The present study was a patient-level pooled analysis of three prospective registries whose results have been previously published.4)5)7)8)9) The first study prospectively enrolled consecutive patients from five university hospitals in Korea (519 patients, 737 vessels), each undergoing clinically indicated invasive coronary angiography, and FFR, CFR, and IMR measurement for at least one coronary artery.5) The second study was an institutional registry of Tsuchiura Kyodo General Hospital, Ibaraki, Japan that included 643 patients (643 vessels) submitting to invasive angiography and physiologic assessment, including FFR, CFR, and IMR.7) The third study prospectively enrolled patients with FFR, CFR, and IMR data on at least one intermediate-grade stenosis from Hospital Clinico San Carlos, Madrid, Spain.8) In all such studies, patients with hemodynamic instability, left ventricular dysfunction, or a culprit vessel of acute coronary syndrome were excluded. Individual patient data for pooled analysis were collected using standardized spreadsheets. For all variables included, standardized definitions were used. Invasive physiologic indices were also cross-checked and confirmed by each study's principal investigators.

Among the 1,397 patients (1,694 vessels) enrolled overall, those undergoing percutaneous coronary intervention (PCI) were excluded, leaving 867 patients (1,152 vessels) who did not undergo PCI for the current analysis. Only one of the deferred patients was lost to follow-up. Study protocols were designed in accord with the Declaration of Helsinki and were authorized by Institutional Review Boards or Ethics Committees at corresponding centers. All patient participants granted written informed consent. The study protocols were in accordance with the Declaration of Helsinki. The study protocol of International Collaboration of Comprehensive Physiologic Assessment was registered at ClinicalTrials.gov (NCT03690713).

Coronary angiography was performed by standard techniques. All angiograms were analyzed at local core laboratories in blinded fashion. Percent diameter stenosis, minimum luminal diameter, reference-vessel size, and lesion length were measured. All coronary physiologic parameters were measured following diagnostic angiography. A guide catheter (5-7 Fr) without side holes was used to engage coronary arteries, and a pressure/temperature-sensor guide wire (St. Jude Medical [Abbott Vascular], St. Paul, MN, USA) was used to measure FFR and CFR.

To derive resting mean transit time (Tmn), a thermodilution curve was obtained by delivering three injections (3–4 mL each) of room-temperature saline. Hyperemia was induced by intravenous infusion of adenosine (140 µg/kg/min) via peripheral or central vein. Proximal aortic pressure (Pa), distal arterial pressure (Pd), and hyperemic Tmn were then measured during sustained hyperemia. Once measurements were complete, the guidewire was pulled back to the guide catheter, and potential pressure drift was checked. CFR was calculated as resting Tmn/hyperemic Tmn. For lesions with low FFR (≤0.80), PCI was recommended, as stipulated by current guidelines. However, decisions regarding PCI were at the discretion of operators. Patients who underwent PCI were excluded from the final analysis. The uncorrected IMR was calculated as Pd × Tmn during hyperemia. All IMR values were corrected by Yong's formula (Pa × Tmn × ([1.35 × Pd/Pa] − 0.32).

Cutpoints were set at FFR ≤0.80 (low FFR), CFR ≤2 (low CFR), and IMR ≥23 U (high IMR), as previously described.4)5)10)11)12) Patients with high FFR (>0.80) were grouped by CFR and IMR values as follows: 1) Group A (high CFR and low IMR); 2) Group B (high CFR and high IMR); 3) Group C (low CFR and low IMR); and 4) Group D (low CFR and high IMR).

Clinical data were obtained at outpatient clinic visits or by telephone contact if needed. The primary study endpoint was patient-oriented composite outcome (POCO), including all-cause mortality, any MI, and any revascularization. Secondary endpoints were individual components of POCO. As a sensitivity analysis, vessel-oriented composite outcome (VOCO), which included vessel-related cardiac death, vessel-specific myocardial infarction (MI), and vessel-specific revascularization was assessed. All clinical outcomes were defined as stipulated by the Academic Research Consortium, including the addendum to the definition of MI.13) In the absence of a clear non-cardiac cause, all deaths were considered cardiac related.

Categorical variables were expressed as numbers and relative frequencies (percentages), and continuous variables as means and standard deviations or medians with interquartile ranges (Q1–Q3) according to related distributions, which were checked by the Kolmogorov-Smirnov test. Linear regression analysis was used to estimate correlation coefficients of quantitative variables. Kappa values were calculated to gauge classification agreement, using established cutpoints of each invasive physiologic index. Event rates were calculated from Kaplan-Meier censoring estimates and presented with cumulative incidences, using log-rank or Breslow test to compare survival curves between groups. Cox proportional hazard regression was used to calculate hazard ratios (HRs) and 95% confidence intervals (CIs) for between-group comparisons. The assumption of proportionality was assessed graphically by log-minus-log plot, and Cox proportional hazard models for all clinical outcomes satisfied the proportional hazards assumption.

Data were analyzed on a per-patient basis for clinical characteristics and outcomes, and on a per-vessel basis for comparison of lesion characteristics and physiologic indices. The vessel of lowest FFR served as representative for per-patient analysis in instances of multivessel measurements. In per-vessel analyses, a generalized estimating equation (GEE) was applied to adjust for intrasubject variability among vessels of the same patient. Estimated means and 95% CIs were presented as summary statistics. A GEE procedure with pairwise comparison was invoked to address per-vessel variables in the four-group classification. No post-hoc adjustment was done. For the per-vessel comparison of the risk of VOCO, a marginal Cox proportional hazard regression model was used.

To identify independent predictors of POCO, a multivariable Cox model was used. Covariates of clinical relevance or showing univariate relations with outcomes (p<0.1) were entered into marginal multivariable Cox models. Candidate variables were carefully chosen, given the number of events available, to ensure parsimony of final models. We calculated Harrell's c-statistics (with 95% CI) to validate each model's discriminant function, assessing gains (with or without integrated physiologic indices) as relative integrated discrimination improvement (IDI). Each analysis incorporated a participating center as a random effect. All probability values were 2-sided, with p values <0.05 considered statistically significant. Standard software applications (SPSS v18.0 [SPSS Inc, Chicago, IL, USA] and SAS v9.3 [SAS Institute, Inc., Cary, NC, USA]) were used for statistical analyses.

Table 1 shows clinical, angiographic, and physiologic characteristics of the 867 patients (1,152 vessels) selected for study. Most presented with stable ischemic heart disease (IHD), and in those with acute coronary syndrome, non-culprit vessels were physiologically interrogated. Anatomic severity of epicardial coronary stenosis was largely of intermediate grade (mean diameter stenosis, 40.9±15.6%; mean FFR, 0.88±0.07), and FFR values were ≤0.80 in 164 vessels (14.2%). Mean CFR was 3.03±1.28, with CFR ≤2.0 in 300 vessels (26.0%), and median IMR was 17.4 U (Q1–Q3, 12.8–24.3 U), 324 vessels (28.1%) showing high IMR values. Supplementary Figure 1 depicts the distributions of FFR, CFR, and IMR.

FFR and CFR showed modest correlation (r=0.113, p<0.001) and low categorical agreement (κ=0.095, p<0.001), with discordance in 268 patients (30.9%) (Figure 1). IMR was highest in patients of high FFR and low CFR (overall comparison p<0.001) (Table 1).

Figure 2 presents the 5-year risk of POCO among deferred patients with high or low FFR according to the CFR level. In patients with high FFR, those with low CFR were at significantly greater risk in terms of POCO (14.9% vs. 8.6%, HR, 1.846; 95% CI, 1.083–3.146; p=0.024) and cardiac death or MI (HR, 5.662; 95% CI, 1.984–16.154; p=0.001) than those with high CFR. Similarly, those with low CFR showed significantly greater risk of POCO than those with high CFR in patients with low FFR (34.8% vs. 14.0%, HR, 2.383; 95% CI, 1.066–5.329; p=0.034) (Figure 2 and Supplementary Table 1).

When evaluating distributions of CFR and IMR in patients with high FFR, modest correlation (r=0.230, p<0.001) and low categorical agreement (κ=0.159, p<0.001) were evident, but discordance of CFR and IMR was demonstrable in 33.8% (242/715) of patients with high FFR (Figure 3 and Table 2). On the other hand, severity of angiographic stenosis did not differ in the 4-group classifications (by CFR and IMR), and although FFR values differed in the four groups, the absolute difference was within 0.02. Distributions of clinical risk factors were mostly similar, except for age and proportion of multivessel disease. Group D showed the lowest CFR and highest IMR values (both p<0.001) overall (Table 2).

Among these patients, the total cumulative incidence of POCO and VOCO were 11.9% (83 events) and 7.3% (70 events) at 5 years. Risks of POCO were comparable in groups A, B, and C, as were individual clinical outcomes. Group D (low CFR and high IMR, indicating overt microvascular disease), showed significantly higher risk of POCO than the other groups (overall comparison p=0.013), surpassing that of group A by nearly 3-fold (21.2% vs. 8.0%, HR, 2.873; 95% CI, 1.476–5.594; p=0.002). The risk of cardiac death or MI in Group D was also higher than that of Group A (10.4% vs. 2.3%, HR, 5.662; 95% CI, 1.984–16.154; p=0.001) (Figure 4 and Table 3). Per-vessel analysis also showed significantly greater risk of VOCO only in group D, compared with other groups (overall comparison p=0.018), again surpassing that of group A by nearly 3-fold (HR, 3.029; 95% CI, 1.468–6.248; p=0.003) (Supplementary Figure 2).

In patients with high FFR, the presence of diabetes mellitus or multivessel disease showed an independent association with occurrence of POCO. When incorporating CFR and IMR into the multivariable model, the presence of overt microvascular disease also proved independently predictive of POCO; and this model with physiologic classification showed significantly more improvement in relative IDI (0.248, p=0.046) (Table 4).