual evidence for the effect of LVAT_max and the other independent predictors of reverse remodeling is provided in Figure 1. Predicted probability of reverse remodeling is greatest for longer baseline LVAT_max in the presence of post-CRT ventricular fusion, indicated by larger changes in R amplitude in leads V₁ through V₂ (top right) and LADEV→RADEV axis quadrant shift (lower left). In contrast, the positive effects of increasing LVAT_max are offset by higher QRS scores such that longer LVAT_max in the presence of a high QRS score has lower predicted probability of reverse remodeling than very short LVAT_max in the presence of a low QRS score (lower right).

Predicting Reverse Remodeling Using a Global Measure of Ventricular Fusion

As for most of the R-wave amplitude directional change variables, there was no change in probability of reverse remodeling for the lowest values (Figure 2). Probability of reverse remodeling did not start to increase with larger CHRLP until post-CRT values were greater than baseline (OR [CI]=3.14 [0.97, 10.15] for each 1× increased >1; P=0.056), whereas all patients with CHRLP <1 had equivalent probability of response, regardless of the actual CHRLP value.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2. Predicting reverse remodeling using a global measure of ventricular fusion. Greater changes in R-wave amplitudes after CRT, indicative of wavefront fusion, predict higher probability of response. The points along the line are individual predicted values.

Tiers of Reverse Remodeling Response

Response tiers are shown in Table 5. The multivariable predictive model showed good discrimination (c-index 0.74). Low responders had ≥1 adverse value for baseline predictors (ie, QRS score >17, LVAT_max <60 ms), or offsetting combinations (ie, QRS score 3 to 5+LVAT_max ≤90 ms, or LVAT_max >110 ms+QRS score ≥12), or midrange values (ie, LVAT_max 90 to 110 ms, QRS score 6 to 11). Very short LVAT_max (48 to 62 ms) occurred only in this group, with no QRS scores 0 to 2. These patients also had limited or absent fusion evidence after CRT (Figure 3).

• Download figure
• Open in new tab
• Download powerpoint

Figure 3. Top, High responder, baseline: LVAT_max=142.7 ms, QRS score=1. Determination of RVAT_max with the use of QRS notching, present in 2 regions, is shown at the bottom. Time between QRS onset (first arrow) and notch nadir (second arrow) is RVAT. Shortest RVAT_max=45.3 ms. LVAT_max=QRSd (188 ms)−45.3 ms=142.7 ms. Post-CRT: LADEV→RADEV=yes (L→R activation reversal, indicated by R→QS in I, R→qR in aVL), change in R in V₁ through V₂=4.8 (P→A activation reversal, indicated by QS→R in V₁, QS→rS in V₂). Predicted probability of response=0.9. Actual change in ESV=−24%. Bottom, Low responder, baseline: LVAT_max=48.2 ms, QRS score=14, by anatomic region: anterosuperior (ASUP), anteroseptal (AS), apical (AP), inferior (INF) and posterolateral (PL) walls. Post-CRT: LADEV→RADEV=yes (L→R activation reversal, indicated by R→QS in I, qR→QS in aVL), change in R in V₁ through V₂=0 (incomplete P→A activation reversal, indicated by QS→rS in V₁, rS→QS in V₂). Predicted probability of response=0.21. Actual change in ESV=+20%. Pt indicates QRS score point.

High responders had ≥1 favorable predictor (ie, QRS score <3 or change in R amplitude >5) or a combination of favorable predictors (ie, QRS score 3 and LVAT_max >110 ms, or QRS score 4 to 9 and LVAT_max >110 ms and quadrant shift). Alternately, high responders had fair values for 1 predictor that were offset by very favorable values for others (ie, smaller change in R amplitude but LVAT_max >130 ms). There were no QRS scores >12, whereas all changes in R amplitude >5 occurred in the high responders (Figure 3).

Middle-range responders had intermediate values for predictors (ie, LVAT_max 90 to 110 ms, QRS score 6 to 11, quadrant shift, midrange change in R amplitude) or an offsetting combination of very favorable and midrange predictor values (ie, long LVAT_max but high QRS score and small change in R amplitude).

Predictors of Supplemental Reverse Remodeling End Points

Multivariable models for supplemental reverse remodeling end points yielded predictors similar to those in the ≥10% ESV reduction model (Table 6). The ≥30% ESV reduction end point was met by 59 patients (29.2%). Each 10-ms increase in LVAT_max was associated with increased probability of ≥30% ESV reduction, slightly less versus the ≥10% ESV model (OR=1.23 versus 1.30). The relationship of QRS score, for its entire range, to a ≥30% reduction in ESV was similar to the relationship of QRS score >4 to a ≥10% reduction in ESV (OR=0.91 versus 0.92 for each 1-point increase). Ventricular fusion (V₂ R amplitude change from <1 to >5 mm) also predicted ≥30% ESV reduction (OR=2.90). Axis quadrant shift was not significant, possibly because of the lower event rate.

The ≥10% EDV reduction end point was met by 45.5% of patients (92/202). Greater LVAT_max was directly associated with EDV reduction (OR=1.92 for each 10-ms increase up to 95), whereas higher QRS scores were inversely related (OR=0.87 for each 1-point increase). Ventricular fusion (V₁ R amplitude change from <1 to >5 mm) also predicted EDV reduction (OR=3.03); axis quadrant shift did not.

Limitations

Echocardiographic measures of LV volumes may be less accurate than other methods. The surface ECG may not accurately reflect important changes in LV activation. QRS score is modestly less accurate than newer imaging methods for calculating scar volume. We did not evaluate other forms of ventricular conduction delay.

Conclusions

Ventricular activation on the surface ECG accurately predicts ventricular reverse remodeling during CRT. Greater LVAT_max and smaller scar volume on the baseline LBBB ECG, combined with wavefront fusion on the paced ECG, are associated with higher probability of ESV reduction.