Predicting Survival in Repaired Tetralogy of Fallot

Objectives This study sought to identify patients with repaired tetralogy of Fallot (rTOF) at high risk of death and malignant ventricular arrhythmia (VA). Background To date there is no robust risk stratification scheme to predict outcomes in adults with rTOF. Methods Consecutive patients were prospectively recruited for late gadolinium enhancement (LGE) cardiovascular magnetic resonance (CMR) to define right and left ventricular (RV, LV) fibrosis in addition to proven risk markers. Results The primary endpoint was all-cause mortality. Of the 550 patients (median age 32 years, 56% male), 27 died (mean follow-up 6.4 ± 5.8; total 3,512 years). Mortality was independently predicted by RVLGE extent, presence of LVLGE, RV ejection fraction ≤47%, LV ejection fraction ≤55%, B-type natriuretic peptide ≥127 ng/L, peak exercise oxygen uptake (V02) ≤17 mL/kg/min, prior sustained atrial arrhythmia, and age ≥50 years. The weighted scores for each of the preceding independent predictors differentiated a high-risk subgroup of patients with a 4.4%, annual risk of mortality (area under the curve [AUC]: 0.87; P < 0.001). The secondary endpoint (VA), a composite of life-threatening sustained ventricular tachycardia/resuscitated ventricular fibrillation/sudden cardiac death occurred in 29. Weighted scores that included several predictors of mortality and RV outflow tract akinetic length ≥55 mm and RV systolic pressure ≥47 mm Hg identified high-risk patients with a 3.7% annual risk of VA (AUC: 0.79; P < 0.001) RVLGE was heavily weighted in both risk scores caused by its strong relative prognostic value. Conclusions We present a score integrating multiple appropriately weighted risk factors to identify the subgroup of patients with rTOF who are at high annual risk of death who may benefit from targeted therapy.

The lack of large prospective studies to support evidence-based decisions and therefore how to apply current clinical guidelines to individual patients is problematic (4,5). Pulmonary regurgitation (PR) is now a widely recognized hemodynamic substrate for VA and SCD, and considerable progress has been made in defining timing of pulmonary valve implantation (PVR) to counter it. Timely PVR alone, however, does not appear to abort the SCD risk, as myocardial fibrosis, a clear arrhythmic substrate for macro-re-entry VT remains (6,7). Multiple hemodynamic, structural, and electrophysiological risk factors have been described, although none sensitive and or specific enough to predict VT and SCD when used in isolation (8)(9)(10). The challenge, therefore, remains in selecting high-risk patients from a much larger rTOF cohort that overall has only a 0.15% annual risk of SCD (11) without contaminating the lives of remaining patients with implantable cardiac defibrillator (ICD) therapy with the physical and mental health issues associated with living with an ICD (11)(12)(13). A robust risk scheme integrating multiple risk factors appropriately is required (4).
Noninvasive assessment of VT substrates has been made possible using late gadolinium enhancement (LGE) cardiovascular magnetic resonance (CMR). We (14), and others, demonstrated association of LGE with right ventricular (RV) dysfunction, impaired exercise capacity, increased neurohormonal activation, and, importantly, sustained arrhythmia (atrial or ventricular) or syncope in cross-sectional studies (14).
The aim of this prospective study was to examine the prognostic value of LGE and to construct a weightedrisk score for death and VA incorporating all independent risk factors in order to help identify high-risk patients who require consideration of ICD, and other interventions, such as preventive VT ablation or further optimization of heart failure therapy. CMR IMAGE ACQUISITION AND ANALYSIS. A standardized CMR protocol for rTOF assessment was acquired in all patients in line with our published protocol (14). Short-axis Cines were acquired for calculation of volumes with 7-mm slice thickness and 3-mm gap (spatial resolution 1.9 Â 1.9 Â 7 mm).
Gadolinium-DTPA 0.1 mmol/kg intravenously was administered, and images were acquired from at least 8 to 10 minutes typically until at least 25 minutes after gadolinium was given. LGE images were obtained using an inversion-recovery gradient-echo sequence (spatial resolution 0.7 Â 0.7 Â 7 mm) with inversion times optimized to null normal myocardium by meticulous visual inspection of each image.
Images were repeated in 2 separate phase-encoding directions or cross-cut to exclude possible areas of artifact and to define subtle RVLGE. Ventricular volumes analysis excluded trabeculations from RV and left ventricular (LV) blood pool (14). Maximum length of RV outflow tract (RVOT) akinetic region and indexed right atrial area (RAAi) were measured as previously reported (9). RVLGE was semiquantified by 2 experienced operators blinded to clinical data using the previously published segmental scoring system designed by our group to account for the unique geometry of the RV (Figure 1) (14). LGE was considered present when in locations that either did not alter when re-imaged in the same plane with a phase-swap or remained visible in a second orthogonal or cross-cut plane. LVLGE was scored using the standard 17-segment LV model. LVLGE related to apical vent at time of surgery and RV/LV septal insertion points were not included in the analysis, as previously described (14). Interstudy reproducibility of RV and LV LGE scoring was tested by repeating scans and analysis in 20 patients performed by 2 different operators blinded to previous study (14).
The index LGE CMR study performed at the start of each patient's recruitment was included for analysis. STANDARD OF CARE INVESTIGATIONS. As part of routine care, patients undergo periodic 12-lead electrocardiograms (ECG), blood sampling for B-type natriuretic peptide (BNP), echocardiography, and cardiopulmonary exercise testing (15)(16)(17). RV restrictive physiology was defined by laminar anterograde Doppler flow in the pulmonary artery in late diastole present throughout the respiratory cycle ("a" wave).
We further classified restrictive RV physiology into the so-called primary phenotype and secondary Predicting Survival in Repaired Tetralogy of Fallot F E B R U A R Y 2 0 2 2 : 2 5 7 -2 6 8 phenotypes with a small or large RV, respectively.
Tricuspid regurgitation, RV systolic pressure (RVSP), tricuspid annulus planar excursion, and LV diastolic filling parameters were reported from echocardiography. Peak oxygen uptake (peak VO 2 ) was included when respiratory exchange ratio was >1. Ambulatory ECG Holter monitoring was performed if clinically indicated. Nonsustained VT (NSVT) was defined as $3 consecutive ventricular beats $100 beats per minute for #30 seconds' duration and sustained atrial arrhythmia was defined as $30 seconds' duration.
These were recorded from Holter readings, routine pacemaker device interrogations, and medical notes.

RV LV
Example of severe RV LGE extent in our study participant. The total RV LGE score was 12. RV-LV insertion point LGE, LV apical vent site LGE, and LV papillary muscle enhancement (asterisks) were not included in the score and were common. Patients with a total RV LGE score $8 were in the top quartile for RV LGE burden and were graded as severe extent. Patients with a total RV LGE score 5-7 were graded as moderate extent where score of 5 was the median. A total RV LGE score <5 was deemed as minimal or mild. The segmental system used for scoring RV LGE has been previously published. The RV is divided into 6 segments (yellow numbers 1-6). Regions of RV LGE were scored according to linear extent (0 ¼ no enhancement, 1 ¼ up to 2 cm, 2 ¼ up to 3 cm, 3 ¼ 3 or more cm in length) and number of trabeculations enhanced including the moderator band (0 ¼ no enhancement, 1 ¼ 1 trabeculation, 2 ¼ 2-4). Scoring of LVLGE was performed using the universally accepted 17segment LV model (14). Points were attributed to the proportion of LGE present in each myocardial segment, as visually judged: 0 ¼ no LGE, 1 ¼ up to 25%, 2 ¼ up to 50%, 3 ¼ up to 75%, and 4 ¼ up to and including 100% of the myocardium enhanced.
LGE ¼ late gadolinium enhancement; LV ¼ left ventricle; RV ¼ right ventricle.  values of only the variables that remained independently predictive of the outcomes and unrelated to one another in bivariable analysis were used as a guide to assign a relative weighting to each variable.
A weighted-risk stratification score was thus derived for mortality and the secondary endpoint VA, respectively. Receiver operating characteristic curves were used to determine whether risk scores for mortality and VA could be used to predict outcome. The thresholds for each risk category for mortality and VA were selected based on sensitivity and specificity for outcome.
Cox proportional hazards survival plots where generated to illustrate the survival differences between high-, intermediate-, and low-risk categories for mortality and VA. Patients who already had the endpoints were censored at baseline. The relative performance of our risk score was compared using receiver-operating characteristic analysis with 3 existing risk scores (8,11,18) and our proposed score with and without LGE. LGE related to extension of VSD patch or spontaneously closed VSD. Interobserver reproducibility of

RESULTS
LGE scoring was highly reproducible (intraclass correlation coefficient 0.97 and 1 for the RV and LV, respectively) (14). Clinical events at study end in relation to RVLGE extent are summarized in Supplemental Table 1.

ALL-CAUSE
MORTALITY. During the follow-up period, a total of 27 deaths were recorded (13 SCDs, 12 deaths caused by heart failure, and 2 noncardiac deaths). Univariable predictors are summarized in Table 1 and were consistent with previous reports/ consensus aside our novel finding that supramedian RVLGE score ($5) is associated with higher mortality.
Mortality is related to RVLGE extent ( Figure 2). NSVT, previous palliative shunt, ventriculotomy, and QRS duration >180 ms were not univariable predictors of mortality. Restrictive RV physiology was not predictive of mortality. In bivariable analysis, supramedian  Table 2).

RISK SCORE FOR PREDICTING MORTALITY.
A total weighted-risk score $51 demonstrated a 93% specificity vs sensitivity 51% for predicting all-cause mortality and was chosen as the lower threshold for the highest risk of death category. Conversely a total score #20 had a sensitivity 93% vs specificity 42% and was used as the upper threshold for the low-risk category. The applied risk score (Figure 3)  Freedom from VA was compromised as RVLGE extent increases ( Figure 2). Univariable predictors are summarized in Table 1. Restrictive RV physiology was predictive of VA only when associated with RV dilation but was not independent of RV dilation alone. In bivariable analyses, RVLGE score $5 remained an independent predictor, as did LVLGE, RVEF #35%,    Table 2). A total weighted-risk score $40 demonstrated the most favorable specificity 91% vs sensitivity 52% for predicting VA, hence was chosen as the lower threshold for the highest risk of VA category.
Conversely, a total score #20 with the most favorable sensitivity 90% over specificity 42% was used as the upper threshold for the low-risk category. The applied risk score (Figure 4)

DISCUSSION
We have shown how to identify patients with rTOF who are at high annual risk of death by using a weighted-risk score that integrates clinical, LGE CMR, exercise, and BNP measurement (Central Illustration).
This performs better than previously proposed risk models (8,11,18). We have also enabled personalized risk stratification specific to malignant VA. This is the largest prospective study to date that also examines LGE extent and long-term outcomes, in a highly characterized adult rTOF cohort with a considerable follow-up period and hard clinical endpoints. We show for the first time that the extent of LGE is a significant and independent predictor of mortality, justifying its routine and periodic inclusion in the clinical surveillance of adults with rTOF. applicable. In addition, other noninvasive risk predictors for outcomes have since been described (8,9,15,17). In keeping with the largest multicenter observational registry study to date (International Multicenter TOF Registry), atrial arrhythmia and CMR-derived RVEF and LVEF were predictors of outcome in our study (8). Prospective studies have also identified other CMR-derived factors including RAAi $16 cm 9 and increased RVOT akinetic length (9), reduced peak VO 2 , (15,16) increased BNP, (17) and increased RVSP (8). These parameters were also univariable predictors of outcome in our study. In Scores were calculated using the model in Figure 3 without the inclusion of LGE cardiovascular magnetic resonance, given that LGE is not in routine clinical practice for this condition. b To enable testing of existing models on our study cohort, points were allocated, using a similar approach to our study to the predictive cutoffs reported by Valente et al (8)  to be benign; it was neither associated with mortality nor significant VA, calling into question guidelines suggesting NSVT should be considered for ICD or its use as a surrogate secondary endpoint in rTOF. Our secondary composite endpoint for VA did not include appropriate ICD shock to ensure it was robust and avoid concerns that appropriate ICD (24) therapy could be delivered for potentially benign NSVT. Furthermore, our study was to predict prognosis and not device outcomes (11). STUDY LIMITATIONS. This was a single-center study, yet our cohort was large and followed for a long period, with rTOF repair at many centers (7 in the United Kingdom and other international), hence reflective of various surgical eras and approaches and representative of this heterogeneous population.
RVLGE CMR acquisition requires training to avoid false negative reporting (14), although recent CMR sequences have made LGE acquisition less operator-dependent, making wider uptake easier (25) and in future enabling comprehensive highresolution coverage (26,27). We continued to use our previously published RV segmental scoring system for LGE (14), given its high reproducibility and its simplicity, and for consistency in this prospective study. Signal-intensity-based thresholds might be considered an alternative for quantifying RVLGE but are limited by partial volume effects, sternal wire artifact, epicardial fat, and the thin RV wall. No studies to date have validated this in the uniquely shaped RV after rTOF repair (28,29).
External validation of these risk score algorithms in a new cohort will be possible once similar data are collected systematically at scale. FUTURE DIRECTION. RVLGE was heavily weighted in both risk scores caused by its strong relative prognostic value. Future studies of total fibrosis burden will also quantify LV interstitial fibrosis (T1 mapping CMR) and require bespoke approaches for the RV (30) and there may be ways to measure fibrosis activity. Risk score identifies contemporary adult patients with repaired tetralogy of Fallot at high annual risk of death. Abbreviations as in Figures 1, 2, and 4. can expect long and healthy lives, but a small minority are at much higher risk for premature cardiovascular death.
For the first time, we show LGE extent is prognostic, justifying its inclusion in clinical practice. We present a weighted-risk score to identify the subgroup of patients with rTOF who are at high annual risk of death who may benefit from targeted therapy with ICDs, VT ablation, or heart failure therapy.