Denervated Myocardium is Preferentially Associated with Sudden Cardiac Arrest in Ischemic Cardiomyopathy: A Pilot Competing Risks Analysis of Cause-Specific Mortality

Abstract

Background

Previous studies have identified multiple risk factors that are associated with total cardiac mortality. Nevertheless, identifying specific factors that distinguish patients at risk of arrhythmic death versus heart failure could better target patients likely to benefit from implantable cardiac defibrillators (ICDs), which have no impact on non-sudden cardiac death (NSCD).

Methods and Results

We performed a pilot competing risks analysis of the NIH-sponsored PAREPET trial (Prediction of ARrhythmic Events with Positron Emission Tomography). Death from cardiac causes was ascertained in subjects with ischemic cardiomyopathy (n=204) eligible for an ICD for the primary prevention of sudden cardiac arrest (SCA) after baseline clinical evaluation and imaging at enrollment (PET and 2D-echo). Mean age was 67±11 years with an EF of 27±9%, and 90% were male. Over 4.1 years of follow-up, there were 33 SCAs (arrhythmic death or ICD discharge for ventricular fibrillation or ventricular tachycardia >240 bpm) and 36 NSCDs. SCA was correlated with a greater volume of denervated myocardium (defect of the PET norepinephrine analog ¹¹C-hydroxyephedrine), lack of angiotensin inhibition therapy, elevated B-type natriuretic peptide, and larger LV end-diastolic volume index (LVEDVI). In contrast, NSCD was associated with a higher resting heart rate, older age, elevated creatinine, larger left atrial volume index, and larger LVEDVI.

Conclusions

Distinct clinical, laboratory and imaging variables are associated with cause-specific cardiac mortality in primary prevention candidates with ischemic cardiomyopathy. If prospectively validated, these multi-variable associations may help target specific therapies to those at greatest risk of sudden and non-sudden cardiac death.

INTRODUCTION

Most previous studies have evaluated risk factors for cardiovascular death and the impact of therapeutic interventions in terms of either all-cause mortality or total cardiac mortality. While death in patients with ischemic cardiomyopathy is usually due to underlying cardiac disease, the cause-specific mechanism is almost equally split between sudden cardiac arrest (SCA) from arrhythmic events and non-sudden cardiac death (NSCD) due to pump failure. Implantable cardiac defibrillators (ICDs) can effectively terminate lethal ventricular arrhythmias and their prophylactic use has been shown to improve survival^{1, 2}. Nevertheless, they do nothing to mitigate NSCD. As a result, their overall clinical benefit may become limited in circumstances where the competing risk from NSCD and/or non-cardiac death is high. For example, advanced heart failure symptoms predict a high likelihood of death from pump failure and because of this, patients with New York Heart Association (NYHA) Function Class IV heart failure symptoms are generally not considered candidates for a primary prevention ICD. Despite intense investigation, identification of patients at high risk of arrhythmic death is currently based primarily on a depressed left ventricular (LV) ejection fraction (EF). Unfortunately, EF does not accurately distinguish cause-specific mortality since the risk of arrhythmic death as well as death from heart failure both increase as EF declines.

Targeting of ICD therapy to patients at highest risk could be significantly advanced by identifying factors that predict cause-specific mortality from SCA^3–8. The few studies that have considered SCA have usually ignored the fact that NSCD is an equally frequent competing end-point. The ability to risk stratify patients in terms of cause-specific mortality would advance our ability to identify those most likely to benefit from an ICD^{8, 9}. In addition, predicting an increased risk of NSCD could be used to identify patients who might benefit from alternate or aggressive up-titration of medical therapy for heart failure, or even a patient cohort at such high risk of NSCD that ICD therapy should perhaps be withheld^{5, 10, 11}. Towards this end, we determined whether specific risk factors could be identified that were associated with SCA versus NSCD using a competing risks analysis of the Prediction of ARrhythmic Events with Positron Emission Tomography (PAREPET) trial. The results identify independent risk factors that are associated with cause-specific cardiac mortality in a cohort with ischemic cardiomyopathy eligible for a primary prevention ICD.

METHODS

Study Population

The NIH-sponsored PAREPET trial (NCT01400334, Supplement)^{12, 13} enrolled 204 subjects with ischemic cardiomyopathy who were eligible to receive an ICD for the primary prevention of SCA^{12, 13}. ICD implantation was refused by 29 subjects (14%), with no significant differences in any study variable between those with and without an ICD (Supplement). All subjects gave informed consent and the study was approved by the University at Buffalo and Veterans Affairs Western New York Healthcare System Institutional Review Boards.

Positron Emission Tomography

PET imaging was performed as previously described^{12, 13}. Briefly, myocardial sympathetic innervation was assessed with ¹¹C-meta-hydroxyephedrine (HED), ¹³N-ammonia, (NH₃) was used to measure perfusion, and ¹⁸F-2-deoxyglucose (FDG) during a hyperinsulinemic-euglycemic clamp was used to confirm metabolic viability^{12, 13}. PET images were analyzed with FlowQuant® software (University of Ottawa Heart Institute)¹⁴. Tracers were quantified in each of 496 myocardial sectors relative to the peak tracer activity determined from the highest 5% of sectors^{12, 14}. HED uptake <75% of the peak activity defined denervated myocardium^{12, 13}. Sectors with NH₃ or FDG ≥80% of the peak activity were considered normal. Infarcted and hibernating myocardium were quantified from a mismatch analysis between NH₃ and FDG^{13, 14}. All defects were quantified as the integrated reduction in each sector with reduced uptake, and expressed as % LV¹³.

Echocardiography and Electrocardiography

Transthoracic echocardiography was performed with a 2.25 MHz phased-array transducer with harmonic imaging (SONOS 7500, Philips Medical Inc.)¹². Chamber volumes and LVEF were quantified as recommended by the American Society of Echocardiography. A resting supine electrocardiogram was obtained at 1000 Hz (H12+ recorders, V3.12, Mortara Instruments)¹⁵.

Clinical End-points

The primary end-point was SCA (including ICD equivalent), with a secondary end-point of total cardiac mortality (Supplement)^{12, 13}. Since ICD therapy overestimates the frequency of SCA in control subjects, the ICD equivalent for SCA was conservatively defined as ICD discharge for fast ventricular tachycardia (>240 bpm) or ventricular fibrillation^{12, 13}. The secondary end-point of total cardiac mortality included SCA as well as NSCD. This was blindly adjudicated by three board-certified cardiologists¹².

Statistical Analyses

All values are mean ± SD. Differences between groups were compared using unpaired t-tests for continuous data and the χ² test for categorical values (Microsoft Excel). Variables were tested for their association with total cardiac mortality, SCA, and NSCD (Supplement)¹³. Multivariable analyses for these individual end-points were performed with backward selection (SAS; Cary, NC). Statistical significance was defined as a two-tailed p≤0.05. The time to end-points was analyzed using the Kaplan-Meier method, and the difference between groups was assessed with the log-rank statistic.

A competing risks analysis was performed to simultaneously assess factors associated with SCA and NSCD (Supplement). The variables that resulted from the multivariable analyses of the individual end-points (total cardiac mortality, SCA, or NSCD) were included in the competing risks analysis. The competing risks analysis was performed using the “cmprsk” package¹⁶ within the R statistical software¹⁷, which fits proportional sub-distribution hazard regression models¹⁸. Backward selection was used to find all statistically significant variables for both SCA and NSCD. Follow up analyses of the significant variables from the competing risks analysis used model-based standardized risk estimates¹⁹.

RESULTS

Table 1 summarizes the key variables. Almost all subjects received anti-platelet therapy or warfarin (99%), β-blockers (96%), and angiotensin inhibition therapy (angiotensin converting enzyme inhibitor or angiotensin receptor antagonist, 90%). Over a median follow-up of 4.1±1.9 years there were 69 cardiac deaths (34%), of which 33 (16%, ~4.1%/year) were SCA and 36 (18%, ~4.3%/year) were NSCD (Figure 1). Of the 33 SCA events, 11 were ICD equivalents. Variables associated with time to the various end-points (total cardiac mortality, SCA, and NSCD) are summarized in Table 2 and Supplemental Table 2. Of the PET variables, only denervated myocardium was associated with total cardiac mortality, and denervated and viable denervated myocardium (mismatch between FDG and HED uptake) were associated with SCA.

Table 1.

Clinical and Demographic Variables

Variable	Average (n=204)
PET Variables (% of LV)
Denervated Myocardium	27±11
Viable Denervated Myocardium	8±6
Infarcted Myocardium	20±9
Hibernating Myocardium	3±3
Echocardiographic and Electrocardiographic Variables
LV End-Diastolic Volume Index (ml/m2)	89±30
LV End-Systolic Volume Index (ml/m2)	66±26
Mitral Regurgitation Severity (0 to 4)	1.6±1.2
Left Atrial Volume Index (ml/m2)	43±16
LV Ejection Fraction (%)	27±9
QRS Duration (ms)	136±35
LV Mass Index (g/m2)	153±46
Laboratory Variables
Creatinine (mg/dL)	1.4±0.8
B-type Natriuretic Peptide (ng/L)	466±532
Hematocrit (%)	41±5
Demographics/Medications
Diabetes Mellitus	95 (47%)
Digoxin Therapy	79 (39%)
History of Revascularization	159 (78%)
Age (years)	67±11
Amiodarone Therapy	19 (9%)
Resting Heart Rate (bpm)	66±12
No Angiotensin Inhibition Therapy	20 (10%)
Rate • Pressure Product (per 100 bpm•mm Hg)	8300±2200
Female	21 (10%)

LV – Left Ventricular. Values are mean±SD or n (%).

Figure 1. Cumulative Distribution Functions for Sudden Cardiac Arrest (SCA), Non-Sudden Cardiac Death (NSCD), and Total Cardiac Mortality.

These curves show the cumulative cardiac mortality in the PAREPET trial to 5 years. Adjudicated SCA (red line) included ICD discharge for ventricular tachycardia >240 bpm and ventricular fibrillation. Total cardiac mortality (gray line) included SCA as well as NSCD (blue line).

Table 2.

Univariate Associations with Various End-Points

Variable	Total Cardiac Mortality		Sudden Cardiac Arrest		Non-Sudden Cardiac Death
	Hazard Ratio	p-value	Hazard Ratio	p-value	Hazard Ratio	p-value
PET Variables (per 10% of LV)
Denervated Myocardium	1.36	0.007	1.72	0.001	1.11	0.49
Viable Denervated Myocardium	1.41	0.092	1.94	0.022	1.05	0.86
Infarcted Myocardium	1.17	0.23	1.31	0.17	1.07	0.72
Hibernating Myocardium	0.83	0.69	0.62	0.51	1.04	0.95
Echocardiographic and Electrocardiographic Variables
LV End-Diastolic Volume Index (per 10 ml/m2)	1.19	<0.0001	1.24	<0.0001	1.14	0.015
LV End-Systolic Volume Index (per 10 ml/m2)	1.21	<0.0001	1.25	<0.0001	1.17	0.004
Mitral Regurgitation Severity (0 to 4)	1.57	<0.0001	1.49	0.008	1.65	0.0007
Left Atrial Volume Index (per 10 ml/m2)	1.23	0.0002	1.18	0.063	1.27	0.0009
LV Ejection Fraction (per 10%)	0.62	0.0007	0.61	0.017	0.63	0.016
QRS Duration (per 10 ms)	1.07	0.062	1.00	0.97	1.12	0.013
LV Mass Index (per 10 g/m2)	1.05	0.041	1.07	0.043	1.03	0.37
Laboratory Variables
Creatinine (per 0.1 mg/dL)	1.06	<0.0001	1.05	0.001	1.07	<0.0001
B-type Natriuretic Peptide (per 100 ng/L)	1.06	0.0002	1.07	0.0004	1.05	0.019
Hematocrit (per 1%)	0.92	0.002	0.93	0.079	0.91	0.007
Demographics/Medications
Diabetes Mellitus	0.63	0.056	1.18	0.63	2.11	0.032
Digoxin Therapy	1.73	0.024	1.06	0.87	2.65	0.004
History of Revascularization	0.56	0.028	0.42	0.019	0.74	0.44
Age (per 10 years)	1.23	0.059	0.94	0.69	1.60	0.004
Amiodarone Therapy	1.96	0.50	1.13	0.84	2.84	0.013
Resting Heart Rate (per 10 bpm)	1.20	0.055	0.87	0.42	1.48	0.0007
No Angiotensin Inhibition Therapy	2.20	0.021	3.33	0.005	1.24	0.72
Rate • Pressure Product (per 100 bpm•mm Hg)	1.01	0.082	1.00	0.55	1.02	0.003
Female	1.10	0.82	1.10	0.99	0.53	0.15

LV – Left Ventricular. Statistical significance is highlighted in bold.

Individual multivariable analyses were performed to determine the independent factors for time to total cardiac mortality, SCA, and NSCD (Table 3). The variables that were independently associated with total cardiac mortality included LVEDVI, BNP, creatinine, and angiotensin inhibition therapy. Independent variables for SCA also included LVEDVI, creatinine, and angiotensin inhibition therapy, as well as denervated myocardium. Four different variables correlated with NSCD: age, heart rate, LA volume index, and LVEF. The nine variables that were identified using these analyses (Table 3) were included in the competing risks model (details below). Cumulative distribution functions comparing total and cause-specific mortality for each of these variables is illustrated in Figures 2, 3, and 4.

Table 3.

Multivariable Analyses for Time to Various Cardiac End-points.

End-point	Variable	Estimate	Hazard Ratio	p-value
Total Cardiac Mortality	Continuous
	LV End-Diastolic Volume Index (per 10 ml/m2)	0.22	1.25	<0.0001
	B-type Natriuretic Peptide (per 100 ng/L)	0.05	1.05	0.007
	Creatinine (per 0.1 mg/dL)	0.06	1.06	0.007
	Dichotomized
	No Angiotensin Inhibition Therapy	1.93	6.92	<0.0001
Sudden Cardiac Arrest	Continuous
	Denervated Myocardium (per 10% of LV)	0.67	1.95	0.004
	LV End-Diastolic Volume Index (per 10 ml/m2)	0.27	1.31	0.005
	Creatinine (per 0.1 mg/dL)	0.09	1.09	0.003
	Dichotomized
	No Angiotensin Inhibition Therapy	1.97	7.72	0.003
Non-Sudden Cardiac Death	Continuous
	Age (per 10 years)	0.66	1.94	0.010
	Resting Heart Rate (per 10 bpm)	0.35	1.42	0.026
	Left Atrial Volume Index (per 10 ml/m2)	0.44	1.55	0.032
	LV Ejection Fraction (per 10%)	-0.50	0.61	0.045

LV – Left Ventricular.

Figure 2. Cumulative Distribution Functions for Variables Associated with Both Sudden Cardiac Arrest and Non-Sudden Cardiac Death.

Left ventricular end-diastolic volume index (LVEDVI), B-type natriuretic peptide (BNP), creatinine, and LV ejection fraction (EF) were all associated with total cardiac mortality (left graphs) as well as its components of sudden cardiac arrest (middle graphs) and non-sudden cardiac death (right graphs). For all variables, the event rates for the high risk (red lines; LVEDVI >99 ml/m², BNP >446 ng/L, creatinine >1.4 mg/dL, EF <24%), intermediate risk (yellow lines) and low risk (green lines; LVEDVI <74 ml/m², BNP <191 ng/L, creatinine <1.1 mg/dL, EF >31%) tertiles are shown. Corresponding p-values for the univariate analyses using continuous variables are in Table 2.

Figure 3. Cumulative Distribution Functions for Variables Preferentially Associated with Sudden Cardiac Arrest.

In addition to associations with total cardiac mortality (left graphs), denervated myocardium and angiotensin inhibition therapy were also significantly associated with sudden cardiac arrest (middle graphs). However, neither variable was associated with non-sudden cardiac death (right graphs). The event rates for the high risk (red lines; denervated myocardium >32.8% of LV and no angiotensin inhibition therapy), intermediate risk (yellow lines) and low risk (green lines; denervated myocardium <22.4% of LV and angiotensin inhibition therapy) tertiles are shown. Corresponding p-values for the univariate analyses using continuous variables are in Table 2.

Figure 4. Cumulative Distribution Functions for Variables Preferentially Associated with Non-Sudden Cardiac Death.

Resting heart rate, age and LA volume index were significantly associated with non-sudden cardiac death (right graphs). None of the variables were associated with sudden cardiac arrest (middle graphs), and only LA volume index was significantly associated with total cardiac mortality (left graphs). For all variables, the event rates for the high risk (red lines; heart rate >69 bpm, age >74 years, LA volume index >46.3 ml/m²), intermediate risk (yellow lines) and low risk (green lines; heart rate <60 bpm, age <61 years, LA volume index <33.7 ml/m²) tertiles are shown. Corresponding p-values for the univariate analyses using continuous variables are in Table 2.

Competing Risks Model

In the competing risks analysis, LVEF was not associated with NSCD (p=0.21) or SCA (p=0.82). The remaining eight variables were significantly associated with SCA and/or NSCD (Table 4). After adjustment for competing risks, SCA was associated with a greater volume of denervated myocardium, elevated BNP, no angiotensin inhibition therapy, and larger LVEDVI. NSCD was associated with elevated LVEDVI, higher resting heart rate, larger LA volume index, elevated creatinine, and older age. If denervated myocardium was excluded from the competing risks model, there were only minor changes (Table 5). Figures 5 and 6 illustrate the effect of each of these variables on the probability of SCA and NSCD using model-based standardized risk estimates while controlling for the other variables¹⁹. For the continuous variables, significant relationships with individual end-points are reflected by changes in risk across the range of data, whereas non-significant relationships produce more horizontal curves. Importantly, when assessed as continuous variables, the 5-year event rate for SCA risk progressively rose with increases in the amount of denervated myocardium, LVEDVI and BNP while NSCD declined. In contrast, the 5-year NSCD event rate progressively rose with increases in resting heart rate, creatinine, LA volume index and age while SCA declined or remained unchanged.

Table 4.

Competing Risks Model for Sudden Cardiac Arrest versus Non-Sudden Cardiac Death.

	Sudden Cardiac Arrest		Non-Sudden Cardiac Death
Variable	Hazard Ratio (95% CI)	p-value	Hazard Ratio (95% CI)	p-value
Denervated Myocardium (per 10% of LV)	1.45 (1.00–2.10)	0.05	0.93 (0.70–1.24)	0.62
B-type Natriuretic Protein (per 100 ng/L)	1.04 (1.01–1.08)	0.01	1.01 (0.95–1.06)	0.87
No Angiotensin Inhibition Therapy (dichotomized)	5.62 (2.18–14.3)	0.0003	1.22 (0.22–6.76)	0.82
LV End-Diastolic Volume Index (per 10 ml/m2)	1.32 (1.10–1.57)	0.002	1.13 (1.00–1.26)	0.04
Resting Heart Rate (per 10 bpm)	0.84 (0.47–1.51)	0.56	1.50 (1.15–1.95)	0.003
Creatinine (per 0.1 mg/dL)	1.01 (0.97–1.04)	0.66	1.04 (1.01–1.08)	0.02
Left Atrial Volume Index (per 10 ml/m2)	0.89 (0.69–1.14)	0.35	1.21 (1.06–1.38)	0.04
Age (per 10 years)	0.81 (0.59–1.11)	0.19	1.45 (1.00–2.11)	0.05

LV – left ventricular. Statistical significance is highlighted in bold.

Table 5.

Competing Risks Model Excluding PET Variables.

	Sudden Cardiac Arrest		Non-Sudden Cardiac Death
Variable	Hazard Ratio (95% CI)	p-value	Hazard Ratio (95% CI)	p-value
LV End-Diastolic Volume Index (per 10 ml/m2)	1.28 (1.12–1.46)	0.0002	1.12 (0.99–1.26)	0.068
No Angiotensin Inhibition Therapy (dichotomized)	5.42 (2.16–13.9)	0.0007	1.30 (0.25–6.78)	0.77
B-type Natriuretic Protein (per 100 ng/L)	1.05 (1.02–1.09)	0.002	1.00 (0.95–1.06)	0.94
Resting Heart Rate (per 10 bpm)	0.76 (0.46–1.25)	0.27	1.56 (1.21–2.01)	0.0006
Left Atrial Volume Index (per 10 ml/m2)	1.02 (0.83–1.25)	0.89	1.21 (1.07–1.37)	0.003
Creatinine (per 0.1 mg/dL)	1.02 (0.98–1.05)	0.33	1.04 (1.01–1.08)	0.02
Age (per 10 years)	0.81 (0.62–1.05)	0.11	1.56 (1.07–2.27)	0.02

LV – left ventricular. Statistical significance is highlighted in bold.

Figure 5. Estimated Event Rates from the Competing Risks Analysis for Variables Preferentially Associated with Sudden Cardiac Arrest.

The upper graph of each pair shows the anticipated effect of each variable on sudden cardiac arrest (SCA, red lines) and non-sudden cardiac death (NSCD, blue lines) within 5 years. Denervated myocardium, B-type natriuretic peptide (BNP), left ventricular end-diastolic volume index (LVEDVI), and angiotensin inhibition were independently associated with SCA. LVEDVI was also weakly correlated with NSCD. The probability of total cardiac mortality is also shown (black lines), although this was not directly derived from the competing risks model. The histogram associated with each variable illustrates the frequency of values in the PAREPET cohort, as well as the frequency of outliers (<5% of the total). For these figures the effect of each variable is assessed individually while controlling for the other variables. The p-values are derived from the competing risks analysis.

Figure 6. Estimated Event Rates from the Competing Risks Analysis for Variables Preferentially Associated with Non-Sudden Cardiac Death.

The upper graph of each pair shows the anticipated effect of each variable on sudden cardiac arrest (SCA, red lines) and non-sudden cardiac death (NSCD, blue lines) within 5 years. Resting heart rate, creatinine, left atrial (LA) volume index, and age were associated with NSCD. The probability of total cardiac mortality is also shown (black lines), although this was not directly derived from the competing risks model. The histogram associated with each variable illustrates the frequency of values in the PAREPET study, as well as the frequency of outliers (<5% of the total). For these figures the effect of each variable is assessed individually while controlling for the other variables. The p-values are derived from the competing risks analysis.

DISCUSSION

Patients with ischemic cardiomyopathy who are eligible for an ICD for the primary prevention of SCA constitute a cohort at very high risk of cardiac mortality (34% at a median follow-up of 4.1 years in the present study). Multivariable analysis demonstrates independent variables associated with an increased risk of total cardiac mortality. These included LVEDVI, BNP, creatinine and the lack of angiotensin inhibition therapy. While evaluating the risk of cardiac mortality provides prognostic information, cost-effective decision making with regard to ICD therapy for primary prevention requires approaches to discriminate patients with high risk of SCA from those more likely to succumb to heart failure and pump dysfunction^{8, 20}. In this regard, our pilot competing risks analysis identifies a set of variables that are independently associated with SCA versus NSCD in this high-risk cohort with ischemic cardiomyopathy. If prospectively validated in a larger population, this could provide an approach to improve the targeting of ICD therapy to patients at greatest risk of arrhythmic death¹³.

Variables Associated with Sudden Versus Non-Sudden Cardiac Death

Relatively few studies have evaluated cause-specific mortality and most, including the primary analysis of the PAREPET trial¹³, focused on SCA versus death from all other causes. The present study extends our original report by providing novel information from a competing risks analysis. The competing risks methodology allows for the simultaneous assessment of the mutually exclusive components of cardiac mortality, namely SCA and NSCD. With regard to SCA, three of the variables remained the same as in the original multivariable analysis of SCA alone¹³. These included greater extent of denervated myocardium, larger LVEDVI, and lack of angiotensin inhibition therapy. The competing risks analysis replaced an elevated creatinine with an elevated BNP. This change is consistent with a previous study that evaluated clinical, hemodynamic, and neurohormonal factors in patients with heart failure, and found that BNP was the only independent variable associated with SCA³. The factors we identified suggest that SCA risk is not only influenced by potential alterations in ventricular repolarization (due to regional sympathetic denervation)¹³, but also by ventricular remodeling (as reflected by elevated LVEDVI and prevented by angiotensin inhibition therapy) and chronic elevations in LV end-diastolic pressure (BNP). These factors would also tend to promote interstitial fibrosis, which would delay ventricular impulse conduction and increase the risk of reentry. They could also lead to stretch-induced arrhythmogenesis which has been observed in an animal model of sudden death from chronic ischemia without infarction²¹.

In terms of NSCD, the competing risks analysis identified elevated resting heart rate, larger LA volume index, elevated creatinine, and older age in addition to larger LVEDVI. This association of older age with NSCD is consistent with previous findings. While the incidence of both SCA and NSCD rises with age²², the proportion of deaths that are sudden declines, reflecting both a greater increase in NSCD as well as an increasing likelihood of non-cardiac causes^{7, 8, 22}. While the other variables have been previously associated with total cardiac mortality, there is limited data regarding their utility for cause-specific cardiovascular mortality. The significant association between creatinine and NSCD in the competing risks analysis corroborates the conclusions of previous studies which have suggested that ICD therapy may be less effective among those with advanced chronic kidney disease^{5, 8, 23}.

Comparison with Previous Studies

To our knowledge, only two prior studies^{6, 7} have performed competing risks analysis of SCA versus NSCD without the confounding effects of including patients with a history of resuscitated SCA^{10, 11}. In the ATLAS trial (Assessment of Treatment with Lisinopril And Survival, conducted from 1992–7), both sudden and non-sudden death were associated with older age, ischemic LV dysfunction, lower EF, and elevated creatinine. In addition, SCA was correlated with nitrate therapy, pre-study ACE inhibitor therapy, lack of β-blocker therapy, and lack of antiarrhythmic therapy. Additional risk factors for NSCD included a lower systolic blood pressure, elevated resting heart rate, lower serum sodium, antiarrhythmic therapy, and lack of aspirin therapy. Although there are some similarities with the results of the present analysis, direct comparison is limited due to our inclusion of PET and echocardiographic imaging, as well as advances in contemporary medical therapy. A recent competing risk analysis of ambulatory heart failure patients did not identify any significant factors associated with SCA⁷. NSCD was correlated with older age, higher NYHA Functional Class, and greater Charlson comorbidity index⁷.

An alternative to a competing risk analysis was the application of the Seattle Heart Failure Model, which was initially derived for total cardiac mortality, to cause-specific cardiac mortality⁴. When tested in over 10,000 subjects with heart failure, the model successfully identified SCA and pump-failure death⁴. Assessment of the area under the receiver operating characteristic curves (AUC) demonstrated that the model was significantly better for pump-failure (AUC=0.85, 95% CI, 0.83 – 0.87) than SCA (AUC=0.68, 95% CI, 0.65 – 0.70)⁴. Mortality decreased in those with lower scores, yet the cause of death was more likely to be sudden. With higher scores, pump-failure death predominated⁴. This observation was corroborated in an analysis of the MADIT-II database. Among a pre-specified subgroup at very high risk of death (blood urea nitrogen ≥50 mg/dL or creatinine ≥2.5 mg/dL), randomization to ICD therapy conferred no improvement in survival⁵. This important influence of renal function is consistent with our results, where elevated creatinine was an independent variable for NSCD but not SCA in the competing risks analysis (Table 4). Recently, the Seattle Proportional Risk Model, which was developed to predict the mode of death (sudden versus non-sudden, as opposed to risk), importantly showed that ICD benefit could in fact be predicted by a greater risk of sudden versus non-sudden death⁸. Although methodologic differences preclude accurate comparison, many variables (age, sex, renal function, diabetes mellitus, digoxin use) show similar associations with SCA versus NSCD in the present study⁸.

PET Variables Associated with Cardiac Mortality

PAREPET was specifically designed to determine whether PET variables could improve the prediction of SCA (primary end-point)¹³ and/or total cardiac mortality (secondary end-point). The volume of denervated myocardium was the only PET variable that showed an association with total cardiac mortality (Table 1). As shown in Table 2 and Figure 3, this primarily reflected the association of denervated myocardium with SCA¹³. In contrast to a previous study that used magnetic resonance imaging (MRI) to quantify infarct size²⁴, we found no correlation between the volume of infarcted myocardium and total cardiac mortality. This was likely due to the fact that by restricting our cohort to primary prevention ICD candidates, coronary artery disease and heart failure were more advanced. In this regard, our results are consistent with other MRI data showing that the peri-infarct zone of hyper-enhancement, rather than core infarct size, is associated with mortality²⁵. As we have previously discussed, the peri-infarct region likely contains denervated but viable myocardium, which in addition to the infarct region defines the total volume of denervated myocardium¹³.

Numerous investigations have determined an association between altered sympathetic innervation and SCD, with clinical studies showing denervation as a result of infarction, reversible ischemia, and non-ischemic etiologies²⁰. Although the mechanism(s) remains to be definitively determined, most evidence supports the development of an arrhythmogenic substrate due to inhomogeneity in sympathetic innervation as a result of partial denervation, nerve sprouting, and/or denervation hypersensitivity²⁶. This results in regional inhomogeneity in ventricular repolarization, which is exacerbated during periods of sympathetic activation^{26, 27}.

Limitations

In order to account for SCA events among subjects with an ICD, we included ICD discharges for ventricular fibrillation or fast ventricular tachycardia (>240 bpm), as the frequency of these arrhythmic events approximated the benefit of an ICD on survival in MADIT II^{28, 29}. A less restrictive threshold would obviously increase the frequency of arrhythmic events (potential SCA equivalents), and could alter the results of this analysis. Nevertheless, we have recently reported that ICD discharges for ventricular tachycardia <240 bpm appear to be associated with a different cardiac substrate (significantly less denervated myocardium) than was found among those with the present definition of SCA³⁰. Secondly, ICD programming was not standardized and the actual programming at the time of events was not recorded. This limitation may have affected the frequency of SCA equivalents. Third, adjudicated sudden death occurred in patients receiving an ICD. It is well known that ICDs don’t prevent all sudden deaths from tachyarrhythmias for reasons including shock resistant VT/VF, post-shock electromechanical dissociation, ICD under sensing and lead malfunction as well as causes of sudden death that are not related to ventricular tachyarrhythmias³¹. While very few device trials have adjudicated cause specific mortality, our incidence of 10% is similar to the MUSTT trial where the 5-year adjudicated sudden death rate using identical classification criteria was 9% in patients receiving an ICD³². Finally, this is a pilot study of modest size with a limited number of events; therefore, our final model is at risk of over-fitting and will require prospective validation.

Clinical Implications

While the presence of ischemic cardiomyopathy identifies a cohort that is at very high risk of cardiac death, distinct subsets of clinical and imaging variables can differentially identify risk of SCA versus NSCD. Our results are consistent with previously published findings indicating that that there are subgroups of primary prevention ICD recipients in which the likelihood of NSCD without prior device therapy can significantly exceed the survival benefit of an ICD¹⁰. Thus, there is a significant proportion of patients at low risk of SCA when candidacy for primary prevention ICD therapy is identified solely on the basis of EF. The primary results of the PAREPET trial corroborated this by demonstrating that the risk of SCA was low (<1% per year) in subjects with none of four risk factors identified by multivariable analysis (retrospectively optimized cut-points, denervated myocardium >37.6% of LV, LVEDVI >99 ml/m², creatinine >1.49 mg/dL, and no angiotensin inhibition therapy)¹³. This group represented nearly half (44%) of our patient population. The current study extends this conclusion using a competing risk analysis which replaced creatinine with BNP as an independent factor associated with SCA. In addition, this methodology identified independent risk factors for an increased likelihood of NSCD. If this model can be prospectively validated, it may be possible to divide patients with ischemic cardiomyopathy and an EF ≤35% into those who are likely to benefit from ICD therapy as opposed to those at high risk of NSCD which would limit the impact of an ICD on survival.

CLINICAL PERSPECTIVE.

The presence of ischemic cardiomyopathy identifies a cohort at very high risk of cardiac death. Our results suggests that distinct subsets of clinical and imaging variables can differentially identify risk of SCA versus NSCD, and are consistent with previously published findings indicating that that there are subgroups of primary prevention ICD recipients in which the risk of NSCD without prior device therapy can significantly exceed the survival benefit of an ICD. Thus, there is a significant proportion of patients at low risk of SCA when candidacy for primary prevention ICD therapy is identified solely on the basis of EF. The primary results of the PAREPET trial corroborated this by demonstrating that the risk of SCA was low (37.6% of LV, LVEDVI >99 ml/m², creatinine >1.49 mg/dL, and no angiotensin inhibition therapy). This group represented nearly half (44%) of our patient population. The current study extends this conclusion using a competing risk analysis which replaced creatinine with BNP as an independent factor associated with SCA. In addition, this methodology identified independent risk factors for an increased likelihood of NSCD. If this model can be prospectively validated, it may be possible to divide patients with ischemic cardiomyopathy and an EF ≤35% into those who are likely to benefit from ICD therapy as opposed to those at high risk of NSCD which would limit the impact of an ICD on survival.