Prognostic value of CT-derived fractional flow reserve for major adverse cardiovascular events in patients with suspected or known coronary artery disease: a systematic review and meta-analysis

Abstract

Background

Fractional flow reserve derived from CT (FFR-CT) enables non-invasive functional assessment of coronary stenoses in patients with suspected or known coronary artery disease, but evidence regarding its prognostic value remains fragmented. We conducted a systematic review and meta-analysis to quantify the association between abnormal FFR-CT and major adverse cardiovascular events (MACE), updating the 2022 meta-analysis by Nørgaard et al

Methods

We searched PubMed, Embase and Scopus through December 2025 for studies comparing outcomes in patients with suspected or known coronary artery disease with FFR-CT ≤0.80 versus >0.80 (PROSPERO: CRD420261276897). Random-effects meta-analysis with Hartung-Knapp-Sidik-Jonkman adjustment was performed. Subgroup analyses examined FFR-CT technology, geography and follow-up duration. Risk of bias was assessed using the Quality in Prognosis Studies (QUIPS) tool and certainty of evidence using Grading of Recommendations, Assessment, Development and Evaluations (GRADE) method.

Results

Twenty-two studies with 20 067 patients were included, representing a fourfold increase from the prior meta-analysis. Follow-up ranged from 12 to 120 months. Patients with FFR-CT ≤0.80 had significantly higher MACE risk (HR 3.94; 95% CI 2.92 to 5.31; p<0.0001) with substantial heterogeneity (I²=79.9%). However, restricting to hard endpoints (death/myocardial infarction; k=6) yielded HR 3.28 (95% CI 2.25 to 4.79) with zero heterogeneity (I²=0%). No significant differences emerged across FFR-CT technologies (p=0.812), including HeartFlow, machine learning and deep learning algorithms. Trim-and-fill analysis suggested possible publication bias, with adjusted HR 2.35. GRADE certainty was moderate.

Conclusions

Abnormal FFR-CT is associated with nearly fourfold increased risk of MACE in patients with suspected or known coronary artery disease, with consistent prognostic value across technologies. The absence of heterogeneity for hard endpoints supports FFR-CT as a reliable prognostic marker. These findings address the evidence gap identified by current guidelines regarding FFR-CT prognostic utility.

PROSPERO registration number

CRD420261276897.

Clinical trial number

Not applicable (as this is a Systematic Review and Meta-Analysis and not a Clinical Trial).

WHAT IS ALREADY KNOWN ON THIS TOPIC.

WHAT THIS STUDY ADDS

• This updated meta-analysis of 22 studies with 20 067 patients demonstrates that abnormal FFR-CT (≤0.80) is associated with nearly fourfold increased major adverse cardiovascular events risk across multiple technologies, geographic regions and follow-up extending to 10 years, with zero heterogeneity for hard endpoints of death and myocardial infarction.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• These findings address the evidence gap identified by current European Society of Cardiology guidelines regarding FFR-CT prognostic utility, supporting its integration into clinical pathways as an independent prognostic marker for cardiovascular risk stratification beyond its established role as a diagnostic tool.

Background

Coronary artery disease remains the leading cause of death worldwide. According to the Global Burden of Disease Study, an estimated 239 million people were living with ischaemic heart disease in 2021, contributing to nearly 20 million cardiovascular deaths annually, roughly one-third of all global mortality.¹ These numbers continue to rise, with projections suggesting a 90% increase in cardiovascular disease prevalence by 2050.² This staggering burden underscores the need for accurate, non-invasive diagnostic and prognostic tools that can guide clinical decision-making without exposing patients to unnecessary invasive procedures.

Over the past decade, coronary CT angiography (CTA) has emerged as the preferred first-line imaging modality for patients with suspected coronary artery disease. Current European Society of Cardiology (ESC) guidelines now assign CTA a class I, level A recommendation for patients with low-to-moderate pretest probability of obstructive disease,³ while American Heart Association guidelines similarly endorse CTA as a first-line test for both stable and acute chest pain evaluation.⁴ The SCOT-HEART trial provided compelling evidence for this approach, demonstrating that CTA-guided management reduced fatal and non-fatal myocardial infarction by 41% at 5 years compared with standard care.⁵

However, anatomic assessment alone has limitations. Coronary stenosis severity on CTA often correlates poorly with haemodynamic significance, and many patients with moderate lesions proceed to invasive angiography only to find non-obstructive disease. Fractional flow reserve derived from CT (FFR-CT) addresses this gap by providing functional assessment from the same anatomic dataset. The NXT trial established that FFR-CT achieves an area under the curve of 0.90 for detecting haemodynamically significant stenosis, with 93% negative predictive value.⁶ The PLATFORM study subsequently demonstrated that FFR-CT guidance could safely cancel 60% of planned invasive angiograms while reducing cases of non-obstructive disease at catheterisation by 83%.⁷ Real-world data from the ADVANCE registry, encompassing over 5000 patients across 38 international sites, confirmed that FFR-CT modifies treatment plans in two-thirds of cases.⁸

While the diagnostic accuracy and clinical utility of FFR-CT are now well established, evidence regarding its prognostic value has been slower to accumulate. In 2022, Nørgaard and colleagues published the first systematic review and meta-analysis specifically addressing FFR-CT prognosis.⁹ That analysis included five studies with 5460 patients and found that abnormal FFR-CT (≤0.80) was associated with a 2.3-fold increased risk of death or myocardial infarction at 12 months compared with normal FFR-CT (RR 2.31; 95% CI 1.29 to 4.13). However, the analysis was limited to HeartFlow technology, short-term follow-up and predominantly Western populations.

Since then, the evidence base has expanded considerably. Large-scale registries such as FISH&CHIPS, the largest FFR-CT outcome study to date with over 90 000 patients from the UK National Health Service, have demonstrated population-level mortality benefits.¹⁰ Long-term follow-up data have emerged, including 10-year outcomes from DISCOVER-FLOW showing durable prognostic value¹¹ and 3-year results from ADVANCE-DK confirming risk stratification across coronary calcium strata.¹² Randomised trials using novel machine learning and deep learning algorithms, such as TARGET¹³ and FORECAST,¹⁴ have added prospective outcome data with non-HeartFlow technologies. Meanwhile, Asian populations, previously under-represented, now constitute the majority of published prognostic cohorts.

Despite this wealth of new evidence, current guidelines still list FFR-CT prognostic utility as an evidence gap requiring further research.³ We therefore conducted this systematic review and meta-analysis as a comprehensive update to Nørgaard et al, incorporating all available prognostic studies through December 2025. Our objectives were to quantify the association between abnormal FFR-CT and major adverse cardiovascular events (MACE), explore sources of heterogeneity across technologies and populations and assess the certainty of evidence supporting FFR-CT for long-term risk stratification.

Methods

study design

This systematic review and meta-analysis followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses V.2020 guidelines¹⁵ and was registered with PROSPERO (CRD420261276897). We designed this study as an update to the meta-analysis published by Nørgaard and colleagues in Heart in 2022,⁹ which included five studies with approximately 5460 patients and follow-up limited to 1 year.

Data sources and search strategy

We searched PubMed, Embase (Ovid) and Scopus from database inception through December 2025. Our search combined terms related to CT-derived fractional flow reserve (FFR-CT, CT-FFR, FFRCT, virtual FFR, non-invasive FFR) with coronary CTA and prognostic outcomes. We placed no language restrictions and manually screened reference lists of included studies for additional relevant publications. Complete search strategy is found in online supplemental table S1.

Study selection

We included studies that: (1) enrolled adults with suspected or known coronary artery disease who underwent FFR-CT analysis; (2) compared clinical outcomes between patients with abnormal FFR-CT (≤0.80) and normal FFR-CT (>0.80) the threshold validated by the FAME trials as identifying ischaemia-causing stenoses with greater than 90% accuracy ^{16 17} ; (3) reported MACE, death, myocardial infarction or unplanned revascularisation; (4) had at least 6 months of follow-up and (5) provided data allowing extraction or calculation of HRs with CIs.

We excluded studies that exclusively enrolled acute coronary syndrome patients; evaluated only bypass grafts, stents or transplant vasculopathy; reported only diagnostic accuracy; used non-standard FFR-CT thresholds or continuous variables; represented overlapping cohorts with included studies (keeping the larger or longer study); or were conference abstracts superseded by full publications.

Two reviewers independently screened titles, abstracts and full texts. Disagreements were resolved through discussion.

Data extraction

We extracted study characteristics (author, year, country, design, sample size, follow-up), patient demographics (age, sex, diabetes, hypertension), FFR-CT technology (HeartFlow computational fluid dynamics, machine learning, deep learning or on-site methods), outcome definitions, and effect estimates.

For studies presenting Kaplan-Meier curves without explicit HRs, we extracted survival probabilities and calculated HRs using established methods.^{18 19} We validated this approach against studies reporting both Kaplan-Meier data and Cox regression HRs.

Risk of bias assessment

We used the Quality in Prognosis Studies (QUIPS) tool to assess risk of bias across six domains: study participation, attrition, prognostic factor measurement, outcome measurement, confounding and statistical analysis.²⁰ Each study received an overall rating of low, moderate or high risk. The risk of bias was assessed by two independent reviewers.

Statistical analysis

We performed random effects meta-analysis using restricted maximum likelihood estimation with the Hartung-Knapp-Sidik-Jonkman adjustment for CIs, which provides more appropriate coverage than the standard DerSimonian-Laird method, particularly when the number of studies is limited.²¹ We report pooled HRs with 95% CIs and 95% prediction intervals.

Heterogeneity was quantified using I² and τ², with I² values of 25%, 50% and 75%, indicating low, moderate and substantial heterogeneity.

Prespecified subgroup analyses examined FFR-CT technology, geographic region, follow-up duration and study design. We performed leave-one-out sensitivity analysis and identified influential studies using Cook’s distance and Difference in Beta Statistics (DFBETAS).

Publication bias was assessed using funnel plots, Egger’s test, Begg’s test and trim-and-fill analysis. Evidence certainty was evaluated using GRADE methodology adapted for prognostic studies.²²

Analyses were conducted in R using the meta and metafor packages.

Results

Search results and study selection

Our database search identified 4521 records: 1002 from PubMed, 939 from Embase and 2580 from Scopus. After removing 1201 duplicates, we screened 3320 records and excluded 3256 at the title/abstract stage. We assessed 64 full-text articles and excluded 18, leaving 46 studies for qualitative synthesis (figure 1).

Figure 1. PRISMA flow diagram: illustrates the systematic search and selection process across three databases (PubMed, Embase, Scopus), yielding 4521 initial records. After removing 1201 duplicates and screening 3320 records, 46 studies underwent full-text review, with 22 studies (N=20 067 patients) ultimately included in the quantitative meta-analysis. The primary exclusion reasons were non-extractable hazard ratios (n=10), different comparisons (n=6) and missing coronary imaging data (n=4). PRISMA, Preferred Reporting Items for Systematic Reviews and Meta-Analyses. FU= Follow-up; CI= Confidence Interval.

Of these 46 studies, 24 could not contribute to the pooled analysis, 10 reported only risk ratios, ORs or risk differences rather than HRs; six used different comparisons (such as FFR-CT-guided strategy vs usual care); four had missing CIs; three had insufficient follow-up for the threshold comparison; and one had too few events (online supplemental table S2).

The final meta-analysis included 22 studies with 20 067 patients.1113 23,42

Characteristics of included studies

The 22 studies were published between 2018 and 2025 (table 1). Sample sizes ranged from 102 to 3674 (median 622). Follow-up ranged from 12 to 120 months (median ~40 months). Most studies were conducted in Asia (18 studies, predominantly China and Japan), with four from Western countries (Denmark, multinational).

Table 1. Characteristics of included studies.

Study	Year	Country	Design	N	Follow-up (M)	Technology	Outcome
Nørgaard	2018	Denmark	Retrospective	3674	24	HeartFlow	Composite
Li Z	2025	China	Retrospective	2566	72	DL	MACE
Wang Z	2025	China	Retrospective	1944	73	DL	MACE
Lan	2023	China	Prospective	1797	48	ML	MACE
Yu Y	2023	China	Retrospective	1334	25	DL	MACE
Yang J	2023	China	RCT	1216	12	DL	MACE
Liu Z	2023	China	Prospective	1215	60	DL	MACE
Wada	2025	Japan	Prospective	987	48	HeartFlow	MACE
Zhou F	2024	China	Prospective	783	35	DL	MACE
Sato	2024	Japan	Retrospective	745	60	On-site	Cardiac events
Wang ZQ	2024	China	Retrospective	622	72	DL	MACE
Li SY	2024	China	Prospective	500	36	ML	MACE
Qian	2025	China	Prospective	461	24	DL	MACE
Nozaki	2022	Japan	Retrospective	370	35	On-site	MACE
Dahl	2021	Denmark	Retrospective	340	40	HeartFlow	MACE
Liu Y	2021	China	Retrospective	284	48	ML	MACE
Huang W	2023	China	Retrospective	251	12	DL	MACE
Wang YY	2025	China	Retrospective	231	24	Other	MACE
Zhang R	2025	China	Prospective	222	22	ML	MACE
Liu J	2023	China	Retrospective	217	24	Other	MACE
Ihdayhid	2019	Multinational	Prospective	206	60	HeartFlow	Composite
Yang S	2024	Multinational	Prospective	102	120	HeartFlow	TVF

Twenty-two studies (N=20 067 patients) from 2018 to 2025 across Denmark, China, Japan and multinational cohorts, with prospective (n=9), retrospective (n=12) and RCT (n=1) designs. Follow-up ranged from 12 to 120 months, utilising HeartFlow (n=5), machine learning (n=4), deep learning (n=9) and on-site CFD (n=2) technologies, with MACE or composite endpoints as primary outcomes.

CFD, computational fluid dynamics; DL, deep learning; MACE, major adverse cardiovascular events; ML, machine learning; RCT, randomised controlled trial; TVF, Target Vessel Failure.

Study designs included prospective cohorts, retrospective cohorts and a single randomised trial.¹³ FFR-CT technologies varied: HeartFlow computational fluid dynamics (five studies),^{11 23 29 32 35} machine learning algorithms (four studies),^{27 34 36 39} deep learning (nine studies)^{13 24 25 28 31 33 37 38 42} and on-site computational methods (two studies).^{30 40} Two studies used other or unspecified technology.^{26 41}

Mean patient age ranged from 59 to 69 years. Male sex predominated 52%–76%. Diabetes prevalence varied from 11% to 100% (one study specifically enrolled diabetic patients),³² and hypertension affected 44%–73% of participants were reported.

Primary meta-analysis

Patients with FFR-CT ≤0.80 had nearly fourfold higher risk of MACE compared with those with FFR-CT >0.80: pooled HR 3.94 (95% CI 2.92 to 5.31; p<0.0001) (figure 2, table 2). The 95% prediction interval was 1.14 to 13.67, indicating considerable variability in expected effects across settings. Heterogeneity was substantial (I²=79.9%; τ²=0.34; Q=104.5, p<0.0001).

Figure 2. Forest plot of primary meta-analysis for MACE risk prediction: forest plot displaying HRs and 95% CIs for the primary meta-analysis across 22 studies, demonstrating significant association between FFR-CT and major adverse cardiovascular events (pooled HR 3.94, 95% CI 2.92 to 5.31, p<0.0001). Substantial heterogeneity was observed (I²=79.9%, p<0.0001) with prediction intervals ranging from 1.14 to 13.67, indicating variable effect sizes across included studies. Individual study HRs ranged from 0.88¹³ to 16.91,³³ with most studies favouring FFR-CT for MACE prediction. FFR-CT, Fractional flow reserve derived from CT; MACE, major adverse cardiovascular events; HK, Hartung-Knapp-Sidik-Jonkman adjustment for confidence intervals.

Table 2. Summary of meta-analysis results.

Analysis	k	HR (95% CI)	P	I² (%)	PI
Primary analysis	22	3.94 (2.92 to 5.31)	<0.0001	79.9	1.14–13.67
By technology (p=0.812)
HeartFlow CFD	5	4.29 (1.85 to 9.95)	0.0086	55.8	—
Machine learning	4	3.32 (2.02 to 5.47)	0.0046	0.0	—
Deep learning	9	4.19 (2.21 to 7.92)	0.0008	91.0	—
On-site CFD	2	5.67 (0.00 to 206 279) *	0.2829	65.4	—
Other	2	3.17 (0.27 to 37.15) *	0.1059	0.0	—
By geography (p=0.409)
Western	4	3.18 (1.46 to 6.90)	0.0178	15.6	—
Asia	18	4.05 (2.86 to 5.75)	<0.0001	83.1	—
By follow-up (p=0.443)
1–3 years	10	3.28 (1.86 to 5.79)	0.0011	87.2	—
3–5 years	5	4.07 (2.17 to 7.62)	0.0034	46.8	—
≥5 years	7	4.84 (3.13 to 7.47)	0.0001	45.7	—
By study design (p<0.0001)
Prospective	9	5.02 (2.96 to 8.50)	0.0001	72.0	—
Retrospective	12	3.85 (2.89 to 5.12)	<0.0001	48.3	—
RCT	1	0.88 (0.59 to 1.31)	0.5299	NA	—
Hard endpoints (death/MI)	6	3.28 (2.25 to 4.79)	0.0005	0.0	2.09–5.16
Sensitivity
Excluding outliers	20	3.51 (2.69 to 4.57)	<0.0001	74.4	—
HeartFlow only	5	4.29 (1.85 to 9.95)	0.009	55.8	—
Trim-and-fill adjusted	32	2.35 (1.62 to 3.41)	0.0001	88.4%	—

Primary analysis showed significant MACE association (HR 3.94, 95% CI 2.92 to 5.31, p<0.0001, I²=79.9%). Subgroup analyses revealed no significant differences by technology (p=0.812), geography (p=0.409) or follow-up duration (p=0.443). But it was significant for study design (p=<0.0001). Sensitivity analyses confirmed robustness, with hard endpoints showing minimal heterogeneity (I²=0.0%).

^*Imprecise estimate with extremely wide CI due to limited studies.

CFD, computational fluid dynamics; k, number of studies; MACE, major adverse cardiovascular event; MI, Myocardial Infarction; PI, prediction interval; RCT, randomised controlled trial.

Subgroup analyses

Effect estimates were generally consistent across subgroups (table 2, figure 3).

Figure 3. Subgroup analysis by FFR-CT technology type: forest plots stratified by FFR-CT computational technology demonstrate consistent prognostic value across different platforms: HeartFlow CFD (5 studies, HR 4.29, 95% CI 1.85 to 9.95, I²=55.8%), deep learning (9 studies, HR 4.19, 95% CI 2.21 to 7.92, I²=91.0%), machine learning (4 studies, HR 3.32, 95% CI 2.02 to 5.47, I²=0%), other technologies (2 studies, HR 3.17, 95% CI 0.27 to 37.15, I²=0%) and on-site CFD (2 studies, HR 5.67, 95% CI 0.00 to 206 279, I²=65.4%) and no significant subgroup differences were detected (p=0.812), suggesting technology-independent prognostic performance despite varying heterogeneity levels. CFD, computational fluid dynamics; FFR-CT, Fractional flow reserve derived from CT.

By technology

HeartFlow studies (k=5) yielded HR 4.29 (95% CI 1.85 to 9.95; I² = 55.8%). Machine learning studies (k=4) showed the most homogeneous results: HR 3.32 (95% CI 2.02 to 5.47; I²=0%). Deep learning studies (k=9) had HR 4.19 (95% CI 2.21 to 7.92) but very high heterogeneity (I²=91.0%), largely driven by the TARGET trial,¹³ which evaluated FFR-CT-guided management rather than threshold-based prognosis. On-site CFD (k=2) and other technologies (k=2) showed similar point estimates (HR 5.67 and 3.17, respectively) but with extremely wide CIs precluding meaningful interpretation. The test for subgroup differences was not significant (p=0.812).

By geography

Western studies (k=4) showed HR 3.18 (95% CI 1.46 to 6.90; I²=15.6%), while Asian studies (k=18) showed HR 4.05 (95% CI 2.86 to 5.75; I²=83.1%). The lower heterogeneity among Western studies likely reflects their more standardised methodology and predominant use of HeartFlow technology. The test for subgroup differences was not significant (p=0.409) (online supplemental figure S1).

By follow-up duration

Studies were categorised into three follow-up periods. Short-term studies (1–3 years; k=10) showed HR 3.28 (95% CI 1.86 to 5.79; I²=87.2%). Medium-term studies (3–5 years; k=5) demonstrated HR 4.07 (95% CI 2.17 to 7.62; I²=46.8%). Long-term studies (≥5 years; k=7) yielded the highest estimate: HR 4.84 (95% CI 3.13 to 7.47; I²=45.7%). This gradient suggests the prognostic value of abnormal FFR-CT may strengthen over extended observation, though the test for subgroup differences was not significant (p=0.443) (online supplemental figure S2).

By study design

Studies were categorised into three design types. Retrospective studies (k=13) showed a pooled HR of 3.85 (95% CI 2.89 to 5.12; I²=48.3%), while prospective studies (k=10) demonstrated a higher estimate of HR 5.02 (95% CI 2.96 to 8.50; I²=72%). The single randomised controlled trial yielded the lowest estimate: HR 0.88 (95% CI 0.59 to 1.31). These differences across study designs suggest that retrospective and prospective observational studies consistently identified a significant prognostic association, whereas the only available RCT did not. Notably, among all subgroup analyses performed, study design was the only subgrouping for which the test for subgroup differences reached statistical significance (p<0.0001), highlighting study design as a meaningful source of heterogeneity in this meta-analysis (online supplemental figure S3).

Hard endpoints analysis

Six studies reported outcomes restricted to death and/or myocardial infarction. The pooled HR was 3.28 (95% CI 2.25 to 4.79; p=0.0005) with no heterogeneity whatsoever (I²=0.0%; Q=3.46, p=0.63). The prediction interval was correspondingly narrow: 2.09 to 5.16. This remarkable consistency suggests that hard endpoints may be more reliable than composite MACE for assessing FFR-CT prognostic value (online supplemental figure S4).

Sensitivity analyses

Leave-one-out analysis

Sequentially excluding each study, the pooled HR ranged from 3.60 to 4.29, a reassuringly stable result (online supplemental table S3), (online supplemental figure S5). The most influential studies were Yang et al. 2023 (TARGET),¹³ whose exclusion raised HR to 4.29 and dropped I² to 62.1%, and Zhou et al. 2024³³ whose exclusion lowered HR to 3.60.

Influence diagnostics

Two studies met criteria for outliers: Yang et al. 2023 (TARGET)¹³ with a protective effect (HR 0.88), and Zhou et al. 2024³³ with an unusually large effect (HR 16.91). The protective finding in TARGET likely reflects that both trial arms received appropriate FFR-CT-guided treatment. Excluding both outliers yielded HR 3.91 (95% CI 3.17 to 4.82) with dramatically reduced heterogeneity (I² = 39.0%) (online supplemental table S4, online supplemental figure S6).

HeartFlow-only analysis

Restricting to HeartFlow studies (k=5) for direct comparison with ⁹ gave HR 4.29 (95% CI 1.85 to 9.95; I²=55.8%).

Risk of bias assessment

Among 22 studies, 17 (77%) had low overall risk of bias and 5 (23%) had moderate risk. None had high risk. The most common concerns were potential selection bias in retrospective designs and incomplete confounder adjustment (online supplemental figures S7 and S8).

Publication bias

The funnel plot showed some asymmetry (online supplemental figure S9). Neither Egger’s test (p=0.13) nor Begg’s test (p=0.35) reached significance. However, trim-and-fill analysis imputed 10 potentially missing studies, yielding an adjusted HR of 2.35 (95% CI 1.62 to 3.42). Even with this conservative adjustment, abnormal FFR-CT remains associated with more than doubled MACE risk.

GRADE assessment

Following GRADE Guidelines 28 for prognostic factor research,²² we rated certainty of evidence starting from HIGH as observational studies represent the optimal design for prognosis research. We downgraded one level for inconsistency (substantial heterogeneity I²=79.9%, wide prediction interval). We did not downgrade for publication bias because the trim-and-fill adjusted estimate (HR 2.35) remained statistically significant and clinically meaningful. No serious concerns were identified for risk of bias, indirectness or imprecision. Final certainty: MODERATE (online supplemental table S5).

Comparison with prior meta-analysis

When comparing our findings with Nørgaard et al,⁹ the meta-analysis included five studies with 5460 patients and reported that the primary endpoint (death or MI at 12 months) occurred in just 60 patients (1.1% overall), 0.6% (13/2126) with FFRCT >0.80 vs 1.4% (47/3334) with FFRCT ≤0.80, yielding RR 2.31 (95% CI 1.29 to 4.13, p=0.005) with I²=0%. They also found that each 0.10-unit FFRCT reduction was associated with RR 1.67 (95% CI 1.47 to 1.87).

Our update substantially expands this evidence base: 22 studies, 20 067 patients, follow-up extending to 10 years and multiple FFR-CT technologies beyond HeartFlow alone. Our larger pooled effect (HR 3.94 vs RR 2.31) likely reflects several factors: longer follow-up capturing more events, inclusion of revascularisation in most composite endpoints and the shift from risk ratios to HRs, which better account for time-to-event. Our higher heterogeneity (I²=79.9% vs 0%) reflects the now-substantial diversity in technologies, populations and study designs. Notably, when we restricted to hard endpoints (death/MI only), heterogeneity dropped to 0%, mirroring the homogeneity Nørgaard observed.

Discussion

Principal findings

This systematic review and meta-analysis represents the most comprehensive assessment of FFR-CT prognostic value to date, pooling data from 22 studies with over 20 000 patients followed for up to 10 years. Our principal finding, that FFR-CT ≤0.80 is associated with a nearly fourfold increased risk of MACEs compared with normal FFR-CT substantially, extends the evidence base established by Nørgaard and colleagues in 2022,⁹ whereas that analysis included five studies with 5460 patients and maximum 12-month follow-up, we now provide effect estimates derived from more than four times as many patients with median follow-up approaching 3 years. Perhaps most importantly, our finding that hard endpoints (death and myocardial infarction) show zero heterogeneity across studies suggests that FFR-CT provides consistent prognostic discrimination regardless of technology platform, geographic setting or patient population.

Comparison with other prognostic modalities

The prognostic value we observed for FFR-CT closely mirrors that established for invasive fractional flow reserve in landmark trials. The DEFER trial, with 15-year follow-up, demonstrated approximately 4.8-fold higher rates of cardiac death and myocardial infarction in patients with functionally significant lesions compared with those with FFR above threshold.⁴³ Similarly, the FAME 2 trial showed that patients with FFR ≤0.80 randomised to medical therapy alone experienced event rates roughly double those of patients undergoing FFR-guided revascularisation, with HRs around 0.46 favouring intervention.⁴⁴ These findings from invasive assessment provide important context for interpreting our FFR-CT results. The close agreement between non-invasive and invasive FFR prognostic discrimination, both identifying approximately three to fourfold risk differences, supports the physiological validity of CT-derived measurements and their potential to replace invasive assessment for risk stratification purposes.

Our results also compare favourably with anatomic assessment alone. The Coronary CT Angiography Evaluation for Clinical Outcomes: An International Multicenter (CONFIRM) Registry, encompassing nearly 24 000 patients, established that obstructive coronary disease on CTA carries HRs ranging from 2.77 for single-vessel disease to 7.91 for three-vessel or left main involvement.⁴⁵ More recent work validating the CAD-RADS reporting system has shown HRs of 3.5–4.0 for moderate stenosis categories.⁴⁶ Our pooled FFR-CT estimate of 3.94 falls squarely within this range, suggesting that functional assessment provides risk stratification comparable to anatomic grading. However, FFR-CT offers a critical advantage: it identifies haemodynamic significance that pure anatomic assessment cannot capture. A stenosis classified as CAD-RADS 3 or 4 may or may not be flow limiting, whereas FFR-CT ≤0.80 confirms functional ischaemia regardless of visual stenosis severity.

When compared with other non-invasive functional imaging modalities, FFR-CT demonstrates prognostic discrimination that is remarkably consistent with stress cardiac MRI and comparable to or better than nuclear techniques. A recent meta-analysis of stress CMR encompassing nearly 59 000 patients reported a HR of 3.90 for MACE with inducible ischaemia,⁴⁷ virtually identical to our FFR-CT finding of 3.94. Stress myocardial perfusion echocardiography shows similarly strong prognostic value, with pooled HRs around 4.75 in meta-analytic estimates.⁴⁸ In contrast, positron emission tomography myocardial perfusion imaging demonstrates somewhat lower prognostic discrimination, with HRs typically around 2.30 for abnormal perfusion.⁴⁹ These comparisons position FFR-CT among the most prognostically powerful non-invasive cardiac imaging modalities currently available, with the added advantage of deriving functional information from a single anatomic CT acquisition without additional stress testing, radiation exposure from nuclear tracers or contraindications associated with pharmacologic stress agents.

The relationship between FFR-CT and coronary artery calcium scoring deserves particular attention. The Multi-Ethnic Study of Atherosclerosis established that CAC scores above 300 carry HRs exceeding 9.0 compared with zero calcium,⁵⁰ and subsequent work in symptomatic populations has confirmed strong prognostic value with risk ratios around 5.7 for any detectable calcium.⁵¹ However, CAC reflects atherosclerotic burden rather than haemodynamic significance, and high calcium scores paradoxically reduce FFR-CT diagnostic accuracy due to blooming artefacts. Our finding that FFR-CT maintains prognostic value across the evidence base suggests it provides complementary information to calcium scoring. Indeed, data from the ADVANCE-DK registry demonstrated that adding FFR-CT to models containing CAC and stenosis improved the area under the curve from 0.62 to 0.74,¹² confirming incremental prognostic value of functional assessment beyond anatomic markers.

Technology platform equivalence and heterogeneity

A notable finding of our analysis is the absence of significant differences in prognostic performance across FFR-CT technologies. HeartFlow computational fluid dynamics, machine learning algorithms and deep learning approaches all demonstrated similar HRs, with the test for subgroup differences yielding p=0.812. This equivalence aligns with diagnostic accuracy data showing that on-site tools achieve 84.1% accuracy compared with 79.4% for off-site processing,⁵² and with the MACHINE consortium’s finding of near-perfect correlation (R=0.997) between machine learning and computational fluid dynamics approaches.⁵³ The clinical implication is substantial: as multiple platforms achieve regulatory clearance and broader availability, clinicians can expect consistent prognostic information regardless of which FFR-CT algorithm their institution employs. This technology equivalence supports the broader adoption of FFR-CT beyond centres with access to specific vendor platforms.

The heterogeneity patterns in our analysis merit careful interpretation. Overall heterogeneity was substantial for the composite MACE endpoint (I²=79.9%), yet dropped to zero when restricting to hard endpoints of death and myocardial infarction. This discrepancy likely reflects two related phenomena. First, MACE definitions vary considerably across cardiovascular research, a systematic review found that fewer than 9% of observational studies use standardised three-component definitions.⁵⁴ Second, revascularisation endpoints are particularly susceptible to practice pattern variation across centres and geographic regions; studies have documented 25% or greater variation in percutaneous intervention rates across institutions.⁵⁵ The inclusion of revascularisation in MACE composites creates a self-fulfilling prophecy: abnormal FFR-CT identifies lesions that clinicians then revascularise, and that revascularisation is counted as an adverse event. This circularity does not apply to death or myocardial infarction, which are objectively defined and not subjected to physician discretion. We, therefore, consider the hard endpoint analysis (HR 3.28, I²=0%) to represent the most reliable estimate of FFR-CT prognostic value, free from definitional variability and practice pattern confounding.

Geographical variations

The geographic composition of our analysis—with 82% of studies from Asian populations—warrants consideration. Asian populations demonstrated a numerically higher hazard ratio (4.05) compared with Western populations (3.18), though this difference was not statistically significant (p=0.41). Several factors may contribute to this observation. Asian populations, particularly those in East Asia, have undergone rapid epidemiological transition with increasing prevalence of metabolic syndrome, diabetes and obesity.⁵⁶ Studies have documented that Asian individuals develop insulin resistance and cardiovascular risk at lower body mass index thresholds compared with Western populations, leading the WHO to recommend different BMI cut-offs for defining overweight and obesity in Asians.⁵⁷ Additionally, the prevalence of diabetes in our Asian cohorts ranged as high as 100% in some studies, reflecting both the metabolic burden of these populations and the use of FFR-CT in high-risk patients. Despite these differences, the similarity of Western and Asian point estimates provides reassurance that FFR-CT prognostic value generalises across diverse populations with varying cardiometabolic profiles.

Clinical implications

Our findings directly address evidence gaps identified by major clinical guidelines. The ESC 2024 chronic coronary syndromes guidelines acknowledge FFR-CT’s diagnostic utility but assign only a Class IIb recommendation, noting limited availability and need for further validation.³ The Society of Cardiovascular Computed Tomography (SCCT) 2021 expert consensus explicitly stated that intermediate-term prognostic data for FFR-CT had not been previously reported in real-world settings.⁵⁸ Our meta-analysis, incorporating studies with follow-up extending to ten years and representing diverse international populations, provides the prognostic evidence these guidelines call for. The consistency of our findings across technologies and populations strengthens the case for upgrading FFR-CT recommendations from diagnostic adjunct to established prognostic tool.

Limitations

Several limitations warrant acknowledgement. The predominance of Asian studies (82%) raises questions about generalisability, though Western studies showed similar point estimates (HR 3.18). Publication bias appears present based on funnel plot asymmetry, and trim-and-fill adjustment reduced our estimate to 2.35, still clinically meaningful but suggesting some overestimation in the primary analysis. We pooled studies using HRs, excluding those reporting only risk ratios or ORs, which may have introduced selection effects. Additionally, the timing of FFR-CT assessment relative to symptom onset or initial diagnosis was not consistently reported across studies, precluding subgroup analysis by disease stage. Future primary studies should systematically report timing of FFR-CT to enable such analyses. Finally, outcome definitions varied across studies, though our subgroup analysis of hard endpoints addresses this concern.

Conclusions

In conclusion, this meta-analysis demonstrates that FFR-CT ≤0.80 identifies patients at nearly fourfold increased risk of MACEs, with consistent prognostic value across technologies, geographic regions and follow-up durations. The prognostic discrimination of FFR-CT is comparable to invasive FFR and stress cardiac MRI, superior to nuclear perfusion imaging and provides incremental value beyond anatomic assessment alone. These findings support the integration of FFR-CT into clinical pathways not merely as a diagnostic gatekeeper but as an independent prognostic marker for cardiovascular risk stratification.

Data availability statement

Data are available in a public, open access repository. Data are available upon reasonable request. All data relevant to the study are included in the article or uploaded as supplementary information.