This study was a substudy of the prospective and multicentered CT-FFR CHINA trial which is registered at www.ClinicalTrials.gov (NCT03692936). Participants underwent cardiac CT angiography ≤7 days before scheduled, non-emergent, and clinically indicated ICA between November 2018 and March 2020. Ref. (46) has explained the inclusion and exclusion criteria of the trial in detail. Inclusion criteria included participants aged ≥18 years and one or more lesions with 30–90% diameter stenosis in a ≥2.0 mm diameter vessel according to cardiac CT angiography, scheduled for ICA based on clinical evaluation. Clinical exclusion criteria included previous percutaneous coronary intervention or bypass surgery, previous myocardial infarction, acute myocardial infarction, allergy to contrast agent, and contraindications to beta-blockers, nitroglycerin, or adenosine, serum creatinine >150 μmol/L, or a glomerular filtration rate <45 ml/kg/1.73 m², severe heart failure, and pregnant state. Cardiac CT angiography exclusion criteria were significant arrhythmia (atrial fibrillation) and poor image quality. Angiographic exclusion criteria were the failure of invasive FFR procedure and incomplete data for QFR computation. The flowchart related to this study is shown in Figure 1. With 28 patients in whom invasive FFR failed to be measured, 317 participants successfully underwent both ICA and FFR. In this study, half of the participants were scheduled for the QFR procedure to explore the value of angiographic lesion morphology for QFR. Therefore, these participants from five clinics were numbered in order, and participants with odd numbers were selected to calculate QFR and record lesion parameters. A total of 148 participants with 175 vessels were included for final analysis in this study. The diagnostic performances of the predictive models combining QFR with angiographic lesion parameters were assessed and compared with QFR. The study protocol was approved at the five centers by each of the local Institutional Review Boards, and participants provided written informed consent.

The ICA and FFR data were acquired and analyzed by a core laboratory in a blinded manner. QFR was computed offline by experienced technicians (L.X. and Z.S., both with more than 3 years of experience) from the cardiovascular intervention department who were blinded to FFR readings. The clinical data of the included participants were collected from electronic medical databases, including age, sex, body mass index, and high-risk factors for coronary artery disease.

Selective ICA was performed using standard catheterization according to the American College of Cardiology Guidelines for Coronary Angiography (47). At least two projection angles were acquired for the optimal view of the stenotic lesion. With clinical indication for FFR measurement, a pressure-monitoring guidewire (St. Jude Medical, Minneapolis, MN, United States) was advanced 1–2 cm distal to the lesion after administration of nitroglycerin. To achieve hyperemia, intravenous adenosine (160 μg/kg/min) was implemented. Pressure data were recorded for at least 3 s of stable value before adenosine administration and at least 10 s of stable value during hyperemia. FFR was defined as the ratio of mean distal coronary pressure to mean aortic pressure during maximum hyperemia. FFR ≤0.80 was considered hemodynamically significant (48).

Quantitative coronary angiography (QCA) analysis was performed based on angiograms using commercial software (QAngioXA 7.3; Medis Medical Imaging System, Leiden, Netherlands). Observers only received diagnostic angiographic runs and were blinded to any potential treatment, FFR results, and QFR results. Optimal projections were selected for the most stenotic lesion, and indices including minimal lumen diameter (MLD), reference lumen diameter (RLD), and lesion length (LL) were measured for analysis. In addition, percent diameter stenosis (%DS) and lesion length/minimal luminal diameter⁴ (LL/MLD⁴) were derived from the above-related indices. Notably, proximal and distal measurement points were approximately 10 mm away from the start and end of the lesion, respectively, if possible. LAD-involved lesions were recorded because the severity of non-LAD involved lesions tended to be overestimated by anatomical parameters of QCA lesions compared with LAD-involved lesions. Lesions with side branch (BL) and multiple lesions with %DS >30% (ML) in one interrogated vessel were recorded as well.

The QFR was calculated offline through a commercial software package (AngioPlus, Pulse Medical Imaging Technology, Shanghai Co., Ltd., Shanghai, China) by experienced technicians who were certified for the software operation and blinded to the FFR readings. Before QFR computation, technicians were informed about the location where the operators had measured the FFR to allow comparison of the QFR to the FFR at the same vessel site. The vessel was reconstructed from two diagnostic angiographic projections ≥25° apart (Figure 2) without a side branch. The lumen contour was automatically delineated, and a manual correction was allowed in cases of poor angiographic image quality or vessel overlap. The reconstructed vessel was automatically divided into several subsegments along the arterial centerline, and the estimated contrast flow velocity was derived via a frame count method (19). The estimated contrast flow velocity was automatically converted into a virtual hyperemic flow velocity using a quadratic function. The pressure drop for each subsegment was calculated using the stenosis geometry and virtual hyperemic flow velocity (19). The pressure drop at every position with respect to the most proximal position was calculated by integrating the pressure drop of all subsegments proximal to that interrogated location (19). Finally, the pressure drop along the segmented vessel enabled QFR reading along the vessel. QFR ≤0.80 was used as the diagnostic cutoff value.

SPSS 22.0 (IBM SPSS Statistics) and MedCalc 18.2 (MedCalc Software) software programs were applied for statistical analysis. All analyses were performed on a per-vessel level. The normality of quantitative data was assessed using the Kolmogorov–Smirnov test. As for quantitative variables with normal distribution, data were expressed as mean ± standard deviation and compared using t-tests. Otherwise, data were expressed as median with interquartile ranges and compared using the Kruskal–Wallis H test. In regard to categorical variables, the Chi-square test or Fisher exact test was used to compare rates as appropriate. The correlation between QFR and FFR was assessed with the Spearman correlation coefficient. A Bland–Altman plot and Wilcoxon signed-rank test were used to visualize and compare QFR and FFR.

All interrogated vessels were assigned to either the ischemia or non-ischemia group based on FFR. Two kinds of logistic regression models with two variables or three variables (one variable was QFR) were constructed to explore the incremental value of angiographic lesion morphology for QFR. Variables that were statistically significant in bivariate logistic regression analyses were included in trivariate logistic regression analyses. The diagnostic accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) of these models as well as QFR were calculated and analyzed. The AUC values for the combined models and QFR were compared with the methods of DeLong et al. (49), which were implemented through MedCalc 18.2 software. Statistical significance was assumed at P < 0.05.

A total of 11 participants with 11 vessels failed for QFR computation mainly due to a lack of proper position or calibrated data. Therefore, 148 participants with 175 vessels were included in this study. The detailed demographics of the 148 participants (mean age of 59.5 ± 9.7 years with 101 males) are presented in Table 1. The interrogated vessels included 119 LAD arteries, 29 left circumflex arteries, and 27 right coronary arteries. Lesions in side branches were classified into the corresponding main vessels.

The lesions had a mean FFR of 0.78 ± 0.13 and a median FFR of 0.81 (interquartile range: 0.70–0.88) (Figure 3A). A positive FFR (≤0.80) was identified in 46.29% (n = 81) of the participants. The comparisons of lesion morphology described by lesion location, lesion features, and QCA indices between ischemic and non-ischemic lesions are displayed in Table 2. There were significantly fewer ischemic lesions in left circumflex arteries and right coronary arteries than in LAD arteries. Moreover, 57.14% of vessels (n = 100) had side branches at the lesion site and 36% of vessels (n = 63) had multiple lesions with %DS >30%. BL and ML were more common in the ischemic group (both P < 0.0001). The mean MLD, RLD, %DS, and LL were 1.37 ± 0.46 mm, 2.66 ± 0.60 mm, 48.29 ± 13.2, and 10.52 ± 4.55 mm, respectively. Overall, lesions with a smaller MLD (1.11 vs. 1.61 mm, P < 0.0001), smaller RLD (2.50 vs. 2.79 mm, P = 0.002), and longer LL (11.64 vs. 9.46 mm, P < 0.0001) were more prone to causing myocardial ischemia. Correspondingly, the %DS (55.52 vs. 41.72, P < 0.0001) and the LL/MLD⁴ Poiseuille-based coronary angiographic index (7.67 vs. 1.47, P < 0.0001) were greater in the myocardial ischemia group compared with the non-ischemia group.

The correlation and agreement between QFR and FFR are shown in Figures 3B,C, respectively. A good correlation (Figure 3B) and agreement (Figure 3C) of QFR and FFR were observed with a correlation coefficient of 0.677 and a mean difference of −0.0015. The mean QFR and median QFR were 0.78 ± 0.15 and 0.82 (interquartile range: 0.70–0.91), respectively. The best cut-off value of QFR was 0.795. QFR correctly classified 82.29% of the vessels using FFR with a cutoff of 0.80 as the reference standard. The AUC of QFR was 0.86 (95% confidence interval, 0.80–0.91) with a sensitivity, specificity, PPV, and NPV of 80.25, 84.04, 81.25, and 83.16%, respectively.

Bivariate logistic regression was applied to construct predictive models combining a single lesion parameter with QFR for the prediction of myocardial ischemia (Table 3). Among the eight lesion parameters used in this study (i.e., LAD, BL, ML, MLD, RLD, %DS, LL, and LL/MLD⁴ with the former three being dichotomous variables), LAD (odds ratio = 0.233, P = 0.002), BL (odds ratio = 0.330, P = 0.006), MLD (odds ratio = 0.090, P < 0.0001), RLD (odds ratio = 0.407, P = 0.009), and %DS (odds ratio = 1.070, P = 0.001) were effective predictors for predicting myocardial ischemia when added to QFR. The accuracy, sensitivity, specificity, PPV, and NPV of the five models with effective predictors are presented in Table 4. Compared with QFR, the sensitivity of the combined models (QFR + BL and QFR + MLD) was significantly improved without compromised specificity and accuracy, as shown in Figures 4A–C. Furthermore, the AUC values and accuracy of the two combined models were higher than those of QFR (AUC: 0.88 vs. 0.86 and 0.90 vs. 0.86, respectively; accuracy: 82.86% vs. 82.29% and 85.72% vs. 82.29%, respectively), but they were not statistically significant (all P > 0.05) (Figure 4D).

Trivariate logistic regression was applied to construct predictive models combining two lesion parameters with QFR for the prediction of myocardial ischemia (Table 5). Five lesion parameters (i.e., LAD, BL, MLD, RLD, and %DS) were selected for trivariate logistic regression analyses because they were effective predictors in bivariate logistic regression. Considering the internal relation among MLD, RLD, and %DS, the three parameters were not combined in one model. Therefore, six combined predictive models were constructed with two lesion parameters as predictors added to QFR. The accuracy, sensitivity, specificity, PPV, and NPV of the six combined models are presented in Table 6. In contrast to QFR, the AUCs of the QFR + LAD + MLD, QFR + LAD + %DS, QFR + BL + MLD, and QFR + BL + %DS combined models were all significantly improved (0.91 vs. 0.86, P = 0.02; 0.91 vs. 0.86, P = 0.02; 0.90 vs. 0.86, P = 0.03, respectively) without compromising other performances (Figure 5D). Notably, the accuracy, sensitivity, specificity, PPV, and NPV of the QFR + BL + MLD combined model were higher than those of QFR (Table 6 and Figure 5), but they were not statistically significant (both P > 0.05).