The present analysis was conducted according to the recommendations of the Cochrane Collaboration6 and Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) statements.7 Studies were considered eligible if they provided comparison between clinical endpoints following FFR-guided or angiography-guided revascularization. Both RCTs and nonrandomized intervention studies (NRSI) were included; studies with either percutaneous coronary intervention (PCI) or coronary artery bypass grafting (CABG) as revascularization strategies were considered. If an NRSI provided both propensity score matched and unmatched cohorts, only matched cohorts were included. From inception through April 2023, PubMed/MEDLINE and Cochrane/CENTRAL databases were systematically searched for studies matching the inclusion criteria. A detailed search algorithm can be found in the supplementary data. A minority of articles was selected through citation searching and from authors’ contributions. Two authors (F. Mangiacapra and L. Paolucci) independently selected eligible works at the title/abstract level and disagreements were resolved by a senior author (E. Barbato). Eligible articles were evaluated at the full-text level. Risk of bias was estimated using the Cochrane RoB 2 tool for RCTs8 and the ROBINS-I tool for NRSI.9

Coprimary endpoints were all-cause death, myocardial infarction (MI) and major adverse cardiovascular events (MACE) at longest available follow-up. Among the 25 studies reporting MACE, 7 reported major adverse cerebral and cardiovascular events.10–16 To ensure that the analysis was sufficiently powered, we decided to merge these outcomes and refer to them simply as “MACE”. The various MACE definitions used are summarized in table 1 of the supplementary data. Secondary endpoints were any revascularization, a combination of target vessel failure/target lesion failure and the occurrence of periprocedural MI.

To compare the extent of revascularization deferral (ie, number of performed PCIs and treated lesions) between FFR- and angiography-guidance, an exploratory analysis was conducted using RCTs with only PCI as the revascularization strategy.

The number of events for each endpoint and the cohort size were extracted by careful evaluation of articles. If there were zero events in both groups for a certain outcome, a continuity factor of 1 was used. Heterogeneity was assessed using I2 statistics.17 Odds ratios (ORs) and 95% confidence intervals (95%CI) were calculated using a random-effects model according to the Der Simonian and Leird method.18 A fixed-effects model according to the Mantel-Haenszel method was used to assess OR heterogeneity (complete absence indicated by I2 statistic values near 0%).18 The absolute number of performed PCIs and treated lesions in the RCTs subgroup were compared using a Pearson chi-square test between the 2 groups. The rates of acute coronary syndromes (ACS) and revascularization deferral in the FFR arms of RCT and NRSI were compared in a similar fashion. Two separate analyses were conducted for the coprimary endpoints. In the primary analysis, both RCT and NRSI were included using a cross study design modelling within a Bayesian framework. With this method, studies with a suspected low level of evidence (ie, NRSI) are progressively undersized by inflating their mean effects variances by a wj factor (minimum relative weight: wj∼0; maximum relative weight: wj=1). No prior distributions of bias vectors were assumed.19 The results of the Bayesian adjustment are reported as logOR and 95%CI.

In the secondary analysis, only RCT and propensity score matching adjusted/prospective NRSI were included.20,21 Publication bias was estimated by Egger's test. Prespecified sensitivity analyses were performed by removing 1 study at a time across the groups and by using fixed-effect models for outcomes previously evaluated with random-effects models and showing low-to-moderate levels of heterogeneity. Subgroup analysis was performed between RCTs and non-RCT studies and formally evaluated with chi-square and I2 tests.22

Random-effects meta-regression analysis was performed to study the influence of revascularization rates and clinical presentation (acute vs chronic) on the coprimary endpoints. The percentage of patients undergoing revascularization and the percentage of patients with ACS enrolled in the FFR arm of each study were used as covariates. For the latter analysis, ACS was defined by the sum of patients with unstable angina (UA), non–ST-segment elevation MI, and ST-segment elevation MI.

Statistical analysis was performed with RevMan version 5.4.1 (The Cochrane Collaboration, The Nordic Cochrane Centre, Copenhagen, Denmark) and STATA 16 (Stata Corp, College Station, United States). The study is registered with PROSPERO (CRD42022344765).

The systematic search led to 4520 records considered for title/abstract level evaluation and 43 articles for full-text review, of which 21 were considered eligible for inclusion. Another 8 studies were added from other sources, eventually leading to the inclusion of 30 studies.5,10–16,23–44 The PRISMA diagram summarizing the research strategy is reported in figure 1 of the supplementary data. The final analysis included 11 RCTs,5,10,12,15,16,30–33,43,45 2 prospective nonrandomized studies,23,24 8 retrospective propensity score-adjusted studies,13,14,29,35–37,39,44 and 9 retrospective unadjusted studies.11,25–28,38,40–42 Risk of bias is shown in figures 2 and 3 of the supplementary data. The analysis involved 393 588 patients: 369 527 underwent angiographic guidance and 24 061 underwent FFR-guidance. The main characteristics of the 30 studies are shown in table 1.

The baseline characteristics of the populations enrolled are summarized in table 2 of the supplementary data. High clinical heterogeneity was observed among the studies, which enrolled patients in different clinical settings including ACS, chronic coronary syndromes (CCS), bypass graft lesions, aortic stenosis, and bifurcation lesions.

The extent of revascularization deferral was reported in 9 PCI RCTs5,15,16,30–33,43,45: 2447 of 2947 (83.0%) patients in the FFR-guided group and 2586 of 2929 (88.2%) patients in angiography-guided group underwent PCI (P< .001). Data on the number of treated lesions were available in 4 studies.5,31–33 Overall, 1485 of 2632 (56.4%) lesions in the FFR-guided group and 2206 of 2606 (86.7%) lesions in the angiography-guided group were treated with PCI (P< .001). Results are shown in table 2. Data on the extent of coronary revascularization in the PCI and CABG studies according to different variables are summarized in tables 3 and 4 of the supplementary data, while periprocedural MI event rates in each study are reported in table 5 of the supplementary data.

All-cause death was reported in 27 studies enrolling a total of 392 242 patients. An FFR-guided strategy was associated with a reduced risk of all-cause death (OR, 0.63; 95%CI, 0.53-0.73; I2=64%). A funnel plot is reported in figure 4 of the supplementary data. A sensitivity analysis performed by leaving out one study at a time showed no meaningful differences, and the fixed-effects model demonstrated comparable results (OR, 0.55; 95%CI, 0.51-0.59). In subgroup analysis, only NRSI FFR-guidance showed an association with reduced all-cause death (RCTs: OR, 0.99; 95%CI, 0.68-1.44, I2=31%; NRSI: OR, 0.56; 95%CI, 0.48-0.65; I2=64%), with a significant difference between the study design subgroups (P=.009). Mean effect variances adjustment of NRSI confirmed these findings up to a wj value of 0.2. The results of the primary analysis are shown in figure 1.

Figure 1.

Primary analysis for all-cause death. A: forest plot. B: mean effect variance adjustment of NRSI. 95%CI, 95% confidence intervals; FFR, fractional flow reserve; NRSI, nonrandomized intervention studies; RCT, randomized clinical trials. The bibliographic references mentioned in this figure correspond to: Puymirat et al.,5 Toth et al.,10 Lunardi et al.,11 Thuesen et al.,12 Di Gioia et al.,14 Rioufol et al.,15 Stables et al.,16 Wongpraparut et al.,23 Puymirat et al.,25 Di Serafino et al.,26 Li J et al.,27 Toth et al.,28 Fröhlich et al.,29 Layland et al.,30 Chen et al.,31 Van Nunen et al.,33 Di Gioia et al.,35 De Backer et al.,36 Fournier et al.,37 Parikh et al.,38 Völz et al.,39 Wong et al.,40 Adjedj et al.,41 Omran et al.,42 Lee et al.,43 Gerhardt et al.,44 Zhang et al.45.

(0.52MB).

The analysis was performed in 25 studies involving 62 212 patients. FFR-guidance was associated with a reduced risk of MI (OR, 0.70; 95%CI, 0.59-0.84; I2=43%). A funnel plot can be found in figure 5 of the supplementary data. Sensitivity analysis performed by removing 1 study at a time showed no significant differences and a fixed-effects model supported the results (OR, 0.70; 95%CI, 0.62-0.79). In subgroup analysis, only NRSI FFR-guidance showed an association with reduced MI (RCTs: OR, 0.84; 95%CI, 0.64-1.10; I2=20%; NRSI: OR, 0.63; 95%CI, 0.50-0.80; I2=49%), with no difference between study design subgroups (P=.120). These findings were consistent after adjustment of NRSI mean effect variances up to a wj value of 0.2. The results are shown in figure 2.

Figure 2.

Primary analysis for MI. A: forest plot. B: mean effect variance adjustment of NRSI. 95%CI, 95% confidence intervals; FFR; fractional flow reserve; MI, myocardial infarction; NRSI, nonrandomized intervention studies; RCT; randomized clinical trials. The bibliographic references mentioned in this figure correspond to: Puymirat et al.,5 Toth et al.,10 Lunardi et al.,11 Thuesen et al.,12 Di Gioia et al.,14 Rioufol et al.,15 Stables et al.,16 Wongpraparut et al.,23 Koo et al.,24 Puymirat et al.,25 Di Serafino et al.,26 Toth et al.,28 Layland et al.,30 Chen et al.,31 Van Nunen et al.,33 Di Gioia et al.,35 De Backer et al.,36 Fournier et al.,37 Parikh et al.,38 Wong et al.,40 Adjedj et al.,41 Omran et al.,42 Lee et al.,43 Gerhardt et al.44.

(0.54MB).

A total of 25 studies with 52 137 patients were included. FFR-guidance was associated with a reduced risk of MACE (OR, 0.77; 95%CI, 0.70-0.85; I2=41%). A funnel plot is shown in figure 6 of the supplementary data. A fixed-effects model supported this finding (OR, 0.76; 95%CI, 0.72-0.82) and sensitivity analysis did not impact the result. In subgroup analysis, only NRSI FFR-guidance showed an association with reduced MACE (RCTs: OR, 0.91; 95%CI, 0.77-1.08; I2=9%; NRSI: OR, 0.72; 95%CI, 0.64-0.81; I2=44%) and a significant difference between study design subgroups was evident (P=.030). These findings were confirmed even for high values of NRSI mean effect variances (wj=0.2). These results are reported in figure 3.

Figure 3.

Primary analysis for MACE. A: forest plot. B: mean effect variance adjustment of NRSI. 95%CI, 95% confidence intervals; FFR; fractional flow reserve; MACE, major adverse cardiovascular events; NRSI, nonrandomized intervention studies; RCT; randomized clinical trials. The bibliographic references mentioned in this figure correspond to: Puymirat et al.,5 Toth et al.,10 Lunardi et al.,11 Thuesen et al.,12 Di Gioia et al.,14 Rioufol et al.,15 Stables et al.,16 Wongpraparut et al.,23 Koo et al.,24 Puymirat et al.,25 Di Serafino et al.,26 Li J et al.,27 Toth et al.,28 Layland et al.,30 Chen et al.,31 Park et al.,32 Van Nunen et al.,33 Di Gioia et al.,35 De Backer et al.,36 Fournier et al.,37 Parikh et al.,38 Adjedj et al.,41 Lee et al.,43 Zhang et al.45.

(0.55MB).

The rates of PCI or CABG deferral in NRSI and RCTs were 63.0% and 12.9%, respectively (P <.001) (figure 4). The association between FFR-guided revascularization and reduced MACE was blunted with increasing numbers of patients treated with PCI or CABG. The MACE logOR increased by 0.032 for every 10% increase of the covariate (95%CI, 0.006-0.075; P=.057). A similar association was evident with all-cause death [logOR, 0.064; 95%CI, 0.027-0.102; P=.001], but not with MI (P=.167).

Figure 4.

Rates of coronary revascularization deferral and ACS in NRSI and RCTs. ACS, acute coronary syndromes; NRSI, nonrandomized intervention studies; RCT, randomized clinical trials.

(0.04MB).

The rate of ACS was lower in the NRSI compared with RCTs (respectively: 44.9% vs 55.7%; P <.001, figure 4). Meta-regression analysis suggested an inverse relationship between the percentage of patients with ACS and the association between FFR-guided revascularization and reduced MACE: logOR for MACE increased by 0.020 for every 10% increase of the covariate (95%CI, 0.001-0.039; P=.039). A tendency toward significance was found for the MI outcome (P=.050), while no association was found for all-cause death (P=.754).

Secondary analyses were performed including only RCTs and adjusted/prospective NRSI. The results of the primary analyses were confirmed for all coprimary endpoints (all-cause death: OR, 0.77; 95%CI, 0.63-0.93; I2=50%; MI: OR, 0.79; 95%CI, 0.67-0.94; I2=9%; MACE: OR, 0.86; 95%CI, 0.78-0.95; I2=2%). Notably, no difference was found between RCTs and adjusted/prospective NRSI for these outcomes. These results are reported in figure 5. Forest plot analyses are reported in figures 7, 8 and 9 of the supplementary data.

Figure 5.

Main results of the secondary analysis for the 3 coprimary outcomes. 95%CI, 95% confidence intervals; FFR, fractional flow reserve; MACE, major adverse cardiovascular events; MI, myocardial infarction; OR, odds ratio.

(0.06MB).

The risk of further revascularizations (OR, 0.93; 95%CI, 0.83-1.04; I2=29%) or target vessel failure/target lesion failure (OR, 0.75; 95%CI, 0.55-1.01; I2=29%) was not increased by FFR. An exploratory analysis using a fixed-effects model showed significant benefit in the FFR-guided group for this latter outcome (OR, 0.75; 95%CI, 0.57-1; I2=29%). The risk of periprocedural MI was reduced in the FFR-guided group (OR, 0.60; 95%CI, 0.39-0.95; I2=41%). Forest plot analyses are reported in figures 10, 11 and 12 of the supplementary data.

Compared with angiography, the adoption of physiology has the advantage of distinguishing epicardial coronary stenoses associated with myocardial ischemia,48 for which revascularization might be warranted, from those where revascularization could be deferred.49 The association between reduced FFR and adverse events has been widely described.50 Studies investigating the natural history of coronary lesions have shown that there is a strong relationship between lesion severity (evaluated with FFR) and rates of MACE and MI.51,52 These data are supported by evidence showing a high grade of inflammatory stress in FFR-positive lesions53 that could increase plaque instability. Consequently, the reduced rates of MI after revascularizing FFR-positive lesions could be associated with reduced mortality and MACE rates during follow-up. Compared with angiography, the addition of FFR maintains the benefits in terms of spontaneous MI reduction. At the same time, by deferring functionally insignificant lesions, FFR decreases the risks associated with procedural myocardial injury,54 which is related to impaired outcomes.55,56 Accordingly, our findings support that FFR-guided revascularization is consistently associated with a lower risk of MI and MACE. Moreover, when considering RCTs of patients treated with PCI, FFR adoption was associated with a lower number of interventions, confirming that deferring PCI in nonfunctionally significant lesions is safe.

Current guidelines recommend the implementation of coronary angiography with physiologic assessment to guide revascularization of intermediate stenoses in patients without prior evidence of myocardial ischemia.1 However, clinical adoption of FFR has been modest over the years.38,57 In addition to several possible logistic reasons limiting FFR penetration, there might also have been conflicting evidence from RCTs following the results of the initial trials.16,30–34 A critical appraisal of the literature suggests that major differences in study designs affecting the ratio of performed/deferred interventions should be thoroughly accounted for when estimating the effectiveness of FFR. In the FAME trial, intermediate lesions considered for invasive functional evaluation with FFR were at least 50% diameter stenosis by visual estimation.2 In contrast, a 30% diameter stenosis was adopted in other major RCTs.30 The inclusion of lesions with less than 30% diameter stenosis can be is expected to dilute the potential clinical benefit associated with functional assessment. In fact, in a hypothetical study randomizing patients to FFR or angiography, the inclusion of mild coronary lesions could be expected to lead to a higher rate of deferred interventions even in patients not treated with physiology guidance. If we assume that the superiority of FFR is related to the proportion of safely deferred nonindicated interventions compared with angiography, this could hinder the detection of a net clinical benefit. Our analysis supports the hypothesis that the benefit of FFR-guidance is more evident when the proportion of revascularization deferral is higher compared with pure angiography-guidance, possibly due to the lower risk of late and acute complications.42,58

We found an association between the rates of patients enrolled with ACS and the benefits of an FFR-guided revascularization strategy. The effectiveness of FFR in detecting significant lesions depends on the achievement of steady-state and maximal coronary hyperemia. In the setting of ACS (especially during ST-segment elevation MI), dynamic changes in the microvascular function of both the jeopardized and nonjeopardized myocardium can lead to increased resting and reduced hyperemic coronary flow. These modifications could hinder the effectiveness of FFR in detecting significant lesions, potentially leading to underestimation.59 Currently, conflicting data are available regarding the reliability of FFR evaluation in this setting and no definitive conclusions can be drawn.60–62 Additionally, further possible explanations could be proposed to justify our finding. In the acute setting, several conditions can hamper the effectiveness of FFR, including hemodynamic instability, challenges in identifying the real nonculprit lesions,30 the need for multiple interventions,5 and the risk of endangering nonculprit vessels during FFR guidewire manipulation.16 Further studies are needed to address this topic, focusing on the relationship between ACS subtypes and the length of follow-up.

One interesting point from our analysis is the striking difference in effect size between RCTs and NRSI. Recently published meta-analyses of RCTs have shown that FFR-guided revascularization was not superior to angiography in improving outcomes.46,47 These results highlight the inconsistency currently existing between data from RCT and NRSI.63 One of the main purposes of our study was to explore the nature of these conflicting data and to evaluate the relative contribution of observational studies in determining the benefits of FFR-guided revascularization. In our study, the influence of NRSI could be largely undersized without compromising the final evidence of a clinical benefit derived from the use of FFR. Similar findings have been confirmed including only adjusted NRSI. It is reasonable to assume that in NRSI, residual confounding from selection bias persists, especially in the absence of statistical adjustment. On the other hand, it could be argued that RCTs are highly selective, plagued by the so-called “entry bias”, and their generalizability is often questionable due to their tendency to exclude higher-risk patients.64 Among potential confounding factors, the appropriateness of physiological assessment based on clinical presentation and a differential risk profile of included patients may account for the observed divergence.63 Besides these considerations, other factors can influence the conflicting results of RCTs and NRSI, including the higher rate of patients deferred in NRSI and the limits of some RCTs in terms of sample sizes and the low number of observed events.63 Nevertheless, the interpretation of NRSI should be cautious and based on a thorough evaluation of inclusion/exclusion criteria and patient risk profiles.

This study has several limitations. First, the lack of patient-level data prevented us from performing prespecified adjustments. This should be particularly considered when evaluating our results on the relationship between clinical outcomes, ACS rates and deferral rates, due to the possibility of significant uncontrolled collinearity between these confounders. Moreover, although we performed several statistical computations to balance the evidence from different study types, we cannot assume that these results are comparable to those from only pooled RCTs in terms of quality. In this regard, propensity score matching is only one of the available methods used to adjust for confounders, and several strategies could have led to slightly different results. Again, when including NRSI in a meta-analytic framework, positive bias related to selective reporting can be difficult to address. Eventually, including a large number of significantly different studies has surely led to a residual unavoidable heterogeneity. Although we showed that there are significant differences between RCTs and NRSI in terms of inclusion criteria (eg, angiographic severity of the lesions), ACS rates, deferral rates and others, we cannot directly assume that our methodological efforts to manage these major sources of heterogeneity have been completely effective. Therefore, we believe that our results can only be considered as hypothesis-generating.