The current study included the patients with available ¹³N-ammonia PET within 3 months of invasive physiologic assessment.12) There was no intervention nor clinical event during the time between the PET scan and angiography/pressure measurement. Among a total of 144 patients, 15 patients with poor PET image quality, 2 patients without adequate coronary angiography data and 2 patients who underwent coronary bypass graft surgery were excluded. Patients with insufficient image quality for QFR analysis (calibration failure, ostial disease, insufficient projections, tortuous vessels, vessel overlap, or poor contrast filling) were also excluded and the specific reasons are presented in Supplementary Figure 1. Finally, 109 patients with 185 vessels were analyzed in this study. 3D QCA and analysis of QFR were retrospectively performed by an experienced analyst. The study protocol was approved by the ethics committee at Seoul National University Hospital (IRB No. H-1106-004-364) and was conducted in accordance with the Declaration of Helsinki.

All ¹³N-ammonia PET images were acquired during resting and hyperemic states by continuous intravenous infusion of adenosine (140 µg·kg⁻¹·min⁻¹) as previously described.13) A bolus of ¹³N-ammonium (370 MBq) was injected via a peripheral vein in both resting and hyperemic states, and list mode dynamic imaging was performed with a Siemens Biograph-40 PET/computed tomography scanner (Siemens Medical Solutions, Erlangen, Germany).14) A 2-compartment model was applied to quantify absolute myocardial blood flow (MBF, mL/min/g). The absolute MBF and physiological indices of a target segment were calculated from PET scan data.14) For image analysis and quantification of resting and stress absolute MBF in milliliters per minute per gram of tissue image acquisition, Carimas software (Turku PET Center, Turku, Finland) was used. Parametric stress MBF polar maps were used to delineate defect areas in target myocardial territories and to obtain MBF values of target segments. PET-derived coronary flow reserve (CFR) was calculated as the ratio of hyperemic MBF to resting MBF in target segments.15) Relative flow reserve (RFR) was calculated as the ratio of stress MBF in target myocardial segments to that of reference myocardial segments.16) To compare the diagnostic performance of pressure-derived physiologic indices, CFR <2.0 and RFR ≤0.80 were used as reference ischemic standards. Coronary flow capacity (CFC) was defined as flow capacity combining maximum absolute MBF and CFR,17) and ischemic cut-off values of 2.0 for CFR and 1.85 mL/min/g for stress MBF were used to define the presence of myocardial ischemia.18) Hyperemic stenosis resistance (HSR) was calculated as mean stenosis pressure gradient divided by hyperemic MBF.

3D QCA and analysis of QFR were performed by an experienced analyst (AngioPlus; Pulse Medical Imaging Technology, Shanghai Co., Ltd., Shanghai, China) from an independent core laboratory at Zhongshan Hospital, Fudan University in a blinded fashion as previously described.5) Briefly, end-diastolic frames of 2 optimal angiography projections, which were separated with angles of at least 25°, were selected and used for 3D model reconstruction. The 3D contour model of the segment of interest and its reference vessel were constructed in an automated manner and manual correction of contour was performed, if necessary. After acquisition of fixed QFR, estimated contrast coronary flow was calculated using thrombolysis in myocardial infarction (TIMI) frame-count adjustment, which indicated the frames where contrast entered and exited the segmented part of the vessel. With application of TIMI frame-count adjustment in the calculation method, the software automatically calculated the contrast QFR value. The ischemic cut-off value of contrast QFR ≤0.80 was used in the current study.

Coronary angiography was performed with standard techniques. Angiographic views were obtained after the administration of intracoronary nitrate (100 or 200 µg). All angiograms were analyzed at a core laboratory (Seoul National University Hospital) by investigators blinded to other data. Two-dimensional (2D) QCA was performed using a validated software (CAAS II; Pie Medical System, Mastricht, Netherlands). Minimal lumen diameter, reference vessel size, and lesion length were measured, and percent diameter stenosis (%DS) was calculated.

All coronary physiological measurements were obtained after diagnostic angiography as previously described. Briefly, a 5F to 7F guide catheter was used to engage the coronary artery (Supplementary Table 1). A pressure-temperature sensor guide wire (Abbott Vascular, Santa Clara, CA, USA) was zeroed and equalized to aortic pressure, and then the pressure sensor was positioned at the distal segment of a target vessel. Intracoronary nitrate (100 or 200 µg) was administered before each physiological measurement. Continuous infusion of adenosine (140 µg·kg⁻¹·min⁻¹) was used to induce hyperemia. FFR was calculated as the mean distal coronary artery pressure/aortic pressure during hyperemia. All pressure data were collected and validated by investigators blinded to other data. iFR was calculated as the mean pressure distal to the stenosis divided by the mean aortic pressure during the diastolic wave-free period. Baseline tracing data of >5 heartbeats before adenosine infusion was extracted and then anonymized and coded as an ASCII text file. Those data were sent to the iFR core laboratory (Imperial College, London, UK), where iFR was calculated with fully automated algorithms acting over the wave-free period of a minimum of 5 beats.19)

Categorical variables were presented as numbers and relative frequencies. Continuous variables were presented as mean±standard deviation or median with interquartile range according to their distributions, which were checked by the Kolmogorov-Smirnov test. The Kruskal-Wallis test was used for comparison of continuous variables among the groups. Spearman correlation coefficients were calculated to evaluate the correlations among pressure physiologic indices, and polynomial regression was used to visualize the relationships between them. The Fisher r- to z-transformation was used to compare the difference between correlation coefficients.

The area under curve (AUC) was calculated from receiver operating characteristic (ROC) curve analysis and compared using with the DeLong method.20) Diagnostic performance of pressure-derived physiological indices was presented with sensitivity, specificity, positive predictive value, negative predictive value, and diagnostic accuracy. They were compared using McNemar test (sensitivity, specificity, and accuracy) or weight generalized score statistic (positive predictive value and negative predictive value).

All probability values were 2-sided, and p value <0.05 was considered statistically significant. The statistical package R, version 3.4.3 (R Foundation for Statistical Computing, Vienna, Austria) was used for statistical analysis.

Baseline patient and lesion characteristics are presented in Table 1. The mean age was 64.3±9.0 years, and 95 patients (87.2%) were male. Mean % diameter stenosis from QCA, FFR, iFR, and QFR were 46.4±16.2%, 0.83±0.11, 0.92±0.12 and 0.84±0.13, respectively. The distributions of angiographic and physiologic indices are shown in Supplementary Figure 2. PET-derived RFR and CFR were 0.83±0.11 and 2.3±0.7, respectively.

Figure 1 shows the correlations between pressured-derived physiological indices and anatomical stenosis severity. All 3 indices showed significant negative correlations with 2D %DS from QCA and 3D %DS from QFR analysis (all p values <0.001), while higher values of correlations were observed with 3D %DS than with 2D %DS. QFR showed higher correlations with %DS than FFR or iFR (all p values for comparison<0.05). All pressure-derived physiologic indices decreased along with the increase of hemodynamic stenosis severity such as RFR, CFR, stress MBF, and HSR (Figure 2 and Supplementary Table 2). The degree of correlation was lower in QFR than that in FFR for RFR (0.655 vs. 0.786, p=0.008) and HSR (−0.797 vs. −0.950, p<0.001) (Supplementary Figure 3 and Supplementary Table 3).

Table 2 and Supplementary Figure 4 show the diagnostic performance of FFR, iFR and QFR against PET-derived RFR, CFR and CFC. With RFR as a reference standard, QFR showed lower sensitivity (58.5%) than FFR (70.8%, p value for comparison=0.011) and lower specificity (85.8%) than iFR (96.6%, p value for comparison=0.002), leading to the lower overall diagnostic accuracy of QFR (76.2%) than that of FFR (83.2%, p value for comparison=0.021) and iFR (84.3%, p value for comparison=0.031). With CFR as a reference, QFR showed lower specificity (80.7%) than iFR (89.6%, p valeu for comparison=0.013), but overall diagnostic accuracies of QFR, FFR and iFR were comparable. With CFC as a reference standard, QFR showed lower sensitivity (58.5%) than FFR (75.6%, p value for comparison=0.008) and lower specificity (78.5%) than iFR (87.1%, p value for comparison=0.008). The overall diagnostic accuracy of QFR (74.1%) was lower than that of iFR (82.0%, p value for comparison=0.031) against CFC.

The discriminant functions of QFR compared to FFR or iFR are shown in Figure 3. With RFR as a reference standard, the AUCs of FFR, iFR, and QFR were 0.899, 0.895 and 0.826, respectively. With CFC as a reference, the AUCs of FFR, iFR, and QFR were 0.788, 0.789 and 0.737, respectively. There was a significant difference only between FFR and QFR with both RFR and CFC as reference standards. With RFR as a reference, QFR showed a trend of lower discriminant function than iFR. There was no difference among AUCs of these 3 indices for CFR.

The diagnostic abilities of QFR in terms of diagnostic performance and discriminant function compared to FFR or iFR showed similar trends regardless of the target vessel location (Supplementary Figures 5 and 6 and Supplementary Tables 4 and 5). In addition, QFR showed high diagnostic performance to define myocardial ischemia in vessels which showed concordance results of pressure-derived and PET-derived indices (Supplementary Table 6).

QFR showed significant correlations with FFR (ρ=0.820, p<0.001) and iFR (ρ=0.742, p<0.001) and 14.6% showed discordant results between QFR and FFR or iFR (Supplementary Figure 7). The anatomic and physiologic parameters in each accordant and discordant group are summarized in Figure 4 and Tables 3 and 4. In both high FFR (A and B) and low FFR groups (C and D), there were differences in 2D and 3D %DS in high and low QFR groups (A vs. B, C vs. D). This finding was consistent with iFR and QFR comparisons. In contrast, PET-derived hemodynamic parameters showed different distribution patterns. For FFR-QFR comparison, there were no differences in RFR, CFR and hyperemic MBF between high and low QFR groups (A vs. B or C vs. D) in both high FFR and low FFR groups. For QFR and iFR comparison, the distribution of PET-derived hemodynamic parameters was similar except for stress MBF (Figure 4).