Introduction

18F-Flurpiridaz is a promising investigational radiotracer for myocardial perfusion PET with favorable properties for quantification of myocardial blood flow (MBF) and reserve (MFR).1,2,3 One phase 3 clinical trial has been completed (“301 trial”4) which established the diagnostic performance of flurpiridaz PET myocardial perfusion imaging,5 and a second phase 3 trial is currently underway (“303 trial”6). Several animal studies have demonstrated excellent tracer extraction and have developed kinetic models using microsphere-derived flow measurements as the gold standard.7,8,9 In addition, the feasibility of flow measurements in humans has been shown in a small, single-center study.10 Prior studies, primarily single-center, with 13N-ammonia,11,12,1315O-water,14,15 and 82Rb16,17 have shown that quantification of stress MBF and MFR improves diagnostic accuracy. In the present study, we sought to evaluate the incremental diagnostic accuracy of 18F-flurpiridaz flow measurements in the multicenter flurpiridaz 301 trial against quantitative coronary angiography.

Methods

Study Population

Patients were enrolled in the 301 trial at 81 sites world-wide. Inclusion criteria included men and women ≥ 18 years of age who underwent invasive coronary angiography (ICA) without intervention, clinically indicated myocardial perfusion SPECT, and were capable of undergoing exercise or pharmacologic stress. Major exclusion criteria were unstable cardiovascular status including myocardial infarction within 6 months of 18F-flurpiridaz PET, percutaneous coronary intervention (PCI) or other invasive coronary procedures within 6 months prior to 18F-flurpiridaz PET, history of coronary artery bypass graft surgery, or current non-ischemic cardiomyopathy. Pre-test probability of CAD was determined according to ACC/AHA guidelines for exercise testing.18 Primary study results and trial design have been previously published.5 The availability of adequate quality dynamic PET images at rest and pharmacologic stress was an additional inclusion criterion for this secondary analysis which was not prespecified (dynamic imaging was optional at the discretion of the recruiting sites). All patients provided written informed consent prior to undergoing study procedures.

Coronary Angiography

Invasive coronary angiography was performed in accordance with institutional practice at each investigational site, and images were evaluated by a single blinded reader at the angiographic core laboratory (Boston Clinical Research Institute, Boston, MA USA) using quantitative methods (QCA PlusPlus, Sanders Data Systems, Palo Alto, CA USA). Angiographic obstructive disease was defined as either ≥ 50% or ≥ 70% reduction in luminal diameter.

Image Acquisition Protocol

Centralized quality control assessment was performed for each PET scanner prior to use in the 301 trial. A total of nine PET/CT scanner models from three vendors were represented in the present substudy (Supplemental Figure 1). 18F-flurpiridaz PET exams were performed within a median of 2 ± 28 days of ICA. PET preceded ICA in 131 (57%) of patients in this study. All patients were instructed to fast for at least 3 h before the imaging exam. PET imaging was performed on qualified scanners in 2D or 3D mode with CT attenuation correction, except one exam which utilized a rotating transmission source for attenuation correction. Dynamic PET acquisition at rest was initiated with intravenous bolus injection of 18F-flurpiridaz (100 ± 7 MBq) immediately followed by a 5-10 mL saline flush. PET images were acquired for 15 minutes. After quality control for alignment of PET and attenuation images,19 PET images were reconstructed with standard PET corrections (attenuation, randoms, scatter, deadtime, decay) without post-smoothing filter, according to prescribed time frames (15 × 10 seconds, 5 × 30 seconds, 5 × 60 seconds, 1 × 300 seconds). Pharmacologic stress was performed with adenosine, dipyridamole, or regadenoson according to local practice and the respective package inserts. At peak stress, dynamic PET image acquisition and reconstruction was repeated with intravenous bolus injection of 18F-flurpiridaz (218 ± 11 MBq). Static images were created by summing the dynamic PET images after the initial 2 minutes. The time between rest and stress 18F-flurpiridaz administration was 52 ± 11 minutes and no shorter than 30 minutes.

Image Analysis

For this secondary analysis, images were processed using Corridor4DM software version 2017 (INVIA Medical Imaging Solutions, Ann Arbor, MI). Static PET images of the heart were transformed to short axis orientation and left ventricular (LV) contours representing the endo- and epicardial boundaries were defined semi-automatically20 and applied to each dynamic frame. Patient motion between dynamic frames was visually identified and manually corrected.21

Relative Tracer Retention

Polar maps of relative tracer retention representing relative perfusion were generated from static PET images and normalized to peak 18F-flurpiridaz uptake in the left ventricle. Thirty-two patients without history of previous PCI, MI, stroke, heart failure, or acute coronary syndrome, and without angiographic evidence of coronary artery disease were selected for inclusion in a database to represent normal myocardial 18F-flurpiridaz distribution. The normal database was then used to estimate stress total perfusion deficit (TPD)22 as a measure of perfusion defect extent and severity in all patients. Vascular territories with evidence of infarction on the rest PET scan (moderate to severe defect corresponding to severity > 5.5 SD using the normal databases) were excluded because MBF and MFR may not be reliably estimated in such regions (N = 37 territories in 27 patients).

Absolute Myocardial Blood Flow

PET time-activity curves (TACs) were automatically sampled from 460 regions in the LV myocardium. An image-derived (whole blood) arterial input function was automatically sampled from a 3D region of interest extending from the left atrium into the left ventricle and centered near the mitral valve plane in each dynamic frame.23,24 Due to the lack of arterial blood samples or validated correction models, no corrections of the input function for metabolites and blood-binding were performed. Residual activity on the stress scan from tracer injection at rest was estimated and subtracted from TACs as previously described7,8,10 after ensuring proper image registration of the initial dynamic frames before arrival of the tracer bolus to the heart. A 3 × 3 mean filter was applied to dynamic polar maps followed by kinetic modeling using a previously described 2-tissue-compartment (also known as 3-compartment) model that was validated in a porcine model against microsphere flow.7,8 The influx rate constant K1 was equated with flow since it is equal to flow times extraction fraction, and previous preclinical data demonstrated very high first-pass extraction fraction of 0.94.25 Polar maps were generated of absolute flow (MBF, in units mL/min/g) and flow reserve (MFR, stress flow divided by rest flow) which are reported as global LV average and per vascular territory.

Statistical Analysis

Continuous variables are reported as mean ± SD unless noted otherwise. Continuous and categorical variables were compared by t tests and χ2 tests, respectively. For global analysis, patients were stratified by number of vessels with obstructive disease, and for per-vessel analysis, vascular territories were stratified by increasing severity of stenosis with the lowest severity subgroups further divided based on the presence of any remote obstructive CAD. Unadjusted receiver operating characteristic (ROC) curves for detecting obstructive CAD were analyzed with stress MBF or MFR and compared using DeLong’s test.26 ROC thresholds were selected with Youden’s index,27 and differences in sensitivity and specificity, and positive and negative predictive values were tested with McNemar’s test28 and Leisenring’s test,29 respectively. The incremental diagnostic value of MBF and MFR was assessed with multivariable logistic modeling that included patient age, sex, BMI, and pre-test likelihood of CAD as covariates, as well as stress TPD. Per-vessel logistic mixed modeling with the same covariates was also performed to account for within-patient correlations between vascular territories. Log transformations toward normality were applied to BMI, TPD, stress MBF and MFR.30 Linearity of predictors and presence of interaction terms were checked by modeling with regression splines31 and by plotting predictor effects with partial residuals.32 The change in discrimination after adding stress MBF or MFR to models containing above covariates was assessed with continuous net reclassification improvement (NRI) and the change in concordance index (c-index).33 Multiple comparisons were performed with an adaptive two-step procedure controlling false discovery rate at a level of 0.05.34 Two-sided P values < 0.05 were considered significant. Statistical analysis was performed using R version 3.6.135 and python version 3.6.5.36

Results

Patient Population

A total of 755 patients were recruited for the flurpiridaz 301 trial, of which 559 underwent pharmacologic stress PET. Within this subset, dynamic 18F-flurpiridaz PET data, as provided by GE Healthcare, were available for 276 patients from 43 clinical imaging sites. After quality control for dynamic PET, 231 of these PET exams (84%) were suitable for evaluation in this post hoc flow study. Reasons for quality control failure included: missing or corrupted dynamic images (N = 8); missed LV first-pass during dynamic scan (N = 12); and scaling errors in the early dynamic frames (N = 25). Patient characteristics of the flow subgroup are shown in Table 1, and were not significantly different compared to the entire 301 trial population (Supplemental Table 1). Patients with CAD (≥ 50% stenosis) were older, predominantly men, with significantly higher incidence of hypercholesterolemia, diabetes, history of MI and previous PCI. Two case examples are shown in Figure 1.

Table 1 Patient characteristics
Full size table
Figure 1
figure 1

Case examples. A female patient, age 60, BMI 51 kg·m2, with chest pain, multiple risk factors (hypertension, hypercholesterolemia, diabetes, smoking), and history of PCI to the LAD. Quantitative coronary angiography showed a 92% lesion in the proximal LAD, 88% lesion in the proximal LCX, and a 40% lesion in the mid RCA. 18F-flurpiridaz PET showed a large reversible defect in the LAD territory with coronary steal apparent in absolute flow but not relative retention polar maps, as well as globally and regionally reduced stress flow and flow reserve. Summed static images (2-15 minutes) are shown in the upper left, and kinetic fits to global LV time-activity curves (yellow points and lines) are shown in the bottom right. B male patient, age 50, BMI 16 kg·m2, asymptomatic, multiple risk factors (hypertension, diabetes, smoking), and abnormal baseline EKG. Quantitative coronary angiography showed a 28% lesion in the proximal LAD, 100% lesion in the mid LCX, and an 82% lesion in the mid RCA. 18F-flurpiridaz PET showed a partially reversible defect in the distal LCX and RCA territories, as well as regionally reduced flow and flow reserve in the LCX and RCA territories

Full size image

Relationship Between Stenosis Severity, Flow, and Flow Reserve

Because reference standard flow values were not available in this retrospective study, we compared our flow estimates to values in the literature. Figure 2 compares 18F-flurpiridaz flow estimates with weighted averages of three cardiac flow tracers (13N-ammonia, 15O-water, and 82Rb) from an extensive literature review that included approximately 15,000 patients in 250 publications (see Table 2 in the online appendix to Ref. 37). For patients with risk factors only (Figure 2A) and patients with CAD (Figure 2B), the average agreement was very good in general, although the average 18F-flurpiridaz flow reserve for CAD patients tended to be somewhat higher than that of other tracers likely due to slightly lower average rest flow (Figure 2B). In Figure 3, the relationship of stress flow to stenosis severity is compared between 18F-flurpiridaz, 15O-water,3813N-ammonia,39 and 82Rb.40 Although the data scatter is large, the degree of stenosis at which the average stress flow was equal to a commonly used lower limit of normal stress flow41 was similar for all four tracers (70-80%).

Figure 2
figure 2

Comparison between absolute flow and flow reserve measured by 18F-flurpiridaz PET and three established PET flow tracers. A Patients with risk factors only (N = 131); B CAD patients (50% stenosis, N = 100). Other tracer data are from Table 2 of the online appendix to Ref. 37. The number of patients contributing to established tracer data points is shown below the graphs. Mean(est. tracers) indicates weighted mean ± pooled SD of established tracers

Full size image
Table 2 18F-flurpiridaz flow and flow reserve
Full size table
Figure 3
figure 3

Stress flow versus stenosis severity per vessel. A18F-flurpiridaz PET from the present study, B 15O-water PET from Ref. 38, C 13N-ammonia PET from Ref. 39, and D 82Rb PET from Ref. 40. Solid black curves are non-parametric locally weighted least squares fits representing average flow, and dashed red lines represent a commonly used lower limit of normal stress flow.41 The crossing point of this lower limit with the average stress flow was between 70 and 80% stenosis for these tracers

Full size image

On a per-patient basis, there was a significant, progressive decline in 18F-flurpiridaz stress flow with increasing extent of disease (Table 2, Figure 4A). Flow reserve exhibited a somewhat flatter response with less dependence on disease extent, although differences between subgroups were statistically significant (Figure 4B). In the regional analysis, there was also a progressive decline in stress flow with increasing %stenosis, and again the flow reserve was less dependent on %stenosis (Table 3, Figure 5). When stratified by patient gender, rest and stress flows were significantly higher in women than men (P < 0.05), and within-gender differences between the first two vascular subgroups (0-19% and 20-49%) were no longer significant (Supplemental Figure 2).

Figure 4
figure 4

18F-flurpiridaz flow and flow reserve. Per-patient distributions of global left ventricular rest and stress MBF (A) and MFR (B) (N= 204 patients; obstructive CAD was defined by a 50% stenosis threshold; 27 patients were excluded with evidence of infarction on the rest PET scan). *P < 0.05

Full size image
Table 3 18F-flurpiridaz flow and flow reserve
Full size table
Figure 5
figure 5

18F-flurpiridaz flow and flow reserve. Per-vessel distributions of regional rest and stress MBF (A) and MFR (B) stratified by stenosis severity (N = 656 territories; 37 vessels were excluded with evidence of infarction on the rest PET scan). *P < 0.05. †P < 0.05, all pairwise differences

Full size image

Diagnostic Performance of Flow and Flow Reserve

In per-vessel unadjusted ROC analysis using a 50% stenosis threshold, the c-index, sensitivity, and NPV were significantly higher for stress flow compared with flow reserve (0.79 vs 0.71, P < 0.001; 77 vs 62%, P < 0.001; 93 vs 89%, P < 0.001) (Figure 6A, Table 4). When a 70% stenosis threshold was used, c-index increased for both flow and MFR but the difference was no longer significant (0.83 vs 0.79, P = 0.0833) (Figure 6B). However, the specificity of flow reserve was slightly higher than that of stress flow without a significant difference in sensitivity (78 vs 74%, P = 0.016; 75 vs 80%, P = 0.248, respectively) (Table 4).

Figure 6
figure 6

Unadjusted ROC analysis of per-vessel CAD diagnosis with stress MBF or MFR. A 50% stenosis threshold and B 70% stenosis threshold

Full size image
Table 4 Unadjusted diagnostic performance of 18F-flurpiridaz stress MBF and MFR for detecting angiographic CAD (see Figure 6)
Full size table

Incremental Diagnostic Value of Flow and Flow Reserve

Four multivariable logistic models were tested for the detection of 70% stenosis. On a per-vessel basis, the incremental value of adding stress TPD, MBF, and MFR on accuracy is shown in Table 5. Both stress MBF and TPD (Table 5, Model 3; Figure 7A) and MFR and TPD (Table 5, Model 4; Figure 7B) were significant independent predictors of a 70% stenosis. Models which included all three variables or stress MBF and MFR without TPD (not shown) did not significantly improve model fit, or increase global χ2 or c-index. After adjustment for clinical covariates and TPD, the c-index of stress MBF and of MFR were not significantly different at 70% stenosis (Figure 7, 0.874 vs 0.869, P = 0.658), and marginally different at 50% stenosis (Supplemental Figure 3, 0.825 vs 0.799, P = 0.043). Similar features were observed for per-patient models of 70% stenosis (Table 6), and 50% stenosis (Supplemental Tables 2, 3, Supplemental Figs. 3-5). Significant interactions were present in all per-vessel models between stress MBF and TPD and between MFR and TPD, such that the importance of low stress MBF or MFR increased with larger and more severe defects (Figure 8, Supplemental Figure 6). There was no statistical evidence of non-linearity for either MBF or MFR. Overall NRI (Table 7) ranged from 0.219 to 0.559 and was significantly greater than zero (P < 0.05) in all but one model (MFR, per patient with 70% threshold, Supplemental Table 4). The improvement in c-index after adding stress MBF or MFR to per-vessel base models plus TPD ranged from 0.006 to 0.043, and was significant for stress MBF at 50% stenosis threshold (P < 0.05) (Table 7).

Table 5 Per-vessel multivariable logistic mixed models for detection of 70% stenosis demonstrating incremental diagnostic value of stress TPD, stress MBF, and MFR (N = 656)
Full size table
Figure 7
figure 7

Adjusted ROC analysis of per-vessel CAD diagnosis (70% stenosis threshold) with stress MBF (A) (Table 5) and MFR (B) (Table 6). Base models included patient age, sex, BMI, and pre-test likelihood of CAD. Figure legends indicate c-index values for each nested model

Full size image
Table 6 Per-patient multivariable logistic models for detection of 70% stenosis demonstrating incremental diagnostic value of stress TPD, stress MBF, and MFR (N = 204)
Full size table
Figure 8
figure 8

Per-vessel interactions for the logistic models in Table 5 (70% stenosis threshold). A Model 3, between stress MBF (left) and TPD (right); and B Model 4, between MFR (left) and TPD (right)

Full size image
Table 7 Continuous net reclassification improvement (NRI) of logistic regression models of CAD detection after including stress MBF or MFR
Full size table

Discussion

The results of this retrospective analysis of 276 patients from the multicenter flurpiridaz 301 trial demonstrate that global and regional 18F-flurpiridaz absolute flow and flow reserve may provide incremental diagnostic value for detecting obstructive CAD. Previous studies have shown similar results for absolute MBF and MFR derived from 13N-ammonia,11,12,1382Rb,16,17 and 15O-water 14,15 cardiac PET. All of these were single-center studies except that of Danad et al.15 which included three highly experienced academic PET centers. Importantly, the present multicenter study included PET data acquired at 43 clinical sites, including many smaller non-academic outpatient imaging centers, and used nine different PET/CT scanner models from three vendors (Supplemental Figure 1), supporting methodologic robustness in clinical practice.

Diagnostic Value of Stress MBF and MFR

In ROC analysis (Figure 6, Table 4), stress MBF provided more accurate detection of CAD than MFR, although this comparison was not adjusted for clinical covariates known to be associated with increased likelihood of CAD. In our study population, patients with obstructive CAD were more likely to be men and those greater than 60 y of age. In addition, both age and sex were more strongly correlated with stress MBF than MFR (multiple R2 0.272 vs 0.029, both P < 0.0001) such that women and younger patients had significantly higher stress MBF. Thus, without adjustment for age and sex, stress MBF would appear to have a diagnostic advantage over MFR based largely on patient demographics. After adjusting for these covariates in our logistic models, stress MBF and MFR performed similarly in terms of overall c-index (0.874 vs 0.869, P = 0.658) (Figure 7; see also Supplemental Figs. 3-5).

In contrast, MFR in the context of prognostic assessment has consistently been a stronger predictor of cardiovascular mortality than stress MBF when adjusted for age, sex, LV ejection fraction, and the extent and severity of ischemia and scar42,43 (see also Ref. 44 which did not meet these conditions). In this case, MFR can be reduced by additional factors not directly related to likelihood of obstructive epicardial CAD, such as resting hyperemia due to increased rate-pressure product, decreased myocardial efficiency, or autonomic dysfunction.45,46

Anatomical assessment by ICA correlates poorly with the hemodynamic functional significance of epicardial stenosis.41 Recent studies14,15 that have defined functional significance by invasive fractional flow reserve (FFR) have tended to report higher unadjusted diagnostic accuracy for stress MBF and MFR (e.g., c-index from 0.84 to 0.94) than the present study (c-index from 0.71 to 0.83, Figure 6). This may reflect the limitations of using solely an anatomical definition of obstructive CAD (as in the present study), as opposed to more contemporary functional definitions.37

Comparison with Established Flow Tracers

The comparison of 18F-flurpiridaz with other tracers illustrates two important points about flow variability and abnormal thresholds. First, for a given tracer, stress flow is widely scattered, particularly for non-obstructive stenoses < 50% (Figure 3). This variability is due to physiological differences in the integrated tissue response to hyperemia as measured by PET, which may be affected by patient age,47 sex,48,49 variations in pharmacologic vasodilation responsiveness,38 and varying degrees of subclinical atherosclerosis and microvascular dysfunction.37 Flow variability between tracers (Figure 2) arises from well-known physical differences in tracer characteristics. The first-pass extraction fraction of 15O-water, which is freely diffusible, is independent of flow, whereas that of retained tracers typically decreases non-linearly with increasing flow.50 Although established tracers have often been validated against microspheres or 15O-water flow in carefully controlled animal studies (see50 for references), most flow models of retained tracers require empirical extraction correction which may introduce characteristic flow differences when applied to clinical patient populations.51 Such absolute flow differences are most noticeable under hyperemic conditions and are progressively less important under resting or low flow conditions (Figure 2).

Second, the stress MBF threshold of 1.5 mL/min/g shown in the figures has been recommended as a lower limit of “definitely abnormal” flow, but is not intended to represent an optimal threshold of obstructive CAD common to all flow tracers.41 While this threshold indicates a similar average stenosis severity (70-80%), there is considerable scatter around this average behavior for all four tracers (Figure 3). Optimal thresholds of significant CAD will generally vary according to tracer characteristics, kinetic modeling methodology (e.g., Supplemental Figure 7), and the definition of obstructive CAD.41 For example, stress MBF thresholds of 1.86,52 2.30,15 and 2.5014,53 have recently been proposed for 15O-water, depending on the choice of stenosis threshold, methodologic variations, and whether the definition of obstructive CAD included functional assessment by FFR.

Although 18F-flurpiridaz stress flow in this study followed an inverse relationship with stenosis severity that was similar to other tracers,38,39,40 we also observed significantly higher rest flow in the lowest severity subgroup (0-19% stenosis) compared to the other subgroups (Figure 5A) which seems to conflict with previous findings.38,39,40 This may be a consequence of significantly higher rest flow in women compared to men in the present study (Supplemental Figure 2), and the fact that women contributed nearly two-thirds (100/159) of vessels in the lowest subgroup, but just 21% (102/497) in all other subgroups combined. Similarly, stress flow was significantly higher in women compared to men in the 3 least severe subgroups, although the corresponding flow reserves did not differ significantly (Supplemental Figure 2), in agreement with previous 15O-water,4913N-ammonia,48 and 82Rb45,46 PET studies.

18F-flurpiridaz Kinetic Model

In this study, we used a 2-tissue-compartment kinetic model previously validated for 18F-flurpiridaz against microsphere flow in pigs.7,8 A prior single-center study of 15 patients,10 which used a novel kinetic model without validation, reported flow estimates generally similar to ours (Table 8), although both studies lacked reference standard flow measurements.

Table 8 Comparison of 18F-flurpiridaz flow and flow reserve per vessel in patients
Full size table

The accuracy of 18F-flurpiridaz flow measurement can potentially be affected by several factors. Substantial residual activity was present during the stress scan which followed the rest scan by 52 ± 11 minutes (approximately 0.5 × 18F half-life). Similar 18F-flurpiridaz protocols have been reported in previous human and animal studies.7,8,10 Stress flow accuracy depends on correcting this residual activity, and these studies have used the simplest approach of estimating residual activity in the early dynamic frames before tracer arrival to the left ventricle and direct subtraction of this estimate from the time-activity curves. Although straightforward, this residual correction can be adversely affected by patient motion and poor count density in the early frames, and more sophisticated modeling approaches have been proposed.9

Another important factor is the use of an image-derived whole blood arterial input function which assumes rapid tracer equilibration between plasma and red blood cells with negligible blood-binding or tracer metabolism. However, in previous 18F-flurpiridaz animal studies, Guehl et al.9 described a blood-binding correction of the input function, while other authors have discussed the potential need for metabolite correction.7,10 Without correction, either of these processes could cause overestimation of the plasma input function, which in turn would result in flow underestimation. In fact, Nekolla et al.7 observed a modest underestimation of flow without metabolite correction, particularly in ischemic regions, while Guehl et al.9 did not observe any bias after blood-binding correction. Although these factors have yet to be fully characterized in humans, our results seem to indicate that their contribution to flow variations is relatively minor.

Study Limitations

This study had several limitations. First, only about half (276/559) of the pharmacologic stress PET data sets acquired in the 301 trial were available for analysis. This limited the number of available diseased vessels in women (Supplemental Figure 2), and likely limited the statistical power of ROC analysis and logistic regression modeling. Second, detailed anatomical information beyond stenosis severity from invasive coronary angiography was not available, precluding assessment of coronary dominance which could confound the evaluation of left circumflex and right coronary artery territories. Third, the acquisition protocol in the 301 trial was suboptimal for estimating residual activity in the stress images which, in many cases, consisted of a single 10-second frame. The residual correction could be improved by acquiring at least 30 seconds of list-mode data prior to initiating pharmacologic stress and tracer administration. Fourth, as mentioned previously, the study lacked reference standard flow measurements which will be necessary in the future to validate appropriate 18F-flurpiridaz kinetic models in humans. Finally, the 32 low-risk patients used to define normal databases of myocardial 18F-flurpiridaz retention were also included in the multivariable logistic regression models, which could potentially overstate the performance of TPD in these models compared to using an externally defined normal database. However, the magnitude of such an effect is small, and would not affect our conclusions regarding the performance MBF or MFR (see Supplemental Table 4).

Conclusion

We have characterized 18F-flurpiridaz flow and flow reserve in a large multicenter patient population with invasive coronary angiographic correlates. Both stress flow and flow reserve per vessel were inversely related to stenosis severity, generally agreed with average values from the extensive literature on other flow tracers, and added incremental diagnostic value beyond clinical characteristics and relative perfusion analysis. 18F-flurpiridaz PET shows promise for routine application of clinical flow quantification.

New Knowledge Gained

18F-flurpiridaz is an investigational radiotracer for PET myocardial perfusion. Using data from the international multicenter phase 3 flurpiridaz 301 trial, we report the first study demonstrating the added diagnostic value of absolute myocardial blood flow (MBF) and flow reserve (MFR) beyond traditional relative perfusion imaging. We show that both stress MBF and MFR meaningfully improve non-invasive diagnosis of obstructive coronary disease identified on invasive coronary angiography.