Stress Cardiac Magnetic Resonance Ischemia Burden and Cardiovascular Events: Post-hoc Analysis from the ISCHEMIA Trial

Abstract

Background:

Research comparing the prognostic value of stress cardiac magnetic resonance (CMR) to other stress modalities in patients with coronary disease is limited.

Objectives:

We compared the prognostic value of stress CMR vs. alternative testing by either single photon emission computed tomography or stress echocardiography (SPECT/ECHO) in the ISCHEMIA trial.

Methods:

CMR vs SPECT/ECHO was compared in 3,909 patients randomized in ISCHEMIA after sites’ interpretation of moderate to severe ischemia. Ischemia and infarct extent, measured by either CMR or SPECT/ECHO, were each associated with the trial’s primary outcome of cardiovascular death, non-fatal myocardial infarction (MI), or hospitalization for unstable angina, heart failure, or resuscitated cardiac arrest, at a median follow-up of 3.37 (2.20–4.56) years.

Results:

Compared to SPECT/ECHO (N=5,627), CMR participants (N=313) were not different in key demographic factors but were more likely to have severe ischemia (57% vs 38%, P<0.001) and to be randomized (N=257, 82%, vs N=3,652, 65%, P<0.001). Ischemia severity (no/mild, moderate, severe) by CMR core laboratory was associated with cumulative 4-year event rates of all trial-specific endpoints, including the primary outcome (P=0.042), cardiovascular death/MI (P=0.041), and non-fatal MI (P=0.03), but SPECT/ECHO ischemia severity was not. No/mild, moderate, and severe ischemia by CMR were associated with 0%, 14%, and 23% 4-year primary outcome rates, respectively, compared to 18%, 15%, and 16%, by SPECT/ECHO. After adjustment for age, estimated glomerular filtration rate (EGFR), and diabetes, the association between ischemia extent and the primary endpoint differed by imaging modality, with each additional ischemic segment on CMR associated with a 13% increase in hazard (interaction p = 0.02). In participants assigned to initial conservative management who had no/mild ischemia on imaging, 4-year rates of invasive referral and coronary revascularization were lower in the CMR than SPECT/ECHO group (16.7% and 0%, respectively, for CMR, and 31% and 13.3%, for SPECT/ECHO).

Conclusion:

Ischemia severity by CMR had a stronger association with all ISCHEMIA trial endpoints compared with SPECT/ECHO.

Condensed abstract

In this post-hoc analysis of the ISCHEMIA trial, the severity of inducible ischemia by stress CMR was more strongly associated with the primary outcome and other study endpoints, than SPECT/ECHO. Multivariable analyses indicated that segmental ischemia extent was significantly associated with the primary outcome only when CMR was used. No or mild inducible ischemia by CMR core laboratory interpretation identified a subgroup with very low cumulative 4-year rates of all study endpoints. On the other hand, patients evaluated by SPECT/ECHO remained at risk of events despite core laboratory finding of no/mild ischemia. Among those assigned to initial conservative management, severe ischemia extent was associated with higher 4-year rates of coronary revascularization in both the CMR and SPECT/ECHO groups, but those with no/mild ischemia by CMR had a lower 4-year rate of coronary revascularization than those with no/mild ischemia by SPECT/ECHO.

Central Illustration

Introduction

The ISCHEMIA (International Study of Comparative Health Effectiveness with Medical and Invasive Approaches) trial observed no reduction of the primary trial composite of adverse cardiovascular outcomes by an early invasive strategy incremental to guideline-directed medical therapy (GDMT) in patients with moderate to severe myocardial ischemia during median follow up of 3.3 years.¹ The results of the ISCHEMIA trial demonstrated the remarkable safety and effectiveness of modern GDMT as part of an initial conservative management strategy for chronic coronary disease (CAD) patients with moderate or severe ischemia by stress testing. Stress cardiac magnetic resonance (CMR), with its technical capability to characterize myocardial perfusion, myocardial infarction (MI), cardiac function, and tissue contrast at high resolution in a single imaging session, has demonstrated robust diagnostic^2–4 and prognostic values^5,6 and reduced costs compared to invasive strategies⁷ in patients with CAD, including the potential to accurately characterize viable and ischemic myocardium. In the current study, we sought to evaluate the prognostic performance of stress CMR, characterizing ischemic burden and infarct scar, and in predicting cardiac events in comparison with alternative stress imaging modalities utilized for enrollment in the ISCHEMIA trial.

Methods

Study Population

ISCHEMIA enrolled and randomized patients with moderate or severe ischemia by noninvasive stress testing as determined locally by sites, 1:1 to an initial conservative strategy of GDMT alone versus a routine invasive strategy with revascularization, if feasible, added to GDMT. Definitions of ischemia severity of all modalities were defined a priori. Stress CMR defined moderate and severe left ventricular ischemia severity by 12.5–24% (based on 4–7 of 32 subsegments) and >25% (8 or more subsegments), respectively. These compared to 10–14% and ≥15% by stress nuclear, respectively, and 3 segments and ≥4 segments of stress induced wall motion abnormality by stress echo, respectively. In the conservative strategy group, patients with failure of GDMT (i.e., refractory angina or an acute ischemic event) could be referred for invasive coronary angiography (ICA) and consideration of revascularization. Each enrolling site determined eligibility based on moderate or severe ischemia, which was then reviewed independently by a blinded core laboratory. Stress testing was referred clinically and imaging protocols were recommended to trial sites but not mandated. Key exclusion criteria included EGFR <30 mL/min per 1.73 m², acute coronary syndrome ≤2 months, left ventricular ejection fraction (LVEF) <35%, New York Heart Association (NYHA) Class IV heart failure or uncontrolled baseline angina, or any coronary revascularization within 1 year before enrollment. Enrolled participants then underwent blinded coronary computed tomography angiography (CCTA) and were excluded for disqualifying coronary anatomy (≥50% left main stenosis or absence of any significant coronary stenosis). CCTA was not performed if eGFR was <60 mL/min or if core laboratory review showed no or mild ischemia. Participants were eligible for randomization if they had an obstructive epicardial arterial stenosis ≥50% on blinded CCTA. In the current study cohort of patients who underwent stress imaging for enrollment, group comparisons were made between patients who underwent a stress single photon emission computed tomography or a stress echocardiography (SPECT/ECHO) versus stress CMR.

Stress CMR and Other Functional Imaging Studies

CMR sites were screened for adequacy of the image quality by core laboratory approval of at least one stress CMR scan that included stress perfusion, cine, and late gadolinium enhancement (LGE) images. CMR sites were allowed to use either a 1.5T or 3T scanner and pulse sequences as provided by their MRI vendors optimized for patient care locally. All CMR studies at all sites were standardized to include vasodilating-stress first-pass CMR perfusion using fast-gradient echo images covering 3–4 short-axis slices every cardiac cycle, cine short-axis stack, and LGE images matching the cine stack. LGE imageing used inversion-recovery-prepared fast-gradient echo with a segmented read-out (standard sequence across MRI vendors at the time). All stress CMR studies were independently reviewed by one of 2 blinded core laboratories (Brigham and Women’s Hospital, Boston, Massachusetts and Montreal Heart Institute) according to published guidelines from the Society of Cardiovascular Magnetic Resonance.⁸ For each of the 16 segments of the American Heart Association (AHA) model, endocardial and epicardial subsegments were scored for presence of stress myocardial perfusion defect and subendocardial infarction. Inducible myocardial ischemia was reported when either endocardial or both subsegments had stress perfusion defect in the absence of subendocardial infarct or when both subsegments had a stress perfusion defect with subendocardial infarct confined to the endocardial subsegment only. Moderate and severe ischemia were defined by the presence of ≥4 and ≥8 ischemic subsegments, respectively. Infarct presence was defined by subendocardial LGE by CMR, irreversible rest/stress perfusion defect by SPECT, and rest wall motion abnormality by stress echocardiography. Unrecognized MI (UMI) was defined by infarct presence by imaging with no documented patient history of infarction. SPECT/ECHO were conducted using exercise, pharmacological stress, or both. For SPECT, segments were scored at rest and stress using a 5-point system (0=normal; 1=equivocal; 2=moderate, 3=severe, and 4=absence of tracer uptake) using the 17-segment model. An ischemic segment on nuclear imaging was defined by stress-rest change ≥1, with exception for segments with scores 4–3 or 3–2 which were considered infarcted. For ECHO, a segment was considered ischemic if there was hypokinesis, akinesis or dyskinesis after stress when resting wall motion was normal or if there was worsening from mild hypokinesis to severe hypokinesis, akinesis or dyskinesis after stress. A total of 16 segments were scored by the ECHO core laboratory, as segment 17 (the apical cap) was not scored. No/mild, moderate, and severe ischemia were defined by 0–2, 3, and ≥4 segments of ischemia myocardium, respectively, in ECHO; whereas 0–9%, 10–14%, and ≥15% ischemia myocardium, respectively, in SPECT. The endocardial and epicardial subsegments as scored by the CMR core laboratory were collapsed into the 16-segment model for alignment with SPECT and ECHO.

Coronary Computed Tomography Angiography

CCTA was acquired per-protocol for evaluation and exclusion of left main CAD and nonobstructive disease. All randomized patients with diagnostic evaluable CCTA received a Duke prognostic index score, which is a validated ordinal categorical scale of severity of CAD.⁹

Assessment of Study Endpoints

Patients were followed for a mean of 3.4 ± 1.4 years for the primary composite outcome of cardiovascular death, non-fatal MI, or hospitalization for unstable angina, heart failure, or resuscitated cardiac arrest. These outcomes were adjudicated by an independent committee, blinded to treatment assignment. The key secondary outcome was the composite of cardiovascular death and acute non-fatal MI. Acute non-fatal MI was further classified as spontaneous or procedure-related. The study protocol was approved by the institutional review board at the NYU Grossman School of Medicine (clinical coordinating center) and at each participating site. All enrolled patients provided written informed consent.

Statistical Analysis

Categorical variables were summarized as counts and percentages and compared using the chi-squared test. Continuous variables were assessed for normality using the Kolmogorov-Smirnov test. Normally distributed variables were reported as mean ± standard deviation and compared using the Student’s t-test; non-normally distributed variables were reported as median (interquartile range) and compared using the Wilcoxon rank-sum test.^10,11.

Annualized cumulative event rates were estimated by follow-up time in the CMR and SPECT/ECHO groups, stratified by ischemia severity or treatment strategy, using the Kaplan-Meier method for non-competing risk endpoints and the non-parametric estimator of the cumulative incidence function (CIF) for endpoints subject to competing risks¹⁰. Trends across ischemia severity were assessed by testing the regression slope against zero¹¹. Adjusted cause-specific Cox regression models included known risk-markers at baseline including patient age at randomization, EGFR, and diabetes. Cox regression models with interaction terms between modality and ischemia extent were used to assess whether the association of ischemia extent with the primary outcome differed between CMR and SPECT/ECHO. Model fit was compared with and without the interaction term using likelihood ratio tests. A sensitivity analysis repeated the model using robust (sandwich) variance estimators to account for potential instability in standard error estimates due to the smaller sample size and number of events in the CMR group. The proportional hazards assumption was evaluated using score tests of the scaled Schoenfeld residuals for each covariate and by visual inspection of residual plots against log time. All statistical analysis was performed by using commercially available software (SAS version 9.4; SAS Institute, Inc) and open-source software (R 4.2.0, R Core Team, 2023). A 2-sided value of P < 0.05 was considered significant.

RESULTS

Study Population

A total of 5,940 participants were enrolled after either a CMR or SPECT/ECHO. Exclusion criteria for enrolled participants who were not randomized are shown in Table 1, where more than 1 criterion may be possible for a participant. A higher proportion of participants enrolled after CMR were randomized compared to after SPECT/ECHO (82% vs 65%, P<0.001), largely due to a lower prevalence of insufficient ischemia (7% vs 15%, P<0.001) or non-obstructive CAD by CCTA (5.8% vs 12.6%, P<0.001) after CMR than after SPECT/ECHO. Unprotected left main CAD was less prevalent after CMR (3.5% vs 4.4%, P<0.001), as was consent withdrawal before randomization (1.3% vs 3.5%, P<0.001). The remaining 3,909 participants, 257 (6.6%) of whom had CMR and 3,652 (93.4%) SPECT/ECHO, were randomized. There was a higher concordance rate between the core lab and the enrolling sites for meeting the moderate/severe ischemia status in the CMR cohort than SPECT/ECHO (91% vs 85%, P=0.016). Participants who had CMR underwent vasodilating stress only, whereas those who underwent SPECT/ECHO underwent exercise stress (1,741, 51%), pharmacological stress (1,345, 40%), both (7.5%), or unknown (51, 1.5%).

Table 1:

ISCHEMIA Trial Enrolled Participants Excluded from Study Randomization, Stratified by Imaging Modalities

Exclusion Criterion	Stress CMR (N=313)	SPECT/ECHO (N=5,627)	P-value
Randomized to Trial, N (%)	257 (82.1%)	3,652 (64.9%)
Exclusion Criteria
Insufficient ischemia, N (%)	22, 7%	856, 15.2%	<0.001
Non-obstructive CAD by CCTA, N (%)	18, 5.8%	708, 12.6%	<0.001
Left main ≥50% by CCTA, N (%)	11, 3.5%	247, 4.4%	0.551
Withdrew consent, N (%)	4, 1.3%	196, 3.5%	0.052
Other reasons, N (%)	7, 2.2%	376, 6.7%	0.003

CAD = coronary artery disease, CMR = cardiac magnetic resonance imaging, CCTA = coronary computed tomography angiography.

^*Other criteria included other CCTA exclusions, intercurrent clinical event, other illness, and other exclusion. Enrolled participant may have > 1 reason for being excluded before randomization and therefore may be counted in >1 screen failure category.

Baseline demographic and imaging characteristics for randomized participants stratified by randomized stress imaging modality are shown in Table 2. All continuous variables violated the assumption of normality; therefore, they were reported as median (interquartile range) and compared using the Wilcoxon rank-sum test. Vasodilating agents used in stress CMR studies were dipyridamole (24%), adenosine (58%), regadenoson (3%), and unreported (14%), whereas nuclear/ECHO used dipyridamole (34%), adenosine (13%), regadenoson (16%), dobutamine (13%), and unreported (24%). The CMR cohort had a higher prevalence of female sex, White race, and former smokers. It was also more likely to have a prior history of MI (29% vs 22%, p=0.04) and CABG (10% vs 5%, p<0.001) than participants randomized after SPECT/ECHO. Compared with participants who underwent SPECT/ECHO, a great proportion of participants in the CMR cohort had evidence of an MI (49% vs 24%, P<0.001) and UMI on imaging (27% vs 14%, P<0.001). CMR participants also on average had a larger number of infarcted segments (1.5 vs 0.6, P<0.001) and ischemic segments (5.9 vs 3.9, P<0.001), lower baseline LVEF (59 ± 9% vs 62 ± 9%, P<0.001), and lower BMI (27 vs 29 kg/m², P<0.001) than SPECT/ECHO participants. Based on core laboratory reads, 9% of the CMR studies in randomized participants were graded as no/mild ischemia, compared with 14% of the SPECT/ECHO studies (P<0.001). More CMR patients had severe ischemia compared with SPECT/ECHO (59% vs 44%, P<0.0001). Female sex, Angina Class, cardiovascular risk factors, triple vessel coronary stenosis by CCTA, and modified Duke prognostic index on CCTA, were not different by stress modality groups.

Table 2:

Baseline Characteristics among ISCHEMIA Randomized Participants, Stratified by Stress CMR versus Stress SPECT or Stress Echocardiography (SPECT/ECHO)

Characteristic	Stress CMR (n=257)	Stress SPECT/ECHO (n=3,652)	P-value	SMD
Demographics
Age (years)	65 (59–71)	66 (59–72)	0.802	0.022
Female	29%	23%	0.270	0.135
Body Mass Index (kg/m2)	27 (25–30)	29 (26–32)	<0.001	0.317
Body Surface Area (m2)	1.92 (1.74–2.06)	1.94 (1.79–2.10)	0.032	0.215
Race			0.384	0.268
American Indian or Alaska Native	0%	0%
Asian	14%	14%
Native Hawaiian / Other Pacific Islander	0%	0%
Black or African American	1%	5%
White	85%	80%
Revascularization strategy (CON)	51%	50%	0.921	0.026
Cardiac History
MI	29%	22%	0.037	0.159
PCI	25%	24%	0.884	0.032
CABG	10%	5%	<0.001	0.215
CAD Risk Factors
Cholesterol (mg/dL)
Total	159.9 (136.7–183.4)	155.0 (131.5–187.1)	0.602	0.020
HDL	45.0 (38.0–55.8)	43.3 (36.7–52.0)	0.019	0.161
LDL	90.0 (67.5–110.0)	83.0 (63.0–111.0)	0.139	0.085
Triglycerides	123.0 (90.0–179.7)	124.0 (90.3–178.0)	0.996	0.026
Smoking			0.065	0.194
Never	34%	40%
Former	57%	47%
Current	9%	13%
Diabetes	43%	41%	0.801	0.043
Insulin-dependent	20%	27%	0.288	0.164
Hypertension	81%	78%	0.632	0.063
Family History Premature CAD	22%	28%	0.160	0.140
CCS class			0.644	0.162
0	25%	25%
1	24%	29%
2	47%	41%
3	4%	5%
4	0%	0%
SAQ Angina Frequency Score	90 (70–100)	90 (70–100)	0.694	0.075
Stress Imaging
Baseline HR (bpm)	69 (59.5–81)	71 (62–81)	0.713	0.134
Peak HR (bpm)	93 (78 –100)	131 (105.25–146)	<0.001	1.387
Baseline Systolic BP (mm Hg)	139 (124–150)	140 (125–160.50)	0.732	0.174
Baseline Diastolic BP (mm Hg)	80 (71–90)	80 (72–86)	0.831	0.186
Peak Systolic BP (mm Hg)	143 (132.75–152.5)	160 (140–180)	<0.001	0.619
Peak Diastolic BP (mm Hg)	79 (70–85)	80 (70–90)	0.465	0.170
Degree of Ischemia			<0.001	0.325
None	3%	6%
Mild	6%	8%
Moderate	32%	42%
Severe	59%	44%
LV end diastolic volume (ml)	135 (113–161.74)	94 (75–116)	<0.001	1.179
LV end systolic volume (ml)	51 (36–68)	37 (27–48)	<0.001	0.710
LVEF (%)	62 (56–67.95)	60 (55–65)	<0.001	0.307
LV mass (g)	123.5 (100–144)	N/A	N/A	N/A
Infarct presence	49%	24%	<0.001	0.528
UMI presence	27%	14%	<0.001	0.314
Number of infarcted segments (median, IQR)	0 (0–3)	0 (0–0)	<0.001	0.495
Number of ischemic segments (median, IQR)	5 (4–8)	4 (3–5)	<0.001	0.767
Coronary CT Angiography Findings
Number of vessels with coronary stenosis ≥70%			0.960	0.124
None	13%	16%
1 vessel	38%	40%
2 vessel	26%	26%
3 vessel	22%	18%
LAD stenosis ≥70%
Proximal	19%	17%	0.908	0.037
Any	64%	61%	0.859	0.051
Modified Duke prognostic index			0.985	0.181
2	0%	0%
3	6%	10%
4	33%	31%
5	32%	35%
6	27%	23%
7	3%	2%

Values were expressed in mean ± standard deviation, or n (%). BP = blood pressure; CABG = coronary artery bypass grafting; CAD = coronary artery disease; CCS = Canadian Cardiovascular Society; CON = early conservative treatment strategy; HDL = high density lipoprotein; HR = heart rate; LAD = left anterior descending coronary artery; LDL = low density lipoprotein; LV = left ventricular; LVEF = left ventricular ejection fraction; MI = myocardial infarction; PCI = percutaneous coronary intervention; SAQ = Seattle Angina Questionnaire. UMI = Unrecognized MI, which was based on imaging evidence of an MI in absence of any clinical history of infarction. CMR evidence of an MI is based on a presence of LGE consistent with an infarction. SPECT and ECHO evidence of MI were based on rest perfusion defect and rest wall motion abnormality, respectively. Modified Duke prognostic index as previously reported, categorizes CAD based on extent, location, and stenosis severity.

Study Endpoints

Follow-up was achieved in 3,867 (98.9%) of randomized participants at a median follow-up of 4.1 (IQR 3.0–5.0) years for CMR and 3.3 (IQR 2.2–4.5) years for SPECT/ECHO, and there were 39 and 529 primary events in the CMR (N = 257) and SPECT/ECHO groups (N = 3,652), respectively. The estimated 4-year event rate for the primary outcome was 16.7% (95% CI 11.8–21.7%) of participants who underwent CMR versus 15.8% (14.5–17.2%) who underwent SPECT/ECHO (P=NS). There were no significant differences between the components of the primary composite for CMR versus SPECT/ECHO including CV death (3.8%, 1.1–6.4% vs 4.8%, 4.0–5.6%), any primary MI (11.6%, 7.4–15.8% vs 10.7%, 9.5–11.8%), all-cause mortality (4.7%, 1.8–7.7% vs 7.1%, 6.1–8.1%), or spontaneous MI (9.7%, 5.7–13.6% vs 8.2%, 7.2–9.3%), P=NS for all.

Univariable and Multivariable Prognostic Association

Figure 1 illustrates a forest plot of the unadjusted association of segmental ischemia extent by stress CMR with all key endpoints. Segmental ischemia extent by CMR demonstrated a strong unadjusted association with the primary outcome, cardiovascular death or MI, cardiovascular death or spontaneous MI, and any acute non-fatal MI. On average, an increment of one ischemic segment was associated with 17%, 17%, and 22% increases in hazard of the primary outcome, cardiovascular death or MI, and MI, respectively.

Figure 1. Unadjusted Association of Segmental Ischemia by Stress CMR with Primary Outcome.

Forest plot of unadjusted association of stress CMR ischemia extent (per segment) with primary outcome. The hazard ratio is per ischemic segment on CMR core laboratory interpretation.

Univariable and multivariable analyses for the primary outcome stratified by imaging groups are shown in Tables 3 and 4, respectively. As shown in Table 3, univariable CMR predictors of outcomes included left ventricular end-systolic volume index, segmental ischemia extent, ischemia in >1 coronary territory, and LGE infarction. When these CMR measures was individually adjusted for patient age at randomization, EGFR, and diabetes, left ventricular end-diastolic and end-systolic volume indices, segmental ischemia extent, ischemia in >1 coronary territory, LGE presence, and UMI each remained a significant multivariable predictor. As shown in Table 4, after adjustment for age, EGFR, and diabetes, only ischemia in >1 territory and number of infarcted segments remain significant for SPECT/ECHO. Segmental ischemia extent by SPECT/ECHO was not significant as a univariable or a multivariable covariate. CMR detected a higher proportion of patients with UMI than SPECT/ECHO (27% vs 14%, P<0.001). UMI by CMR has borderline unadjusted association with increased primary outcome (HR 1.86, P=0.06), but did not modify the association of ischemia severity with primary outcome.

Table 3 –

Univariable and Multivariable Association of Stress CMR Characteristics with Primary Outcome (N=257)

Variable	Univariable		Adjusted*
	Hazard Ratio	P Value	Hazard Ratio	P Value
Stress CMR
LVEDVI	1.01 (1.00, 1.02)	0.051	1.01 (1.00, 1.02)	0.021
LVESVI	1.01 (1.00, 1.02)	0.029	1.01 (1.00, 1.03)	0.014
Number of ischemic segments	1.17 (1.06, 1.28)	0.002	1.16 (1.05, 1.27)	0.003
Ischemia in > 1 coronary territory	7.51 (1.03, 54.70)	0.047	7.71 (1.06, 56.22)	0.044
Ischemia in LAD territory	1.05 (0.50, 2.21)	0.900	1.09 (0.52, 2.30)	0.823
Infarct (LGE) presence	2.02 (1.05, 3.88)	0.036	2.02 (1.05, 3.90)	0.035
Number of infarcted segments based on endocardial LGE	1.11 (0.98, 1.24)	0.096	1.10 (0.98, 1.24)	0.119
UMI presence	1.86 (0.98, 3.55)	0.059	1.94 (1.01, 3.71)	0.046
Infarct (LGE) presence in a coronary territory unaffected by ischemia	2.02 (1.05, 3.88)	0.036	1.43 (0.59, 3.47)	0.430

^*Multivariable Cox regression adjusted by age at randomization, EGFR and diabetes at randomization.

LVEDVI = left ventricular end-diastolic volume index; LVESVI = left ventricular end-systolic volume index; MI = myocardial infarction; LAD = left anterior descending coronary artery; LGE = late gadolinium enhancement consistent with an infarction; UMI = Unrecognized MI. The primary outcome was cardiovascular death, MI or hospitalization for unstable angina, heart failure or resuscitated cardiac arrest.

Table 4 –

Univariable and Multivariable Association of Clinical and SPECT/ECHO Characteristics with Primary Outcome (N=3,652)

Variable	Univariable		Adjusted*
	Hazard Ratio	P Value	Hazard Ratio	P Value
SPECT / ECHO
LVEDVI	1.00 (1.00, 1.01)	0.721	1.00 (1.00, 1.01)	0.453
LVESVI	1.01 (1.00, 1.02)	0.212	1.01 (1.00, 1.02)	0.114
Number of ischemic segments	1.01 (0.96, 1.05)	0.761	1.01 (0.97, 1.06)	0.456
Ischemia in > 1 coronary territory	1.29 (1.03, 1.61)	0.025	1.28 (1.02, 1.60)	0.030
Ischemia in LAD territory	1.12 (0.93, 1.35)	0.214	1.05 (0.82, 1.34)	0.688
Infarct presence	1.23 (1.02, 1.49)	0.035	1.20 (0.99, 1.45)	0.064
Number of infarcted segments	1.08 (1.03, 1.14)	0.004	1.07 (1.02, 1.13)	0.010
UMI presence	1.28 (1.02, 1.61)	0.035	1.24 (0.98, 1.55)	0.071

^*Multivariable Cox regression is adjusted by age, estimated glomerular filtration rate and diabetes at randomization.

LVEDVI = left ventricular end-diastolic volume index; LVESVI = left ventricular end-systolic volume index; MI = myocardial infarction; LAD = left anterior descending coronary artery; MI_URG = no clinical history of infarction or ischemia but evidence of an infarction based on a resting perfusion defect on SPECT or rest wall motion abnormality on ECHO

4-year Cumulative Event Rates by Ischemia Grade

Figure 2 illustrates the cumulative event rates over the 4-year follow-up, grouped by ischemia severity. The cumulative 4-year event rates of the primary outcome were lower in patients with no/mild ischemia by CMR than those with moderate and severe ischemia (0% with no/mild ischemia vs 14.4% with moderate ischemia and vs 23.0% with severe ischemia, P=0.042 across groups). The 4-year event rates of cardiovascular death/MI were also observed to be statistically different across the groups with no/mild, moderate, and severe ischemia by CMR (0% vs 11.8% vs 21.9%, P= 0.041 across groups). The 4-year event rates of any acute non-fatal MI were observed to be different across the groups with no/mild, moderate, and severe ischemia by CMR (0% with no/mild ischemia vs 8.5% vs 17.7%, P=0.030 across groups). By comparison, these strengths of association for these study endpoints were not significant in participants who underwent stress SPECT/ECHO. The 4-year event rates of the primary outcome were not different across the groups with no/mild, moderate, and severe ischemia by SPECT/ECHO (17.5% with no/mild ischemia vs 15.3% vs 15.9%, P=0.555 across groups). The 4-year event rates of cardiovascular death/MI were also not different across the groups with no/mild, moderate, and severe ischemia by SPECT/ECHO (15.6% with no/mild ischemia vs 13% vs 14.4%, P=0.942 across groups). Indeed, there appeared to be a trend toward inverse association between ischemia severity by SPECT/ECHO and 4-year event rates of all-cause mortality (10.1% with no/mild ischemia vs 6.6% with moderate ischemia vs 6.7% with severe ischemia, P=0.087 across groups). Amongst patients reported to have no/mild ischemia, those randomized after CMR did not experience any 4-year primary composite events which compared to 17.5% in those randomized after SPECT/ECHO (statistical comparison not possible due to absence of event in the CMR group). As previously reported, ischemia severity by SPECT/ECHO did not demonstrate association with the primary outcome in either conservative or early invasive arm in adjusted clinical models.¹² Figures 3a, 3b, and 3c demonstrated cumulative incidence curves by ischemia grades of stress CMR and SPECT/ECHO for cumulative 4-year cumulative events in primary outcome, CV death/nonfatal MI, and acute non-fatal MI, respectively. Ischemia grade by CMR demonstrated a trend association with each of these 3 key study endpoints. While the Fine-Gray test did not reach statistical significance, visual inspection of the cumulative incidence curves showed separation between ischemia severity grades, with no events observed in the no/mild group, albeit limited by the small number of patients in this group. In contrast, ischemia grade by SPECT/ECHO was not significant with any of the key endpoint. The primary results of ISCHEMIA trial showed higher earlier risk in the invasive group due to a higher incidence of peri-procedural MI.¹ A similar early curve crossing is evident in Figure 3a in the panel for SPECT/ECHO when comparing the no/mild versus moderate ischemia curves.

Figure 2. Adverse event Rates by Ischemia Extent.

Cumulative 4-year event rates grouped by ischemia severity of imaging modalities.

Figure 3. Cumulative Incidence Curves of Adverse Cardiovascular Events by Stress CMR and Stress SPECT/ECHO.

Figure 3a, 3b, and 3c illustrated cumulative incidence curves of 4-year events of, primary outcome, cardiovascular death or non-fatal MI, and non-fatal MI alone, respectively, by ischemia grade by stress CMR and stress SPECT/ECHO. P-values were generated using Fine-Gray’s test which accounted for competing risks. Patients in the stress CMR no/mild ischemia subgroup observed no adverse events and were not used in the Fine-Gray’s test.

Referral to Invasive Testing in Participants Assigned Initial Conservative Management

Invasive coronary angiography was performed in 96% of the invasive group and 26% of the conservative group in the entire cohort studied here. Coronary revascularization was performed in 80% of the invasive group and 21% of the conservative strategy group (15% before sustaining a primary outcome and 6% after). Figure 3 illustrates the 4-year cumulative rates of invasive coronary angiography and revascularization in participants randomized to the conservative group, categorized based on the type of imaging stress test performed for study qualification. In both CMR and SPECT/ECHO groups, ischemia severity was not associated with the 4-year rates of performing invasive angiography. However, ischemia severity was associated with higher rates of coronary revascularization procedures in both the CMR and SPECT/ECHO groups (P=0.039 and 0.016, respectively). Participants in the CMR group with no/mild ischemia underwent revascularization at 0% at 4-years which compared to 13.3% (95% CI: 8.9%–17.6%) for participants with no/mild ischemia in the SPECT/ECHO group.

Imaging Modality and the Association Between Segmental Ischemia Extent and the Primary Outcome

Table 5 illustrates that imaging modality modified the association between ischemia extent and the primary outcome. In the whole cohort that combined CMR and SPECT/ECHO participants, there was no significant association between the number of ischemic segments and the primary outcome. Adding the interaction term between imaging modality and ischemia extent significantly improved model fit compared to the non-interaction model on the primary outcome (Likelihood Ratio Test: χ² = 7.50, p = 0.006), supporting a differential prognostic effect of ischemia severity by imaging modality, and the interaction was found statistically significant in both the raw and the covariate adjusted model. The association of segmental ischemia extent with the primary outcome was conditional on the status of the imaging modality as CMR (P=0.005 for effect modification). Amongst participants with no ischemia, CMR participants were 66% less likely to experience the primary outcome, than SPECT/ECHO participants (HR 0.34, 0.16–0.76, P=0.008). Per every additional CMR ischemia segment, the hazard rate for primary outcome increased by 16% (HR 1.16, 1.05–1.29, P=0.005). Adjusted for age, EGFR, and diabetes, the primary outcome hazard rate increased by 13% per every additional CMR ischemia segment (HR 1.13, 1.02–1.25, P=0.021). In a sensitivity analysis using robust variance estimators, the interaction term in the adjusted model remained statistically significant (HR 1.13, 1.01–1.26, P=0.028), supporting the robustness of our findings despite the smaller event count in the CMR group. Assessment of proportional hazards showed consistent results across model specifications. The assumption was violated for the segmental ischemia extent (p = 0.04), and the hazard accordingly should be interpreted as a time-averaged association, whereas modality and its interaction with the segmental ischemia extent did not demonstrate violations.

Table 5 –

Association between ischemia extent and the primary outcome, adjusted for stress imaging modality

CMR Variables
	HR (95% CI)	P-value
Ischemia Extent (per segment)	1.01 (0.96, 1.05)	0.743
Imaging Modality
Stress SPECT/ECHO	---	---
Stress CMR	0.34 (0.16, 0.76)**	0.008
Interaction between Ischemia Extent*Imaging Modality
Ischemia Extent*Stress CMR	1.16 (1.05, 1.29)	0.005
Adjusted for Age at randomization, EGFR, and Diabetes
	HR (95% CI)	P-value
Ischemia Extent (per segment)	1.01 (0.97, 1.06)	0.441
Imaging modality
Stress SPECT/ECHO	---	---
Stress CMR	0.40 (0.19, 0.86)**	0.019
Interaction between Ischemia Extent*Imaging Modality
Ischemia Extent*Stress CMR	1.13 (1.02, 1.25)	0.021

HR = Hazard Ratio; CI = Confidence Interval

^**This hazard ratio compares the risk associated with the absence of ischemia on CMR and on SPECT/ECHO. The interaction term demonstrates that for each additional segment that is ischemic, risk is greater if that ischemic segment was determined based on CMR rather than SPECT/ECHO.

Discussion

This post-hoc analysis evaluated the prognostic performance of CMR vs SPECT/ECHO for stable CAD patients with site-determined moderate or severe ischemia who were randomized in the ISCHEMIA trial. Our current study is unique as it compared the performance characteristics of stress CMR vs. SPECT/ECHO when used as guiding tools towards the outcome benefits of conservative vs. early invasive strategies. Despite the small sample size of the CMR group, we observed that the severity of inducible ischemia by CMR demonstrated stronger univariable and multivariable association with the primary outcome and other study endpoints, compared with SPECT/ECHO. Multivariable analyses indicated that segmental ischemia extent had a stronger association with the primary outcome when the enrolling imaging modality was CMR. An adjusted hazard ratio of 1.15 per ischemic segment for the primary outcome was observed in the CMR group, while no significant association was observed in the SPECT/ECHO group. Patients with no/mild ischemia by CMR core lab interpretation experienced no primary outcome at 4 years, in contrast to patients with no/mild ischemia by SPECT/ECHO (0% vs. 17.5%). In patients assigned to initial conservative management, a larger ischemia extent was associated with higher 4-year rates of coronary revascularization in both stress CMR and SPECT/ECHO cohorts, but those with no/mild ischemia by CMR appeared to have a lower 4-year rate of coronary revascularization than those with no/mild ischemia by SPECT/ECHO (0% by CMR vs 13.3%, 95%CI 8.9–17.6%, by SPECT/ECHO). Potential reasons for the difference in revascularization rates could include false negative SPECT/ECHO results in patients with symptomatic disease leading to referral to angiography and revascularization. Alternatively, it is possible that CMR was better able to identify patients who would not fail GDMT than SPECT/ECHO.

The ISCHEMIA trial by study design captured a subset of patients with established CAD who were at substantial clinical risk. We observed that CMR differentiated risk in this CAD cohort better than SPECT/ECHO. Given that CMR exhibited better prognostic prediction than SPECT/ECHO, we suggest that CMR should be prioritized over SPECT/ECHO where there is appropriate imaging equipment and interpretative expertise.

Stress perfusion CMR has several technical merits that may explain these findings. First, stress CMR captures regions of relative hypoperfusion, infarction, and regional wall thickening in matching scan planes at high temporal and spatial resolution. These features allow determination of the size and severity of regional ischemia and infarction, which reflect the range of pathophysiologic states in the spectrum of CAD which are associated with risks of adverse outcome. High spatial resolution also allows CMR to capture ischemia due to multivessel CAD or in smaller hearts with high diagnostic certainty. CMR could also be more sensitive in detecting UMI. Second, relative freedom from artifacts caused by tissue attenuation or large body habitus increases the diagnostic specificity across the spectrum of CAD. The observations from our current study are concordant with the recent GadaCAD2 trial where CMR demonstrated higher diagnostic specificity and accuracy than SPECT.⁴

We observed that interpretation of ischemia extent between sites and the core laboratory was more consistent by CMR than SPECT/ECHO, leading to less exclusion from randomization after CMR due to core laboratory-determined insufficient ischemia. Compared to SPECT/ECHO, participants by CMR were also less likely to be excluded due to non-obstructive disease on CCTA. Potential factors leading to these observations include a small number of CMR sites (31) with potentially higher and more uniform imaging equipment or technical expertise and selective referral of patients with a higher CAD burden for trial entry. On the other hand, better image quality of CMR based on higher spatial/temporal resolutions and matching perfusion and infarction imaging may have contributed to a higher site/core laboratory agreement than SPECT/ECHO. These features, in combination with CMR pulse sequences that yield an excellent tissue contrast needed to detect subendocardial perfusion defects (for which the epicardial and mid-myocardial perfusion can be used as a reference). The prevailing practice by the enrolling sites relied on qualitatively determined ischemia extent on SPECT which are inherently subjective and may have caused disagreement with the core laboratory due to variable image quality, readers’ experience, choice of post-processing and contouring methods, and use of quantitative software at the core laboratory. In the multicenter Fast Myocardial Perfusion Imaging with NExt generation SPECT (REFINE SPECT) Registry, visual interpretation of SPECT demonstrated a mismatch rate as high as 50% compared to quantitative analysis. Increased adverse outcome events were observed in both mismatched groups (3.4 to 3.6% annual event rate of death, MI, unstable angina, and late revascularization).¹³ In the current cohort, the choice of functional stress was left to the discretion of the participating site to conform to real-world practice. Prior studies have indicated that exercise and pharmacological stress had similar diagnostic accuracy^14,15, thus the difference in the chosen stress method does not appear to explain the difference in prognostic values observed between CMR and SPECT/ECHO in the current cohort.

Lawler et al reported that exposure to low-dose ionizing radiation was associated with increased cancer risk in both sexes, with women experienced significantly higher cancer risk than men.¹⁶ Compared to SPECT, CMR has the advantages of not using ionizing radiation and less artifacts due to breast attenuation in women.¹⁷ Indeed, another analysis of cancer risk by low-dose ionizing radiation was performed in the ISCHEMIA trial and found a 4-fold increase in malignancy deaths in patients with 3 radiation exposure examinations vs one or less exposure.¹⁸ These data highlight the need for non-invasive cardiac testing in men and particularly in women that minimizes iatrogenic radiation and both unnecessary diagnostic ICA and revascularization, especially in those at low to intermediate pre-test risk of CAD. Finally, CMR also is more sensitive in detecting UMI which is a high-risk finding and an opportunity to institute appropriate medical therapy.¹⁹

Given the extensive evidence of high diagnostic and prognostic values of CMR from clinical registries⁵, randomized trials^2,20, and cost-effectiveness studies modeled based on a wider clinical adoption of CMR²¹, our study adds to the evidence supporting the first line use of stress CMR. We propose that randomized trials assessing patient outcomes and cost-effectiveness of a stress CMR ischemia-guided strategy against CCTA, is needed to establish the optimal strategies across broad clinical risks. In addition to MR-INFORM²⁰, more trials evaluating whether CMR-based ischemia can guide early invasive intervention towards outcome or cost benefits, are warranted”.

STUDY LIMITATIONS

First, the CMR sample size was too small to permit analysis of whether CMR can guide decisions about utilizing an invasive approach in addition to medical therapy. Small sample size for CMR also limited the number of clinical covariates that could be used for statistical adjustment. The current cohort also cannot address the association of ischemia by CMR with cardiovascular mortality given the small number of this event. Second, the ISCHEMIA trial targeted moderate or severe ischemia at sites, resulting in a lower number of participants with no/mild ischemia compared to clinical practice and limiting our ability to detect associations across the spectrum of ischemia severity. The ISCHEMIA design also prohibited randomized comparison between CMR and SPECT/ECHO, and the observed findings may be subject to confounding bias. Finally, local referral pattern to CMR may have selected a higher risk group with more prior coronary bypass surgery. Local imaging equipment and interpretive expertise may also have affected results of the CMR and SPECT/ECHO comparisons. Future studies should further address the key observed results across patients’ race and sex.

CONCLUSIONS

This post-hoc analysis of the ISCHEMIA trial demonstrated strong prognostic utility of stress CMR in CAD patients with moderate or severe ischemia detected by sites. Agreement between sites and the core laboratory on the presence of moderate or severe ischemia was higher for the stress CMR than for SPECT/ECHO. Stress CMR is a robust method that can discriminate cardiac risk, and is a powerful radiation-free gatekeeper by identifying patients with known coronary stenoses but no need for revascularization. The evidence from this post-hoc analysis of the ISCHEMIA trial supports increased adoption of stress CMR over SPECT/ECHO in patients with CAD at moderate-high risk of cardiac events. In addition, future trials comparing patient outcomes and cost-effectiveness of functional stress CMR guided care against CCTA-guided care in stable chest pain patients are warranted.

Figure 4. Invasive Referrals and Revascularization by Imaging Modalities.

Cumulative 4-year event rates of invasive referrals and revascularization in patients assigned to initial conservative management, stratified by ischemia severity of imaging modalities.

CLINICAL PERSPECTIVES.

This post-hoc analysis of ISCHEMIA demonstrated strong prognostic utility of stress CMR in symptomatic CAD patients with moderate or severe ischemia identified by sites. Stress CMR had a higher rate of patient retention for trial randomization and a higher interpretative agreement between the participating sites and the core laboratory, than stress SPECT/ECHO. Despite its small sample size relative to SPECT/ECHO, ischemia grade in the stress CMR group demonstrated significant discrimination of cumulative 4-year adverse events, and appeared a powerful radiation-free gatekeeper by identifying CAD patients with no need for revascularization. The present post-hoc analysis of the ISCHEMIA trial provides increment evidence supporting an increased adoption of stress CMR over SPECT/ECHO in CAD patients at moderate-high risk of cardiac events.

List of Abbreviations

ISCHEMIA

International Study of Comparative Health Effectiveness with Medical and Invasive Approaches

CMR

cardiac magnetic resonance imaging

SPECT

single photo emission computed tomography

ECHO

echocardiography

CCTA

coronary computed tomographic angiography

GDMT

guideline-directed medical therapy

ICA

invasive coronary angiography

LGE

late gadolinium enhancement

MI

myocardial infarction

CABG

coronary arterial bypass grafting

EGFR

estimated glomerular filtration rate

Data availability

Data will be submitted and available via the NHLBI according to their guidelines.