Improved Cardiac Risk Assessment with Non-Invasive measures of Coronary flow reserve

Abstract

Background

Impaired vasodilator function is an early manifestation of coronary artery disease and may precede angiographic stenosis. It is unknown whether non-invasive assessment of coronary vasodilator function in patients with suspected or known coronary artery disease (CAD) carries incremental prognostic significance.

Methods and Results

2783 consecutive patients referred for rest/stress PET were followed for a median of 1.4 years (inter-quartile range: 0.7–3.2 years). The extent and severity of perfusion abnormalities were quantified by visual evaluation of myocardial perfusion images (MPI). Rest and stress myocardial blood flow (MBF) were calculated using factor analysis and a 2-compartment kinetic model, and were used to compute coronary flow reserve (CFR=stress/rest MBF). The primary endpoint was cardiac death. Overall 3-year cardiac mortality was 8.0%. The lowest tertile of CFR (<1.5) was associated with a 5.6-fold increase in the risk of cardiac death (95%CI 2.5–12.4, p<0.0001) compared to the highest tertile. Incorporation of CFR into cardiac death risk assessment models resulted in an increase in the c-index from 0.82 (95%CI 0.78–0.86) to 0.84 (95%CI 0.80–0.87, p=0.02) and in a net reclassification improvement (NRI) of 0.098 (95%CI 0.025–0.180). Addition of CFR resulted in correct reclassification of 34.8% of intermediate risk patients (NRI=0.487, 95%CI 0.262–0.731). Corresponding improvements in risk assessment for mortality from any cause were also demonstrated.

Conclusions

Non-invasive quantitative assessment of coronary vasodilator function using PET is a powerful, independent predictor of cardiac mortality in patients with known or suspected CAD and provides meaningful incremental risk stratification over clinical and gated MPI variables.

Introduction

Traditional approaches for cardiac risk assessment in coronary artery disease (CAD) include information about left ventricular (LV) function, the angiographic extent and severity of epicardial coronary stenoses, and the magnitude of inducible myocardial ischemia by non-invasive testing (1). Although very powerful, these traditional risk markers do not together fully account for the risk of cardiac death, especially among high risk subgroups (2,3). This suggests that other mechanisms may contribute to clinical risk. One such mechanism may involve the adverse effects of atherosclerosis and its risk factors on coronary epicardial and microcirculatory function, thereby increasing the potential for ischemic and atherothrombotic complications.

Coronary circulatory dysfunction is present in the earliest stages of atherogenesis and precedes significant angiographic stenosis (4,5), identifiable perfusion deficits and wall motion abnormalities with traditional imaging methods (6–8). Furthermore, abnormalities in vascular function identified by specialized invasive methods have been shown to identify patients at increased risk for adverse cardiovascular events (9–11), and to improve selection of patients for revascularization (12). Coronary vasodilator function can also be assessed non-invasively using Positron Emission Tomography (PET).

This study was designed to test the hypothesis that among patients referred for cardiac testing, abnormalities of coronary vasodilator function add incremental value for the identification of those at risk of cardiac death beyond clinical risk factors and traditional semi-quantitative assessment of myocardial ischemia and left ventricular function.

Methods

Study Population

All patients referred for rest/stress cardiac PET at the Brigham & Women’s Hospital (Boston, MA) between January 1, 2006 and June 30, 2010 were included in this study, excluding those whose images were missing or uninterpretable due to poor image quality. In cases of repeat evaluations during the study period, only the earliest evaluable study was included. The study was approved by the Partners Healthcare Institutional Review Board and conducted in accordance with institutional guidelines.

Risk Factor Assessment

Demographic factors and key elements of the patients’ history including risk factors and medication use were ascertained at the time of the study by patient interview and review of medical records.

Positron Emission Tomographic Imaging

Patients were studied using a whole body PET-CT scanner (Discovery RX or STE LightSpeed 64, GE Healthcare, Milwaukee, WI) after an overnight fast. Patients refrained from caffeine and methylxanthine containing substances and drugs for 24 hours prior to their scans. Myocardial blood flow (MBF) was measured during rest and peak stress using ⁸²Rubidium as a perfusion tracer, as described previously (13). Briefly, after transmission imaging and beginning with the intravenous bolus administration of ⁸²Rubidium (1,480–2,200 MBq), list mode images were acquired for seven minutes. Then, a standard intravenous infusion of dipyridamole, adenosine, regadenoson or dobutamine was given. At peak stress, a second dose of ⁸²Rubidium was injected and images were recorded in the same manner. The average radiation exposure per study was 4.6 mSv (14). Heart rate, blood pressure, and 12-lead electrocardiogram were recorded at baseline and every minute during and after pharmacological stress.

Image Analysis

Semiquantitative Analysis of Myocardial Perfusion

Semi-quantitative 17-segment visual interpretation of the gated myocardial perfusion images was performed by experienced observers using a standard 5-point scoring system (15). Summed rest (SRS) and stress scores (SSS) were calculated as the sum of individual segmental scores on the respective images, and their difference was recorded as summed difference score (SDS). For each of these variables, higher scores reflect larger areas of myocardial ischemia and/or scar. Summed rest, stress and difference scores were converted into percentages of total myocardium by division the maximum possible score.

Left Ventricular Systolic Function

Rest and stress LV ejection fraction (LVEF) were calculated from gated myocardial perfusion images using commercially available software. Left ventricular ejection fraction reserve was considered present when LVEF increased from rest to stress.

Quantitative Myocardial Blood Flow and Flow Reserve

Absolute MBF (in ml/g/min) was computed from the dynamic rest and stress imaging series using commercially available software (Corridor4DM; Ann Arbor, Michigan) and previously validated methods (13,16). Automated factor analysis was used to generate blood pool (arterial input function) and tissue time-activity curves (17). Regional and global rest and peak stress MBF were calculated by fitting the ⁸²Rubidium time-activity curves to a two-compartment tracer kinetic model as described previously (16). Per-patient global coronary flow reserve (CFR) was calculated as the ratio of absolute MBF at stress over rest for the entire left ventricle. Quantitation of MBF was performed by four operators. The intra-class correlation coefficient for CFR among these four readers was 0.94 (95%CI 0.88–0.98), indicating excellent reproducibility.

Assessment of Outcomes

The primary outcome was death from any cardiac cause. Patients who died from non-cardiac causes were censored. Vital status of all patients was ascertained by integrating data from the Social Security Death Index, the National Death Index and the Partners Healthcare Research Patient Data Registry. Cause of death was determined by blinded adjudication of hospital records and death certificates. Early revascularization (within 90 days) was ascertained from the Partners Healthcare Research Patient Data Registry and hospital records. Mortality from any cause was used as a secondary endpoint.

Statistical Analysis

Statistical significance was assessed using Wilcoxon tests, Fisher exact and chi-square tests for continuous, dichotomous and categorical variables, respectively. Two sided p-values < 0.05 were considered significant. All statistical analyses were performed using SAS 9.2 (SAS Institute Inc., Cary, NC).

Multivariable Modeling

The Cox proportional hazards model was used to assess the impact of CFR on cardiac mortality after controlling for the effects of critical covariates. A series of models were developed starting with one containing only clinical covariates (i.e. age, gender, hypertension, dyslipidemia, diabetes mellitus, family history of CAD, tobacco use, history of CAD (including recent or prior myocardial infarction and revascularization), body mass index, chest pain, dyspnea and early revascularization). Rest LVEF, the combined extent of myocardial scar and ischemia, stress-induced LVEF augmentation and CFR (in tertiles) were then sequentially incorporated into the model. In order to investigate the effects of absolute peak stress MBF we generated an additional model containing absolute stress MBF instead of CFR. The models were examined for the validity of the proportional hazards assumption and additive value, taking care to avoid over-fitting. Survival was plotted using direct adjusted survival probabilities (18) from the Cox survival model.

To assess for biases introduced by early revascularization, analyses were repeated censoring all patients who underwent early revascularization (19). In an exploratory analysis, we considered the effect of any revascularization, including those >90 days after the PET scan, as a time-dependent covariate.

Assessment of Incremental Value

Incremental prognostic value of CFR was assessed with the likelihood ratio test to determine the improvement in prediction power of each sequential Cox model. Model calibration was evaluated with the Nam-D’Agostino χ² statistic (20). The c-index was calculated for each model (21) with comparisons using the method of Antolini and colleagues (22). The potential impact of CFR on risk stratification was assessed by net reclassification improvement (NRI) (23)at 1.5-years using threshold annual rates of cardiac mortality of 1% and 3% based on ACC/AHA guidelines for management of chronic stable angina (1).

Exploratory Analyses

The effect size of CFR on cardiac mortality was evaluated in a series of subgroups by constructing Cox proportional hazard models for each subgroup containing only CFR tertiles as predictor variables (i.e. unadjusted models). Effect modification was explored using Cox models incorporating interaction terms between key covariates and CFR.

To explore the relative contribution of epicardial stenoses and microvascular dysfunction, findings on coronary angiography within 90 days of PET were correlated to CFR tertile among patients with no evidence of scar or ischemia by visual PET evaluation and without history of coronary artery bypass graft surgery or intervening MI. Lesions greater than 70% stenosis were considered significant. Angiographic disease was categorized as non-obstructive or normal (<50% stenosis), 1-, 2- or 3-vessel disease.

Results

Patient Characteristics

A total of 2783 of 3404 (81.8%) consecutive patients met inclusion criteria during the study period and followed for a median of 1.4 years (IQR: 0.7–3.2 years). Baseline characteristics are given in table 1. PET scans represented 19.8% of all stress imaging scans performed at our institution during the time period of this study. The remainder of stress imaging tests were performed with SPECT (54.6%) or echocardiographic imaging (25.7%). The risk profiles of patients referred for these tests were comparable using the Duke clinical score: 49.2 ± 32.3 (mean ± SD) for PET, 47.8 ± 31.6 for SPECT and 37.9 ± 30.9 for echocardiography (24).

Table 1.

Patient Characteristics

Variable	No Cardiac Death (n=2646)	Cardiac Death (n=137)	All Patients (n=2783)	p-Value
Demographics
Age (y)	64.3 [55.8–74.6]	74.6 [63.7–81.4]	64.8 [56.1–75.2]	<0.0001
Male gender	1242 (46.9)	91 (66.4)	1333 (47.9)	<0.0001
Hispanic	300 (11.3)	3 (2.2)	303 (10.9)	0.0002
Race				0.06
White	1662 (62.8)	100 (73.0)	1762 (63.3)
Black	436 (16.5)	17 (12.4)	453 (16.3)
Other/Unknown	548 (20.7)	20 (14.6)	568 (20.4)
COPD	283 (10.7)	33 (24.1)	316 (11.4)	<0.0001
Risk Factors
BMI (kg/m2)	29 [25.1–34.7]	26.4 [23–30.4]	28.8 [25.1–34.4]	<0.0001
BMI ≥ 30 kg/m2	1167 (44.1)	37 (27.0)	1204 (43.3)	<0.0001
Hypertension	2154 (81.4)	117 (85.4)	2271 (81.6)	0.26
Dyslipidemia	1756 (66.4)	97 (70.8)	1853 (66.6)	0.31
Diabetes	940 (35.5)	68 (49.6)	1008 (36.2)	0.001
Family history of CAD	723 (27.3)	33 (24.1)	756 (27.2)	0.43
Tobacco Use	276 (10.4)	18 (13.1)	294 (10.6)	0.06
Dialysis	118 (4.5)	15 (11.0)	133 (4.8)	0.003
Medications
Aspirin	1613 (61.0)	92 (67.2)	1705 (61.3)	0.15
B-adrenergic blockers	1655 (62.6)	107 (78.1)	1762 (63.3)	0.0002
Cholesterol-lowering agents	1624 (61.4)	99 (72.3)	1723 (61.9)	0.01
Insulin	407 (15.4)	36 (26.3)	443 (15.9)	0.002
Oral hypoglycemic agents	274 (10.4)	10 (7.3)	284 (10.2)	0.31
Ca-channel blockers	594 (22.5)	27 (19.7)	621 (22.3)	0.53
ACE inhibitors	1040 (39.3)	64 (46.7)	1104 (39.7)	0.09
Nitrates	324 (12.2)	32 (23.4)	356 (12.8)	0.0005
Diuretics	948 (35.8)	77 (56.2)	1025 (36.8)	<0.0001
Indications
Chest Pain	1268 (47.9)	38 (27.7)	1306 (46.9)	<0.0001
Dyspnea	791 (29.9)	61 (44.5)	852 (30.6)	0.0006
Post-MI	223 (8.4)	28 (20.4)	251 (9.0)	<0.0001
Pre-operative	380 (14.4)	28 (20.4)	408 (14.7)	0.06
Cardiovascular History
Any prior CAD	1070 (40.4)	105 (76.6)	1175 (42.2)	<0.0001
Recent MI (≤30 days)	279 (10.5)	37 (27.0)	316 (11.4)	<0.0001
Remote MI (>30 days)	462 (17.5)	49 (35.8)	511 (18.4)	<0.0001
Prior PCI	565 (21.4)	47 (34.3)	612 (22.0)	0.0007
Prior CABG	317 (12.0)	52 (38.0)	369 (13.3)	0.0007
Congestive Heart Failure	115 (4.4)	28 (20.4)	143 (5.1)	<0.0001
Cerebrovascular Disease	158 (6.0)	12 (8.8)	170 (6.1)	0.2
Peripheral Vascular Disease	157 (5.9)	24 (17.5)	181 (6.5)	<0.0001
Early Revascularization (≤90 days post-PET)	217 (8.2)	18 (13.1)	235 (8.4)	0.0564
Stress Protocol				<0.0001
Adenosine	205 (7.8)	11 (8.0)	216 (7.8)
Dipyridamole	1280 (48.4)	96 (70.1)	1376 (49.4)
Dobutamine	115 (4.4)	9 (6.6)	124 (4.5)
Regadenoson	1046 (39.5)	21 (15.3)	1067 (38.3)
Imaging Parameters
Rest LVEF(%)	59 [49–66]	39 [27–54]	58 [48–66]	<0.0001
Stress-induced ↑LVEF	2060 (77.9)	87 (63.5)	2147 (77.2)	0.0002
Scar + Ischemic Myocardium (%)	0 [0–10.3]	16.2 [4.4–35.3]	0 [0–10.3]	<0.0001
Ischemic Myocardium(%)	0 [0–4.4]	4.4 [0–10.3]	0 [0–5.9]	<0.0001
Global CFR	1.76 [1.37–2.25]	1.31 [1.12–1.56]	1.73 [1.34–2.22]	<0.0001
Stress Global MBF (ml/g/min)	1.82 [1.25–2.54]	1.26 [0.92–1.82]	1.80[1.23–2.52]	<0.0001
Rest Global MBF (ml/g/min)	1.01 [0.79–1.33]	1.00[0.75–1.34]	1.01 [0.78–1.33]	0.41

Continuous variables are presented as median (inter-quartile range). Dichotomous variables are presented as number (%). Patients whose LVEF at stress was greater than that at rest were considered to have positive stress-induced increase in LVEF. BMI = body mass index. ACE = angiotensin converting enzyme. MI = myocardial infarction. CAD = coronary artery disease. PCI = percutaneous coronary intervention. CABG = coronary artery bypass graft. LVEF = left ventricular ejection fraction. CFR = coronary flow reserve. MBF = myocardial blood flow.

The most common indications for testing were evaluation for chest pain, dyspnea, or their combination. Approximately half of all studies were normal by semi-quantitative visual analysis. Scar and ischemia were more common in patients who experienced cardiac mortality. However, scar rather than ischemia accounted for a larger portion of defects among patients who died.

At rest, myocardial blood flow was comparable across tertiles of CFR (1.17, 1.14 and 1.01 mL/min/g for patients in high, intermediate, and low tertiles of CFR, respectively, p<0.0001; Figure 1). During peak stress, there was a significant stepwise decline in myocardial blood flow with decreasing tertiles of CFR (2.61, 1.98, and 1.35 mL/min/g for patients in high, intermediate, and low tertiles of CFR, respectively, p<0.0001), indicating that reduced CFR reflected primarily a reduction in coronary vasodilator function.

Figure 1. Myocardial Blood Flow by CFRTertile.

Box plots of rest and stress blood flow distributions of the population divided into tertiles by coronary flow reserve (CFR). Overall median blood flow for rest and stress are indicated by the dashed horizontal lines. Although the blood flows significantly vary across tertiles at both rest and stress, the differences are more pronounced at stress and are small in magnitude at rest.

Patient Outcomes

Mortality from any cause occurred in 279 (10.3%) patients, of which 137 (49.1%) were due to cardiac causes (Table 2). Three-year cardiac mortality was 8.0%. Compared to patients without cardiac death, those who experienced cardiac death were older, more likely to be male with diabetes, had a lower body mass index, more likely to be referred for dyspnea evaluation, have prior CAD, lower rest LVEF, and larger abnormalities on PET scans (Table 1).

Table 2.

Causes of Death

Cause of Death	All Patients (n=2783)	Upper CFR Tertile	Middle CFR Tertile	Lower CFR Tertile	p-Value
Cardiac	137 (4.9%)	7 (0.8%)	37 (4.0%)	93 (10.0%)	<0.0001
Vascular	2 (0.1%)	0	0	2 (0.2%)	0.11
Non-hemorrhagic stroke	9 (0.3%)	0	4 (0.4%)	5 (0.5%)	0.06
Other	115 (4.1%)	24(2.6%)	31(3.3%)	60 (6.5%)	<0.0001
Unknown	16 (0.6%)	1 (0.1%)	8 (0.9%)	7 (0.8%)	0.04
Any Cause	279 (10.3%)	32 (3.5%)	80 (8.0%)	167 (18.0%)	<0.0001

Patients were divided into tertiles of CFR corresponding to <1.5, 1.5–2.0 and >2.0. CFR = coronary flow reserve.

The annual rate of cardiac death increased with increasing myocardial scar and ischemia and decreasing CFR (Figure 2a,b). Importantly, lower CFR consistently identified higher risk patients at every level of myocardial scar/ischemia, including among those with visually normal PET scans. Likewise, in each category of LVEF, increasing tertiles of CFR were associated with a significant decrease in the risk of cardiac mortality (Figure 2c).

Figure 2. Unadjusted Cardiac Mortality.

Unadjusted annualized cardiac mortality by tertiles of CFR and categories of myocardial ischemia and scar (panel A); by categories of myocardial ischemia (panel B); and by tertiles of CFR and categories of left ventricular ejection fraction (panel C). The annual rate of cardiac death increased with increasing summed stress score, decreasing LVEF and CFR. Importantly, lower CFR consistently identified higher risk patients at every level of myocardial scar/ischemia and LVEF, including among those with visually normal PET scans and normal LV function.

Univariate Predictors of Mortality

Compared to the highest tertile of CFR (values >2), the lowest tertile (values <1.5) was associated with a 16-fold increased risk whereas the intermediate tertile was associated a 5.7-fold increased risk for cardiac death. Other significant predictors of increased risk included age, male gender, renal failure requiring dialysis and prior CAD (Figure 3). Chest pain as a reason for testing and obesity were unexpectedly associated with a decreased risk, possibly reflecting confounding due to more aggressive testing and treatment in these groups, although alternative mechanisms have been suggested for similar observations in other cohorts (25). In addition, dyspnea, a decrease in rest LVEF, the absence of stress-induced LVEF augmentation, as well as increasing burden of myocardial scar, ischemia or their combination on semi-quantitative visual analysis were all significantly associated with increased risk.

Figure 3. Univariate Predictors of Cardiac Death.

Univariate predictors of cardiac mortality are shown. Hazard ratios are presented for a one unit increase except for age (increase of 10 years), left ventricular ejection fraction (LVEF; decrease of 10%), extent of myocardial ischemia and scar combined and each separately (increase of 10%). CAD indicates patient reported coronary artery disease, known angiographic coronary stenosis, prior myocardial infarction or history of coronary revascularization. Hx = history of. ASA = aspirin. BMI = body mass index.

Multivariable Survival Analysis and Incremental Prognostic Value

A series of multivariable models were constructed to assess the incremental value of CFR after adjustment for critical covariates known to be associated with increased risk of cardiac mortality (Table 3). Addition of CFR was associated with a significant increase in global χ² and decrease in Akaike information criterion, indicating improved model fit, and a significant increase in the c-index from 0.82 to 0.84 (p=0.02). Compared to the highest tertile of CFR, the adjusted hazard ratio for cardiac death was 5.6 (95%CI 2.5–12.4, p<0.0001) for the lowest tertile and 3.4 (95%CI 1.5–7.7, p=0.003) for the intermediate tertile (Figure 4).

Table 3.

Multivariable Survival Analysis

	Model 1		Model 2		Model 3		Model 4		Model 5
	Fit Statistic	p-value	Fit Statistic	p-value	Fit Statistic	p-value	Fit Statistic	p-value	Fit Statistic	p-value
Global χ2*	154.74	ref	209.78	<.0001	227.06	<.0001	227.27	n/s	253.57	<.0001
AIC*	1919.08	ref	1866.03	<.0001	1850.76	<.0001	1852.54	n/s	1830.24	0
Calibration χ2	12.71	0.18	6.2	0.72	10.33	0.32	9.18	0.42	10.97	0.28
c-index*	0.78 (0.74–0.82)	ref	0.81 (0.77–0.85)	0.01	0.82 (0.78–0.86)	0.17	0.82 (0.78–0.86)	0.87	0.84 (0.8–0.87)	0.02
Covariate	HR	p-value	HR	p-value	HR	p-value	HR	p-value	HR	p-value
Age (y)	1.03 (1.01–1.05)	0.0002	1.03 (1.02–1.05)	<.0001	1.03 (1.02–1.05)	<.0001	1.03 (1.02–1.05)	<.0001	1.03 (1.01–1.05)	0.001
Male Gender	1.52 (1.05–2.21)	0.03	0.97 (0.66–1.43)	0.89	0.94 (0.64–1.39)	0.75	0.94 (0.64–1.38)	0.75	1.00 (0.68–1.48)	1.00
Hypertension	1.02 (0.62–1.68)	0.94	1.06 (0.64–1.74)	0.83	1.15 (0.70–1.89)	0.58	1.15 (0.70–1.88)	0.59	1.04 (0.63–1.71)	0.88
Dyslipidemia	0.88 (0.60–1.30)	0.53	0.90 (0.61–1.33)	0.59	0.80 (0.54–1.19)	0.28	0.80 (0.54–1.19)	0.28	0.84 (0.57–1.25)	0.40
Diabetes	1.73 (1.22–2.45)	0.002	1.51 (1.06–2.15)	0.02	1.53 (1.07–2.17)	0.02	1.52 (1.07–2.17)	0.02	1.39 (0.97–1.98)	0.07
Family Hx CAD	0.84 (0.56–1.25)	0.39	0.85 (0.57–1.26)	0.41	0.86 (0.58–1.29)	0.47	0.87 (0.58–1.30)	0.49	0.86 (0.58–1.29)	0.47
Tobacco Use	1.43 (0.85–2.41)	0.17	1.44 (0.86–2.43)	0.17	1.48 (0.88–2.49)	0.14	1.47 (0.88–2.48)	0.14	1.56 (0.92–2.64)	0.10
CAD	3.02 (1.95–4.67)	<.0001	2.26 (1.45–3.55)	0.0004	1.81 (1.13–2.88)	0.01	1.80 (1.13–2.87)	0.01	1.60 (1.01–2.55)	0.05
BMI (kg/m2)	0.94 (0.91–0.97)	0.0003	0.96 (0.94–1)	0.03	0.96 (0.93–0.99)	0.02	0.96 (0.93–0.99)	0.02	0.96 (0.93–0.99)	0.02
Chest Pain	0.55 (0.38–0.81)	0.003	0.7 (0.47–1.03)	0.07	0.71 (0.48–1.05)	0.08	0.71 (0.48–1.05)	0.08	0.79 (0.54–1.17)	0.24
Dyspnea Early	1.69 (1.2–2.37)	0.0025	1.37 (0.97–1.93)	0.07	1.46 (1.03–2.06)	0.03	1.45 (1.03–2.05)	0.03	1.45 (1.03–2.05)	0.03
Revascularization	0.97 (0.58–1.62)	0.91	0.88 (0.53–1.46)	0.62	0.62 (0.37–1.07)	0.08	0.62 (0.36–1.06)	0.08	0.57 (0.33–0.97)	0.04
Rest LVEF (%)			0.96 (0.95–0.97)	<.0001	0.97 (0.96–0.98)	<.0001	0.97 (0.95–0.98)	<.0001	0.97 (0.96–0.98)	<.0001
Ischemia+Scar (%)					1.03 (1.01–1.04)	<.0001	1.03 (1.01–1.04)	<.0001	1.02 (1.01–1.03)	0.0005
LVEF Reserve							0.92 (0.64–1.32)	0.64	1.03 (0.72–1.48)	0.86
CFR Lower Tertile									5.55 (2.49–12.4)	<.0001
CFR Middle Tertile									3.40 (1.50–7.72)	0.003

Summary of characteristics of five nested models.

^*P-values for fit statistics are for comparison of each model to the next simpler model (e.g. model 5 s model 4). C-indices and Hosmer-Lemeshow statistics are calculated for 1.5-year event data. Global χ² = likelihood ratio chi-squared statistic for the entire model. AIC = Akaike Information Criterion. Calibration χ² = Nam-D’Agostino goodness of fit statistic. Hx = history of. CAD = coronary artery disease. BMI = body mass index. LVEF = left ventricular ejection fraction. CFR = coronary flow reserve.

Figure 4. Cardiac Mortality.

Cumulative incidence of cardiac mortality for tertiles of coronary flow reserve presented in Kaplan-Meier format (panel A) and after adjustment (18) for age, sex, body mass index, hypertension, dyslipidemia, diabetes mellitus, family history of coronary artery disease (CAD), tobacco use, prior CAD, chest pain, dyspnea, early revascularization, rest left ventricular ejection fraction (LVEF), summed stress score and LVEF reserve (panel B) showing a significant association between CFR and cardiac mortality. HR = hazard ratio.

In contrast, replacement of CFR by peak stress myocardial blood flow in a model containing clinical covariates, rest LVEF, the combined extent of myocardial scar and ischemia, and stress-induced LVEF augmentation resulted in a more modest increase in the global χ² from 227.3 to 230.6 (p=0.07). The adjusted hazard ratio for cardiac death was 1.1 (95%CI 1.0–1.3, p=0.07).

Risk Reclassification

Addition of CFR estimates to the model resulted in the reclassification of 15%, 48%, and 12% of patients at low, intermediate, and high cardiac risk, respectively, based on the pre-CFR model (Table 3, Model 4) (Figure 5). The net reclassification improvement (NRI) was 0.098 (95%CI 0.025–0.180) across clinically meaningful risk categories of <1, 1–3, and ≥3% annual rate of cardiac death (Table 4). The continuous NRI was 43.5% (95%CI 23.9–62.2). The benefit of CFR on risk reclassification was greatest among patients with an intermediate pre-CFR risk, in whom addition of CFR to risk estimation downgraded risk in 32% (0% observed 3-year cardiac mortality) and upgraded it in 16% (13.6% observed 3-year cardiac mortality). Overall 34.9% of these intermediate pre-CFR risk patients were correctly reclassified. In these patients, the NRI across risk categories of <1%, 1–3 and ≥3% annual rate of cardiac death was 0.484 (95%CI 0.262–0.731). Among patients considered low and high risk without CFR data the NRI was 0.478 (95%CI −0.157–0.933) and 0.112 (95%CI 0.061–0.162), respectively.

Figure 5. Risk Reclassification.

Illustration of risk reclassification by addition of coronary flow reserve (CFR) to a model containing clinical risk factors, left ventricular ejection fraction (LVEF), LVEF reserve and combined extent of myocardial scar and ischemia. The upper horizontal bar graph represents the distribution of risk across categories of <1 (green), 1–3 (blue) and >3% (red) per year risk of cardiac death as estimated by a model containing clinical risk factors, rest LVEF, LVEF reserve and the combination of myocardial scar and ischemia (Model 4, Table 3). The pie graphs represent the proportions of patients in each pre-CFR category reassigned to each risk category after the addition of CFR to the risk model(Model 5, Table 3). The vertical bar charts at the bottom represent the annualized rates of cardiac mortality in each of the post-CFR risk categories.

Table 4.

Risk Reclassification

Model without CFR	Model with CFR
	<1% Annual Risk	1–3% Annual Risk	>3% Annual Risk	Total
Patients with Cardiac Death
<1% Annual Risk	2.2 (38.4)	3.6 (61.6)	0 (0)	5.8
1–3% Annual Risk	1 (3.4)	19.6 (63.9)	10 (32.8)	30.7
>3% Annual Risk	0 (0)	4.7 (4.4)	101 (95.6)	105.6
Total	3.3	27.8	27.8	142.1
Patients without Cardiac Death
<1% Annual Risk	924.8 (84.5)	169.4 (15.5)	0 (0)	1094.2
1–3% Annual Risk	303 (34.9)	425.4 (49)	139 (16)	867.3
>3% Annual Risk	22 (3.2)	84.3 (12.4)	573 (84.3)	679.4
Total	1249.7	679.2	712	2640.9

Reclassification table for censored data using method of Steyerberg and Pencina (37) from 1.5 year event data. Parentheses indicate percentages of each pre-test category reclassified to each post-stress category.

Subgroup Analysis & Effect Modification

The impact of CFR on the risk of cardiac death was consistent across 29 of 33 subgroups evaluated (Figure 6), suggesting the broad potential impact of this measure in clinical practice. Interaction terms between CFR and age, gender and diabetes were non-significant, underscoring the consistent effect of CFR on cardiac mortality.

Figure 6. Subgroup Analysis.

Analysis of the hazard ratio for cardiac death across patient subgroups between the upper and lower tertiles of coronary flow reserve (CFR). The graph demonstrates the consistent effect of impaired CFR on cardiac mortality. Hazard ratios were computed without adjustment for other covariates. The left vertical dashed line indicates unity. The right vertical dashed line corresponds to the point estimate for the hazard ratio in the entire population with the grey bar indicating the 95% confidence interval for this estimate. In three subgroups, each containing 20% or fewer of patients in the study (i.e. normotensive individuals, those without stress-induced left ventricular ejection fraction (LVEF) augmentation and those who underwent early revascularization), the Cox model did not generate meaningful estimates for the hazard ratios (indicated by ★). Hx = history of. CAD = coronary artery disease. LVEF = left ventricular ejection fraction.

Transient ischemic dilation (TID) has been associated with poor prognosis and severe disease on coronary angiography. Among 1883 patients (67.7%) for whom TID was available, incorporation of TID into a model containing clinical risk factors, LVEF, extent of ischemia and scar and stress-induced LVEF augmentation was not associated with an improved model fit, increase in c-index or positive NRI. However, addition of CFR to the comprehensive model containing TID continued to improve model fit (p<0.0001), the c-index (0.84 vs. 0.86, p=0.05) and was associated with favorable risk reclassification (NRI=0.068).

In order to assess the relative impact of epicardial stenosis and microvascular dysfunction on impaired CFR, we correlated CFR findings to angiographic findings in the 41 of 1444 patients (2.8%) with normal PET scans on semi-quantitative visual assessment and no history of coronary artery bypass graft surgery who underwent coronary angiography within 90 days of their PET scan and did not have intervening myocardial infarction or revascularization. Nineteen of the 41 patients (46%) had non-obstructive disease by coronary angiography (<70% stenosis). Likewise, 12 of 25 patients with abnormal CFR (low or intermediate CFR tertiles) had no obstructive epicardial stenosis by angiography.

All-Cause Mortality

Analyses were repeated using mortality from any cause as a secondary outcome and the results remained significant, although somewhat attenuated. After correction for a wide array of risk factors, CFR remained a significant predictor of mortality with the lowest tertile of CFR associated with a hazard ratio of 2.82 (1.87–4.26, p<0.0001) compared to the highest tertile. Similarly, the middle tertile of CFR carried a hazard ratio of 1.79 (1.17–2.72, p=0.007) compared to the highest tertile. The continuous NRI for all-cause mortality was 0.408 (95%CI 0.272–0.537). Using thresholds of 2 and 6% annually for mortality from any cause, the NRI was 0.058 (95%CI 0.006–0.111).

Discussion

These data from a large patient cohort demonstrate the incremental prognostic value of non-invasive measures of coronary flow reserve over the combination of comprehensive clinical assessment, LV systolic function and semi-quantitative measures of myocardial ischemia and scar for identification of patients at risk of cardiac mortality. Indeed, patients with the highest tertile of CFR, indicating preserved vasodilator function, had an extremely low rate of cardiac mortality (<0.5%/year). Conversely, intermediate and severely reduced CFR were associated with an adjusted HR for cardiac mortality of 3.4 and 5.6, respectively. The addition of CFR to clinical variables, rest LVEF, LVEF reserve and the extent of myocardial scar and ischemia resulted in the correct reclassification of approximately one third of all intermediate risk patients.

Prior Studies

Prior studies have evaluated the diagnostic value of CFR compared to coronary angiography and fractional flow reserve, especially with respect to detection of multi-vessel CAD (26,27). More recent studies have examined the prognostic value of non-invasive estimates of CFR in less than 700 patients with limited numbers of outcome events (28–31). Although these studies suggested an association between impaired CFR and adverse cardiovascular events (including hospitalization for cardiac causes and late revascularization), they were underpowered to demonstrate an association between impaired CFR and cardiac mortality. The demonstration of incremental prognostic value was also limited by incomplete adjustment for important markers of clinical risk (28,30) and measures of LV systolic function (29,30).

Our study included a large number of patients with known or suspected CAD encompassing a wide spectrum of clinical risk, and was sufficiently powered to adjust for a broad array of clinical and imaging risk markers, and demonstrate an association between decreased CFR and risk of cardiac mortality. More importantly, the inclusion of CFR in risk prediction models resulted in the correct reclassification of risk in a substantial proportion of patients, including 35% of intermediate risk patients (NRI=60.3%). Improved risk estimates obtained with CFR may impact patient management by triggering more targeted management of risk factors and/or referral for revascularization. These quantitative measures of CFR can be obtained at no additional cost, imaging time or radiation exposure.

Mechanistic Link Between Abnormal Coronary Circulatory Function and Cardiac Mortality

The exact mechanism relating non-invasive measures of coronary flow reserve to increased cardiac mortality cannot be determined from this study. However, several potential mechanisms could explain our findings. The observation that the reduction of CFR was primarily due to progressive reductions in peak stress myocardial blood flow suggests a primary abnormality in coronary vasodilator function.

As our study cohort included patients with both known and suspected CAD, it is likely that the observed abnormalities in coronary function integrate the fluid dynamic effects of epicardial stenosis with those of coronary risk factors and atherosclerosis on microcirculatory function. However, the fact that CFR improved risk stratification not only in the setting of demonstrable ischemia and/or scar as identified by conventional methods, but also among those patients with visually normal PET scans suggests that diffuse atherosclerosis and vascular remodeling, microvascular dysfunction, or both may also partially account for the excess cardiac mortality among those with impaired CFR. Indeed, the fact that approximately half the patients with visually normal PET scans who underwent coronary angiography had non-obstructive CAD, supports the notion that the observed excess coronary risk may be related in part to microvascular dysfunction. Such abnormalities have been described in early stages of atherogenesis in patients without overt CAD or angiographically normal coronary arteries (32), and have been linked to disease progression (10) and adverse cardiovascular events including sudden cardiac death, myocardial infarction, congestive heart failure, and coronary revascularization (9,11,33,34).

Limitations of our analyses merit consideration. The current study is a single-center, non-randomized, observational study and carries all of the inherent limitations of that study design. As such, it is likely that some amount of residual confounding remains, despite adjustment for a wide array of clinically relevant covariates. On the other hand, compared to data derived from patients selectively enrolled in a randomized trial, these data, with very limited exclusion criteria, may be more representative of patients seen in routine clinical practice.

Another side-effect of our broad inclusion criteria is that we may understate the value of parameters which have been previously shown to be markers of poor prognosis in more restricted subgroups (35). The breadth of the inclusion criteria of this study underscores the generalizability of the findings.

The demonstrated improvement in the c-index is only moderate in size, especially when compared to the more substantial impact on risk reclassification. However, this reflects the relative insensitivity of the c-index (36) rather than limited clinical value of CFR measures.

Peak stress blood flow on its own was a less powerful predictor of outcomes than CFR, possibly because CFR taken as the ratio of peak stress and rest blood flows may better isolate vasodilator capacity and reduce systematic errors in measurement.

Finally, we are not able to evaluate the downstream impact of CFR on patient management decisions as referring clinicians were not informed of CFR results in clinical reports. However, this reduces bias in estimates of CFR effect size introduced by subsequent treatment decisions based on this result.

Conclusion

Non-invasive assessment of coronary vasodilator function provides incremental risk stratification beyond routine measures of clinical risk, including estimates of LV systolic function and the extent and severity of myocardial ischemia and scar, and results in a meaningful incremental risk reclassification of patients with known or suspected CAD.

Qyantitative estimates of myocardial blood flow and coronary flow reserve (CFR) integrate the fluid dynamic effects of epicardial atherosclerosis and microvascular function. The present study demonstrates that CFR improves stratification of risk of cardiac and all-cause mortality beyond clinical risk factors, left ventricular ejection fraction (LVEF), extent of myocardial ischemia and scar, and stress-induced LVEF augmentation. Patients with CFR <1.5 were at 5.6-fold increased risk of cardiac mortality compared to those with CFR >2.0. Among patients whose clinical risk factors, LVEF and stress imaging findings placed them at intermediate risk of cardiac death (1–3% per year), 35% were reclassified as either high risk (>3% cardiac mortality per year) or low risk (<1% cardiac mortality per year). These findings demonstrate that incorporation of coronary vasodilator function assessment into stress testing by quantification of CFR improves risk stratification in patients with known or suspected coronary artery disease.