Coronary computed tomography angiography plaque and flow patterns in acute coronary syndrome lesions

Summary

Acute coronary syndrome (ACS) risk stratification requires assessment across different time frames. This multicenter study investigated whether a single coronary computed tomography angiography (CCTA) scan could provide integrated short-term (≤7 days) and long-term (≥30 days) risk assessment. The study developed and validated dual-timeframe models combining population health data, stenosis severity, quantitative low-density plaque burden (LDP%), and the hemodynamic parameter ΔCT-FFR (CT-derived fractional flow reserve [CT-FFR] proximal to the lesion minus CT-FFR distal to the lesion). Key findings indicate that while stenosis and LDP% predicted ACS risk in both time frames, ΔCT-FFR was specifically associated with short-term risk. The combined models demonstrated superior performance and greater net clinical benefit compared to using stenosis severity alone, as confirmed in multiple external cohorts. These results establish that a single CCTA examination can simultaneously assess immediate and future ACS risk.

Graphical abstract

Highlights

• •This is the first dual-timeframe ACS risk model from a single CCTA scan.
• •ΔCT-FFR predicts short-term ACS risk; while stenosis and LDP predict short- and long-term risks
• •The dual-timeframe ACS model improves discrimination and net benefit over stenosis alone

Health Science

Introduction

Acute coronary syndrome (ACS) remains a leading cause of morbidity and mortality worldwide.¹ Histopathological and intracoronary imaging studies have shown that ACS primarily results from the erosion or rupture of atherosclerotic plaques.^2,3 Plaque rupture occurs due to the interaction between the plaque’s intrinsic vulnerability and the extrinsic hemodynamic forces acting on its surface.⁴ In addition to assessing the severity of stenosis, coronary computed tomography angiography (CCTA) can noninvasively characterize adverse plaque characteristics (APCs) and provide quantitative assessments of plaque components.⁵ With advanced computational fluid dynamics analysis, CCTA can also noninvasively generate simulated fractional flow reserve (FFR) results. Numerous studies have confirmed that CT-derived FFR (CT-FFR) correlates well with invasive FFR results.⁶ These quantitative and qualitative plaque and hemodynamic parameters derived from CCTA are useful for predicting future ACS.^7,8

The 2024 European Society of Cardiology (ESC) guidelines for managing chronic coronary syndromes (CCSs)⁹ emphasize the need to integrate both short-term and long-term risk stratification for patients with suspected CCS. Physicians are advised to identify culprit lesions and assess culprit lesion precursors that may lead to adverse events in patients with ACS with multivessel disease. Nearly 50% of patients with ST-segment elevation myocardial infarction (STEMI) reportedly have multivessel coronary artery disease (CAD).¹⁰ Progressive non-culprit coronary lesions have been shown to be independent risk factors for future major adverse cardiac events (MACEs).¹¹ Quantitative plaque analysis using CCTA has revealed a 4-year residual risk in 23.9% of patients with non-ST-segment elevation ACS.¹² The COMPLETE trial (Complete versus Culprit-Only Revascularization Strategies to Treat Multivessel Disease after Early PCI for STEMI) demonstrated that complete revascularization significantly reduces the risk of new myocardial infarction compared to managing only the culprit lesion.¹³ The latest coronary revascularization guidelines from the American Heart Association recommend staged percutaneous coronary intervention (PCI) for severely stenosed non-infarct-related arteries in patients with multivessel STEMI (Class I A recommendation).¹⁴ This underscores the importance of identifying patients who need complete or target vessel revascularization to optimize the clinical management of ACS and CCS.

The objective of this study is to establish a dual model based on the index CCTA that can simultaneously serve distinct clinical scenarios of ACS. We hypothesize that quantitative and qualitative plaque and hemodynamic parameters derived from a single CCTA scan can address two critical challenges in the management of ACS: short-term and long-term risk stratification. By elucidating both the commonalities and differences in plaque characteristics across these two time frames (≤7 days and ≥30 days), this study seeks to demonstrate the capability of a single CCTA examination to concurrently facilitate short-term and long-term risk stratification.

Results

Baseline clinical features of patients in the derivation cohort

The baseline clinical features of the derivation cohort are presented in Tables S2 and S3. Of the 1,688 patients enrolled in the short-term ACS risk stratification group from the derivation cohort, the median age was 65.0 years (interquartile range [IQR], 58.0–73.0), 62.3% (1051/1688) were male, and 154 patients (9.1%) experienced new-onset ACS within 7 days (median interval between CCTA and outcome: 3.4 days) (High exclusion rate [n = 2,443, 1,528 plaque-free] with overall actual ACS event rate of 3.7% in screened population [n = 4,131]). Aside from clinical symptoms at baseline, no statistically significant differences were observed in population health data between the ACS and CCS groups (all p > 0.05). A total of 906 patients were included in the long-term ACS risk stratification group from the derivation cohort, with a median age of 66.0 years (IQR, 58.0–74.0). Among them, 61.5% (557/906) were male, and the median follow-up time was 42.1 months (IQR, 15.7–73.8). Within this cohort, 41 patients developed new-onset ACS by May 31, 2024. At baseline, there were no significant differences in demographic health data or clinical manifestations between the ACS and CCS groups (all p > 0.05).

Coronary computed tomography angiography-based plaque features in culprit vs. non-high-risk lesions, acute coronary syndrome precursors vs. stable lesions

Compared to non-high-risk lesions, the culprit lesions exhibited more severe stenosis (p < 0.001), involved more segments (p < 0.001), and had greater length (19.0 mm vs. 26.4 mm, p < 0.001). Additionally, culprit lesions demonstrated a higher percentage of high-risk plaque (HRP) (32.3% vs. 42.9%, p = 0.008), primarily positive remodeling (PR) (50.9% vs. 64.3%, p = 0.002). Culprit lesions contained more fibro-fatty (14.0% vs. 15.2%, p < 0.001) and low-density (12.4% vs. 16.3%, p < 0.001) components, along with lower lesion-specific CT-FFR values (0.92 vs. 0.86, p < 0.001) and higher ΔCT-FFR (CT-FFR proximal to the lesion minus CT-FFR distal to the lesion) values (0.05 vs. 0.10, p < 0.001). Similar results were observed when comparing ACS precursors with stable lesions (Table 1).

Table 1.

Differences in baseline imaging features between culprit and non-high-risk lesions, culprit lesion precursors and stable lesions in the derivation cohort

Variables	Short-term ACS risk stratification (n = 1688)		P1 value	Long-term ACS risk stratification (n = 906)		P2 value
	Culprit lesions (n = 154)	Non-high-risk lesions (n = 1534)		Culprit lesion precursors (n = 41)	Stable lesions (n = 865)
Vessel, n (%)			0.016			0.903
LMa	57 (37.0)	394 (25.7)		11 (26.8)	226 (26.1)
LAD	65 (42.2)	826 (53.9)		20 (48.8)	466 (53.9)
LCX	18 (11.7)	179 (11.7)		3 (7.3)	52 (6.0)
RCA	15 (9.1)	135 (8.8)		7 (17.1)	121 (14.0)
Location, n (%)			0.785			0.539
Proximal	135 (87.7)	1317 (85.9)		33 (80.5)	748 (86.5)
Mild	18 (11.7)	201 (13.1)		7 (17.1)	105 (12.1)
Distal	1 (0.7)	16 (1.0)		1 (2.4)	12 (1.4)
Lesion nature, n (%)			0.002			0.251
Calcified plaque	17 (11.0)	361 (23.5)		7 (17.1)	223 (25.8)
Non-calcified plaque	50 (32.5)	413 (26.9)		14 (34.1)	209 (24.2)
Mixed plaque	87 (56.5)	760 (49.5)		20 (48.8)	433 (50.1)
Stenosis severityb, n (%)			<0.001			<0.001
<70%	103 (66.9)	1402 (91.4)		30 (73.2)	808 (93.4)
≥70%	51 (33.1)	132 (8.6)		11 (26.8)	57 (6.6)
Involved segments, n (%)			<0.001			0.828
1	46 (29.9)	781 (50.9)		19 (46.3)	412 (47.6)
2	75 (48.7)	590 (38.5)		17 (41.5)	367 (42.4)
3	33 (21.4)	160 (10.4)		5 (12.2)	85 (9.8)
4	0 (0.0)	3 (0.2)		0 (0.0)	1 (0.1)
Lesion length, mm	26.4 (13.6–42.4)	19.0 (10.5–32.9)	<0.001	20.8 (13.1–30.6)	18.5 (10.2–32.7)	0.271
APCs, n (%)
Positive remodeling	99 (64.3)	781 (50.9)	0.002	30 (73.2)	431 (49.8)	0.003
Spotty calcification	31 (20.1)	326 (21.3)	0.745	6 (14.6)	189 (21.8)	0.272
Low-attenuation plaque	57 (37.0)	420 (27.4)	0.011	12 (29.3)	212 (24.5)	0.490
Napkin ring sign	13 (8.4)	68 (4.4)	0.027	4 (9.8)	57 (6.6)	0.349
HRP (≥2 APCs), n (%)	66 (42.9)	496 (32.3)	0.008	16 (39.0)	262 (30.3)	0.236
Quantitative plaque volume (mm3)
Vessel	426.7 (198.1–709.9)	307.6 (161.4–542.5)	<0.001	409.0 (197.4–670.4)	307.4 (159.1–533.0)	0.156
Dense calcium	10.1 (0.1–54.1)	8.7 (0.5–40.9)	0.868	7.3 (0.1–54.5)	10.4 (1.0–44.1)	0.840
Fibrous	65.7 (33.5–129.5)	52.8 (26.9–100.2)	0.003	59.8 (32.4–115.8)	52.3 (27.1–99.6)	0.340
Fibro-fatty	63.2 (28.3–102.9)	43.5 (20.3–78.8)	<0.001	52.1 (26.2–93.3)	42.8 (19.2–75.8)	0.063
Low-density plaque	63.2 (24.4–114.0)	40.7 (12.1–81.8)	<0.001	58.0 (37.5–156.1)	37.1 (10.7–77.7)	<0.001
Percentage of components (%)
Dense calcium	2.5 (0.0–12.8)	4.3 (0.2–13.8)	0.200	4.9 (0.0–11.6)	5.0 (0.3–15.0)	0.341
Fibrous	18.4 (14.3–21.8)	18.0 (14.9–21.4)	0.938	19.4 (10.3–59.6)	17.5 (8.0–37.2)	0.337
Fibro-fatty	15.2 (12.0–19.0)	14.0 (10.6–17.1)	<0.001	14.0 (12.1–18.3)	13.7 (10.2–16.8)	0.151
Low-density plaque	16.3 (8.1–28.7)	12.4 (5.3–21.3)	<0.001	22.8 (11.2–28.0)	11.5 (4.8–20.1)	<0.001
Lesion-specific CT-FFR	0.86 (0.77–0.92)	0.92 (0.85–0.95)	<0.001	0.89 (0.83–0.93)	0.92 (0.87–0.95)	0.008
ΔCT-FFR	0.10 (0.05–0.20)	0.05 (0.02–0.10)	<0.001	0.06 (0.05–0.12)	0.04 (0.02–0.09)	0.010

Values are mean ± standard deviation, number (percentage), or median (interquartile range). n = patients per group.

Group comparisons for categorical variables were performed using the Pearson chi-square test or Fisher’s exact test, as appropriate. Continuous variables were compared using the Mann–Whitney U test, depending on data distribution. P1 values correspond to comparisons between culprit and non-high-risk lesions in the short-term stratification group; P2 values correspond to comparisons between culprit lesion precursors and stable lesions in the long-term stratification group.

ACS, acute coronary syndrome; LM, left main artery; LAD, left anterior descending; LCX, left circumflex artery; RCA, right coronary artery; APC, adverse plaque characteristics; HRP, high-risk plaque; CT-FFR, CT-derived fractional flow reserve; ΔCT-FFR, CT-FFR proximal to the lesion minus CT-FFR distal to the lesion.

^alesions involving LM.

^bCulprit lesions in patients with ACS vs. maximum stenosed lesions in the CCS group.

Association of coronary computed tomography angiography-based plaque features with culprit lesions and acute coronary syndrome precursors

To better understand the contribution of each variable to ACS, we categorized the models as follows: Model 1: includes only population health data; Model 2: based on Model 1, qualitative CCTA plaque characteristics were incorporated, and population health data were fitted using logarithmic transformation; and Model 3: on the basis of Model 2, quantitative CCTA plaque characteristics were incorporated. The results indicated that population health data alone are insufficient for the risk stratification of ACS, either in the short or long term. Multivariable logistic regression analysis showed that when qualitative CCTA parameters were considered, ≥70% stenosis (odds ratio [OR]: 4.60, 95% confidence intervals [CI]: 2.92–7.23, p < 0.001) was the only independent predictor of the occurrence of culprit lesions within 7 days, apart from the synthesized log odds derived from population health data (OR: 2.65, 95% CI: 1.54–4.55, p < 0.001). However, after incorporating quantitative CCTA parameters, both low-density plaque burden (LDP%) (OR per 10-unit change: 1.25, 95% CI: 1.05–1.48, p = 0.010) and ΔCT-FFR (OR per standard deviation [SD] change: 1.42, 95% CI: 1.10–1.83, p = 0.007) also demonstrated independent predictive value for culprit lesions, apart from the synthesized log odds derived from population health data (OR: 2.65, 95% CI: 1.54–4.57, p < 0.001) (Table 2). Unlike the short-term ACS risk stratification group, in the competing risk Cox regression analysis, it was found that in addition to ≥70% stenosis (hazard ratio [HR]: 6.16, 95% CI: 2.44–15.55; p < 0.001), only LDP% (HR per 10 units change: 1.31, 95% CI: 1.01–1.71; p = 0.045) was independently associated with the future ACS event, apart from the synthesized log odds derived from population health data (HR: 1.80, 95% CI: 1.06–3.08, p = 0.049) (Table 3).

Table 2.

Variables for short-term risk stratification in the derivation cohort

Variables	Short-term ACS risk stratification
	Univariable analysis		Multivariable analysis
	OR (95% CI)	p value	OR (95% CI)	p value
Model 1: Population health data
Age (per 10 units)	1.05 (0.90–1.22)	0.510
Sex (male)	1.25 (0.88–1.77)	0.215
Hypertension	1.47 (0.99–2.16)	0.054
Diabetes mellitus	0.90 (0.61–1.34)	0.613
Hyperlipidemia	0.91 (0.64–1.30)	0.603
Current smoker	1.34 (0.94–1.92)	0.109
Current drinker	1.27 (0.87–1.85)	0.215
Model 2: CCTA qualitative features
Propensity score of population health dataa	2.72 (1.61–4.58)	<0.001	2.65 (1.54–4.55)	<0.001
Stenosis severity
<70%	Ref.		Ref.
≥70%	5.26 (3.60–7.69)	<0.001	4.40 (2.81–6.89)	<0.001
Positive remodeling
Negative	Ref.		Ref.
Positive	1.74 (1.23–2.45)	0.002	1.40 (0.96–2.03)	0.082
Low-attenuation plaque
Negative	Ref.		Ref.
Positive	1.56 (1.10–2.20)	0.012	1.25 (0.83–1.88)	0.296
Napkin ring sign
Negative	Ref.		Ref.
Positive	1.99 (1.07–3.69)	0.029	0.94 (0.46–1.93)	0.872
Lesion-specific CT-FFR
>0.80	Ref.		Ref.
≤0.80	2.70 (1.87–3.91)	<0.001	1.27 (0.81–1.98)	0.301
Model 3: CCTA qualitative and quantitative features of CCTA
Propensity score of population health dataa	2.72 (1.61–4.58)	<0.001	2.65 (1.54–4.57)	<0.001
Stenosis severity
<70%	Ref.	<0.001	Ref.	<0.001
≥70%	5.26 (3.60–7.69)		3.30 (2.05–5.30)
Lesion length (per 10-units), mm	1.24 (1.14–1.35)	<0.001	1.04 (0.92–1.17)	0.525
Positive remodeling
Negative	Ref.		Ref.
Positive	1.74 (1.23–2.45)	0.002	1.08 (0.72–1.62)	0.722
Low-attenuation plaque
Negative	Ref.		Ref.
Positive	1.56 (1.10–2.20)	0.012	1.03 (0.66–1.59)	0.907
Napkin ring sign
Negative	Ref.		Ref.
Positive	1.99 (1.07–3.69)	0.029	0.94 (0.45–1.95)	0.871
Lesion-specific CT-FFR
>0.80	Ref.		Ref.
≤0.80	2.70 (1.87–3.91)	<0.001	0.56 (0.29–1.10)	0.094
FF (per 10-units), %	1.87 (1.36–2.59)	<0.001	1.15 (0.78–1.69)	0.492
Low-density plaque (per 10-units), %	1.45 (1.27–1.65)	<0.001	1.25 (1.05–1.48)	0.010
ΔCT-FFR (pe SD)	1.61 (1.42–1.83)	<0.001	1.42 (1.10–1.83)	0.007

Associations between variables and short-term ACS risk were assessed using univariable and multivariable binary logistic regression models, with results presented as ORs and 95% CIs. Variables with P-values ≤0.05 in the univariable regression analysis are included in the subsequent multivariate logistic regression analysis.

ACS, acute coronary syndrome; OR, odds ratio; 95% CI, 95% confidence interval; CCTA, coronary computed tomography angiography; PS, propensity score; CT-FFR, CT-derived fractional flow reserve; SD, standard deviation; FF, fibro-fatty; ACEI, angiotensin-converting enzyme inhibitor; ARB, angiotensin receptor blocker; PS, propensity scores; ΔCT-FFR, CT-FFR proximal to the lesion minus CT-FFR distal to the lesion.

^aGiven the sample size limitations in this study, population health data (age, sex, current smoking, alcohol consumption, hypertension, diabetes mellitus, hyperlipidemia, and medication use [antiplatelet drugs, statin, anticoagulant, beta-blocker, and ACEI/ARB]) were used to simulate a propensity score in multivariable regression models 2–3. The propensity score was then converted into log odds using the formula: log [PS/(1 - PS)]. These log odds were subsequently included in the regression analysis.

Table 3.

Variables for long-term risk stratification in the derivation cohort

Variables	Long-term ACS risk stratification
	Univariable analysis		Multivariable analysis
	HR (95% CI)	p value	HR (95% CI)		p value
Model 1: Population health data
Age (per 10 units)	0.82 (0.61–1.09)	0.174
Sex (male)	1.53 (0.78–3.00)	0.215
Hypertension	1.79 (0.79–4.03)	0.162
Diabetes mellitus	0.97 (0.50–1.91)	0.939
Hyperlipidemia	1.38 (0.74–2.59)	0.312
Current smoker	1.34 (0.67–2.68)	0.412
Current drinker	0.76 (0.32–1.81)	0.537
Model 2: CCTA qualitative features
Propensity score of population health dataa	2.01 (1.20–3.38)	0.008	1.91 (1.11–3.30)		0.020
Stenosis severity
<70%	Ref.		Ref.
≥70%	6.44 (3.02–13.72)	<0.001	5.71 (2.60–12.57)		<0.001
Positive remodeling
Negative	Ref.		Ref.
Positive	2.30 (1.06–4.99)	0.035	1.77 (0.78–4.03)		0.174
Lesion-specific CT-FFR
>0.80	Ref.		Ref.
≤0.80	2.70 (1.24–5.88)	0.013	1.36 (0.58–3.15)		0.479
Model 3: CCTA qualitative and quantitative features of CCTA
Propensity score of population health dataa	2.01 (1.20–3.38)	0.008		1.80 (1.06–3.08)	0.031
Stenosis severity
<70%	Ref.			Ref.
≥70%	6.44 (3.02–13.72)	<0.001		6.16 (2.44–15.55)	<0.001
Positive remodeling
Negative	Ref.			Ref.
Positive	2.30 (1.06–4.99)	0.035		1.36 (0.57–3.23)	0.485
Lesion-specific CT-FFR
>0.80	Ref.		Ref.
≤0.80	2.70 (1.24–5.88)	0.013	1.91 (0.49–7.46)		0.350
Low-density plaque (per 10 units), %	1.56 (1.23–1.97)	<0.001	1.31 (1.01–1.71)		0.045
ΔCT-FFR (per SD)	1.32 (1.09–1.61)	0.006	0.79 (0.44–1.40)		0.419

Associations between variables and long-term ACS risk were assessed using univariable and multivariable Cox proportional hazards regression models, with results presented as HRs and 95% CIs. Variables with P-values ≤0.05 in the univariable regression analysis are included in the subsequent multivariable Cox regression analysis.

HR, hazard ratio. All other abbreviations are consistent with those in Table 2.

^aSame as Table 2.

Evaluation of the models

In the short-term ACS risk stratification group, the combined model—incorporating the propensity score derived from population health data, stenosis severity, ΔCT-FFR (per SD decrease), and LDP% (per 10-unit increase)—demonstrated superior discriminatory performance in differentiating culprits from non-high-risk lesions compared to the model based on stenosis severity alone (areas under the curve [AUC]: 0.70, 95% CI: 0.68–0.72 vs. 0.62, 95% Cl: 0.60–0.65, p < 0.001) (Figure 1A). Figure 1B shows a well-calibrated combined model, without significant evidence of poor fit indicated by the Hosmer-Lemeshow test (p = 0.421). Figure 1C shows that the combined model provides a greater net clinical benefit than the stenosis severity model over a wide range of risk thresholds, according to decision curve analysis (DCA).

Figure 1.

Comprehensive performance assessment of risk models for short-term and long-term ACS stratification in the derivation cohort

(A–C) Short-term ACS stratification results. Model performance was evaluated using ROC curves with DeLong’s test for comparison (A), calibration curves with the Hosmer–Lemeshow test (B), and DCA (C).

(A) ROC curve for short-term ACS risk stratification. The combined model, incorporating the propensity score derived from population health data, stenosis severity, ΔCT-FFR (per SD decrease), and low-density plaque burden (per 10-unit increase) shows superior discriminatory ability in distinguishing between culprit and non-high-risk lesions compared to the model based on stenosis severity alone.

(B) The calibration curve demonstrated that the combined model was well-calibrated, a finding supported by a non-significant Hosmer-Lemeshow goodness-of-fit test result (p = 0.421).

(C) Decision curve analysis demonstrated that the combined model provided a higher net benefit than the stenosis severity model across a wide range of threshold probabilities.

(D–F) Long-term ACS stratification results. Predictive performance was assessed via time-dependent AUC (panel D), NRI, IDI (E), and DCA (F).

(D) Time-dependent AUC analysis demonstrated that the combined model—incorporating the propensity score derived from population health data, stenosis severity, and low-density plaque burden (per 10-unit increase)—had superior overall predictive performance compared to the stenosis severity model.

(E) At a follow-up of 42.1 months, the combined model demonstrated statistically significant improvement over the stenosis severity model, with both the NRI and IDI value > 0.

(F) Decision curve analysis demonstrated that the combined model provided a greater net clinical benefit than the stenosis severity model across a wide range of threshold probabilities at a median follow-up of 42.1 months.

∗The p value for the comparison between groups is < 0.001.

ACS, acute coronary syndrome; AUC, areas under the curve; 95% CI, 95% confidence interval; ROC, receiver operating characteristic; DCA, decision curve analysis; ΔCT-FFR, CT-FFR proximal to the lesion minus CT-FFR distal to the lesion; CT-FFR, CT-derived fractional flow reserve; AUC, area under the curve; NRI, net reclassification index; IDI, integrated discrimination improvement; C index, concordance index.

In the long-term ACS risk stratification group, time-dependent AUC analysis demonstrated that the combined model—incorporating the propensity score derived from population health data, stenosis severity, and LDP% (per 10-unit increase)—had superior overall predictive performance compared to the stenosis severity model (concordance index [C-index]: 0.80, 95% Cl: 0.73–0.87 vs. 0.65, 95% Cl: 0.56–0.74, p = 0.010) (Figure 1D). Compared to the stenosis severity model, the combined model demonstrated a statistically significant net reclassification improvement (NRI = 0.282, 95% CI: 0.019–0.487, p = 0.032) and integrated discrimination improvement (IDI = 0.028, 95% CI: 0.001–0.099, p = 0.040) at a follow-up of 42.1 months (Figure 1E). Figure 1F demonstrates that, based on DCA, the combined model provides a higher net clinical benefit than the stenosis severity model across a wide range of threshold probabilities at a median follow-up of 42.1 months.

External validation of the models

Details of the clinical and imaging features at baseline, as well as between-group comparisons for each external validation cohort, are presented in Tables S4–S6. In terms of validating the short-term ACS risk stratification, the combined model demonstrated a superior discriminative ability in distinguishing between culprit and non-high-risk lesions compared to the model based solely on stenosis severity (AUC = 0.62, 95% Cl: 0.54–0.69 vs. 0.76, 95% Cl: 0.69–0.83, p = 0.001) (Figure 2A). In terms of validating the long-term ACS risk stratification, time-dependent AUC analysis demonstrated that the combined model consistently provided superior discrimination of future ACS risk compared to the stenosis severity-only model in both cohort 3 (C index: 0.76, 95% Cl: 0.66–0.85 vs. 0.60, 95% Cl: 0.50–0.69, p = 0.018) (Figure 2B) and cohort 4 (C index: 0.77, 95% Cl: 0.70–0.84 vs. 0.59, 95% Cl: 0.52–0.66, p < 0.001) (Figure 2C).

Figure 2.

Comparison of model performance for short-term and long-term ACS risk stratification in the external validation cohorts

In cohort 2 (A), model discrimination between culprit and non-high-risk lesions was evaluated using ROC analysis and DeLong’s test for AUC comparison, with IVUS as the reference standard. The validation results demonstrated that the combined model—incorporating the propensity score derived from population health data, stenosis severity, ΔCT-FFR (per SD decrease), and low-density plaque burden (per 10-unit increase)—exhibited superior discriminatory ability in distinguishing between culprit and non-high-risk lesions compared to the model based solely on stenosis severity.

In the prospective cohorts 3 (B) and 4 (C), time-dependent AUC analysis was used to assess the predictive performance for future ACS risk and the results demonstrated that the combined model—which incorporates the propensity score derived from population health data, stenosis severity, and low-density plaque burden (per 10-unit increase)—provided better discrimination of future ACS risk than the model based on stenosis severity alone.

∗: The p value for the comparison between groups is < 0.001.

IVUS, intravascular ultrasound; All other abbreviations are consistent with those in Figure 1.

Discussion

In this study, we demonstrated that a single CCTA examination can simultaneously serve both short-term (≤7 days) and long-term (≥30 days) risk stratification for ACS, with interrelated yet distinct implications. Specifically, we found that ΔCT-FFR (per standard deviation decrease) was significantly associated only with ACS development within 7 days, while the propensity score derived from population health data, stenosis severity, and LDP% (per 10-unit increase) were identified as independent risk factors for short-term ACS occurrence, as well as independent predictors of future ACS events.

The differential predictive value of ΔCT-FFR for short-term risk versus the sustained importance of morphological features (stenosis severity and LDP%) for long-term risk reflects fundamental differences in the biological continuum of coronary plaque progression. Specifically, the short-term ACS risk reflected by ΔCT-FFR may be driven by acute hemodynamic triggers, wherein a significant pressure drop across a lesion indicates the presence of hemodynamic forces that can precipitate fissuring or rupture in vulnerable plaques, such as thin-cap fibroatheromas.¹⁵ In contrast, the long-term risk associated with high LDP% and severe stenosis aligns with a slower, inflammatory-driven morphological evolution. This process involves progressive lipid accumulation, sustained macrophage infiltration, and fibrous cap thinning over months to years, ultimately culminating in plaque rupture upon reaching a critical vulnerability threshold.¹⁶ Our findings thus suggest that ΔCT-FFR serves as a surrogate for immediate hemodynamic stress, whereas quantitative plaque characteristics track the indolent yet high-risk progression of coronary atherosclerosis.

Previous studies have shown that most ACS originate from fibrous cap rupture in thin-cap fibroatheroma (TCFA), which leads to exposure of prothrombotic material and subsequent formation of flow-limiting thrombi.¹⁷ A study by Otsuka K et al. further advocates for increased attention to lesions that are currently non-obstructive but exhibit vulnerable morphological features on CCTA.¹⁸ Results from the ICONIC (Incident Coronary Syndromes Identified by Computed Tomography) case-control study indicated that cross-sectional plaque burden, fibro-fatty and LDP volume, and HRP significantly increased the adjusted hazard ratio for ACS.⁷ Findings from SCOT-HEART (Scottish Computed Tomography of the HEART) revealed that low-attenuated plaque burden was the strongest predictor of both fatal and nonfatal myocardial infarction.¹⁹ Consistent with previous studies,^7,20 qualitative parameters such as PR, low attenuation plaque (LAP), and napkin ring sign (NRS) were statistically significant in group comparisons and univariable regression. However, their effects were overshadowed when quantitative parameters were introduced. This may be due to the reliance of qualitative parameters on clinician experience and visual assessment. Additionally, the prevalence of qualitative parameters can vary significantly among different populations. Contrary to the SCOT-HEART study, our study found a higher threshold for LDP. This discrepancy may be attributed to differences in the definition of LDP. The SCOT-HEART study defined LDPs as those with an attenuation of <30 HU, whereas we defined LDPs as having a CT value between −30 and 75 HU, resulting in a broader range. The establishment of this threshold was based on prior evidence suggesting that a broader attenuation range may better capture high-risk plaque features and demonstrate improved correlation with intravascular ultrasound (IVUS).^21,22

Last but not least, our study found insufficient evidence to show a significant relationship between individual population health data and ACS. However, after performing propensity score matching and linear conversion {log [PS/(1-PS)]}, the comprehensive effect of population health data demonstrated independent predictive value in both multivariable logistic regression and Cox regression. This suggests that multiple clinical risk factors may have a combined predictive effect. The overall level of risk captured by these factors could contribute to the occurrence and progression of ACS.

Limitations of the study

Our study has several limitations. First, the retrospective cohort design presents inherent limitations, such as a high loss to follow-up rate in the long-term ACS risk stratification cohort and potential recall and selection biases. To mitigate these issues, we validated the results in two prospective cohorts. However, the number of patients undergoing IVUS examination was small, leading to potential selection bias. Furthermore, the absence of optical coherence tomography data as a reference standard may limit the detection of superficial high-risk plaque features. Second, due to the time-consuming nature of plaque composition measurements, only one plaque per patient—the culprit lesion and the most severely stenotic non-high-risk lesion—was selected for subsequent analyses, preventing intraindividual comparisons of culprit and non-high-risk lesions. Consequently, this single-lesion selection may underestimate multi-lesion risk dynamics and prevent intraindividual comparisons of culprit and non-high-risk lesions. Third, while medication data were collected and adjustments made, the frequency and dosage of medications used by patients remain unknown. Fourth, despite propensity score matching, a significant residual imbalance in the male sex variable persisted (standardized mean difference [SMD]>0.1) in cohort 4, as male sex is a well-established risk factor for ACS and was therefore more prevalent in the future ACS group. In the subsequent validation of the model, sex was also incorporated into the propensity score derived from population health data. Fifth, a more clinically applicable algorithm is still needed to integrate population health data for estimating the risk of ACS occurrence. Finally, our derivation and validation cohorts were derived from Chinese populations. Although this ensures internal consistency, it may limit the generalizability of our findings to other ethnic groups. Therefore, external validation in multi-ethnic, international cohorts is necessary to further confirm the robustness and applicability of our model before widespread clinical implementation.

Resource availability

Lead contact

Further information and requests for resources should be directed to and will be fulfilled by the lead contact, Long Jiang Zhang (kevinzhlj@163.com).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• •Requests for imaging data used in this work should be directed to the lead contact. The availability of imaging data will be contingent upon the specific request and institutional policies.
• •Any additional information required to reanalyze the data reported in this article is available from the lead contact upon request.

Acknowledgments

This work was supported by the Noncommunicable Chronic Diseases-National Science and Technology Major Project (2024ZD0521700), China; the National Natural Science Foundation of China (No. 82302187 for H.Y.Q); and the National Natural Science Foundation of China (No. 82471979 for Q.C.).

Author contributions

L.J.Z. and P.P.X. were responsible for the concept and design. P.P.X., J.N.Z., M.J.L., Z.H.X., and X.W.T. performed statistical and computational analysis of data. J.N.Z., Q.C., H.Y.Q., F.Z., B.J.G., Y.H., Y.L., L.S.M., and J.H.G. did data collection. P.P.X. and J.N.Z. performed image analysis. F.W., C.S.Z., and X.H. were involved in technical support. P.P.X. wrote the original draft article. H.X. and L.J.Z. reviewed and performed a revision of the article. All authors had full access to all the data and verified the underlying data. All authors read and approved the final article for publication.

Declaration of interests

Zhi Han Xu is a current employee of Siemens Healthineers. Xin Wei Tao is a current employee of Boston Scientific Corporation. All other authors declare no conflicts of interest to disclose.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Deposited data
Derivation cohort	This papter	–
Validation cohort	Previous study	Chen et al., Qiao et al., Zhou et al., Guo et al.23,24,25,26
Software and algorithms
SPSS version 22.0	SPSS	https://www.ibm.com/cn-zh/products/spss
R version 4.2.2	RStudio	https://www.r-project.org

Experimental model and study participant details

Study participants

The study design is shown in the Graphical abstract, and the study cohort composition is illustrated in Figure S1. The study adhered to the principles of the Declaration of Helsinki. Ethical approval was obtained from the independent ethics committees of Jinling Hospital (Approval No.: 2022DZKY-023-01), Nanjing First Hospital (Approval No.: KY20230106-02-KS-01), and General Hospital of the Central Theater Command (Approval No.: [2024]061-01). Written informed consent was waived for the retrospective cohort but was obtained from participants in the prospective cohorts. All study participants were of self-reported East Asian (Chinese) ethnicity. The sex and age distributions for each derivation and validation cohort are provided in the respective results tables and supplemental information.

Derivation cohort (cohort 1)

Derivation cohort consecutively enrolled patients from three participating institutions who underwent CCTA during hospitalization for the indication of chest discomfort. The study population comprised patients with a broad spectrum of clinical indications for CCTA, including those presenting with both stable and acute chest pain. New-onset ACS culprit lesions and precursor lesions were identified within 7 days and at ≥30 days, according to established criteria. This cohort sought to determine the quantitative and qualitative CCTA characteristics associated with these lesions. Information on enrollment and exclusion criteria for derivation cohort can be found in Figure S2.

Validation cohort (cohort 2)

This multicenter retrospective cohort was derived from the development cohort of a previous study²³ and included patients who underwent CCTA in the three months preceding IVUS (supplemental information). This study used IVUS as the reference standard to validate the short-term risk stratification of ACS performance of CCTA features of culprit lesions, identified in the derivation cohort, among chest discomfort patients undergoing IVUS evaluation.

Validation cohort (cohort 3)

This prospective single-center cohort, derived from a previous study (supplemental information),²⁴ included 1,133 patients >18 years old with suspected CAD (stenosis: 20%–80%) for whom CCTA was recommended as a first-line test. In this study, with interim (5-year) follow-up results, we validated the long-term risk stratification of ACS performance of CCTA features in identifying culprit lesion precursors among a suspected CAD population.

Validation cohort (cohort 4)

This is a prospective multicenter nested case-control cohort using data from the prospective multicenter Chinese CT-FFR Study 2 (registration number: ChiCTR2000031410)²⁵ and China CT-FFR Study 3 (registration number: ChiCTR2100053219) (supplemental information).²⁶ In this study, the nested case-control cohort included data from 26 hospitals, aiming to validate the long-term risk stratification of ACS performance of CCTA features of culprit precursors with short-term follow-up. To address the differences in data volume between ACS and CCS cases, and to adjust for variations in clinical risk factors between the groups, we first screened ACS cases. We then matched CCS cases from China CT-FFR Study 2 and China CT-FFR Study 3 in a 1:4 ratio using propensity score matching (PSM) based on age, sex, hypertension, diabetes, hyperlipidemia, and smoking history. This study cohort comprises a suspected CAD population with balanced baseline characteristics between case and control groups through PSM, aiming to further validate the long-term risk stratification of ACS performance of culprit lesion precursor features.

Information on enrollment and exclusion criteria for validation cohorts can be found in Figure S2.

Method details

ACS and culprit lesion/precursors adjudication

ACS including myocardial infarction and unstable angina (supplemental information), was adjudicated through a rigorous multi-modality assessment protocol. Myocardial infarction is defined according to the fourth universal criteria,²⁷ and unstable angina is based on the 2017 cardiovascular and stroke endpoint definitions for clinical trials²⁸ (supplemental information). An independent clinical event committee adjudicated ACS cases. All relevant findings, including ICA, electrocardiographic (ECG), IVUS and wall motion abnormalities on echocardiography were submitted to the core laboratory for vessel-level adjudication.²⁹ The culprit lesion was defined as the atherosclerotic plaque directly associated with the newly diagnosed index ACS event,³⁰ with CCTA examinations restricted to within 7 days of clinical presentation to minimize therapeutic confounding. Patients exhibiting multiple potential culprit plaques within a single artery were categorized as “indeterminate” and excluded from analysis to maintain diagnostic precision. Plaques with the most severe stenosis were designated as non-high-risk lesions for control analysis in patients without ACS within 7 days. Plaques associated with ACS occurring ≥30 days were defined as precursor lesions. In the CCS group, plaques causing the most severe luminal stenosis but not leading to future ACS were identified as stable plaques for analysis. The final follow-up date in the derivation cohort was May 31, 2024.

CCTA analysis

All CCTA images were acquired using scanners with ≥64 detector rows. Details of the CCTA models and scanning protocols can be found in the supplemental information. Data were imported into a post-processing workstation (Syngo.Via VB10, Siemens) for analysis. Image interpretation was equally allocated to four fellowship-trained radiologists with minimum six-year experience in CCTA. Baseline plaque anatomy parameters included plaque location, nature (calcified, non-calcified, or mixed), stenosis severity, length, involved segment, and location. Based on the thresholds of 70%, the degree of diameter stenosis was analyzed in two groups. APCs such as PR, spotty calcification (SC), LAP, and NRS were visually assessed.³¹ HRP was identified by the presence of two or more APCs,²⁵ with detailed definitions in the supplemental information. Plaque composition was measured using semiautomated coronary plaque analysis software (Syngo.Via Frontier, version 4.2.1, Siemens Healthcare), which has shown effective performance in previous studies.^23,32 Radiologists could manually adjust plaque delineation. The software analyzed plaque composition at 1 mm intervals using predefined Hounsfield unit (HU) thresholds: LDP (−30 to 75 HU), fibro-fatty (76–130 HU), fibrous (131–350 HU), and dense calcium (351–2048 HU).³³ The proportion of each component was calculated by dividing its volume by the vessel volume and multiplying by 100.¹⁵ A detailed technical note on the software’s processing time, reproducibility, required expertise, and scalability is provided in the supplemental information.

Fully automated CT-FFR calculations

CT-FFR calculations were performed using dedicated software (skCT-FFR version v0.7.1, Beijing, China). The technical principles and diagnostic accuracy of the fully automated CT-FFR algorithm (skCT-FFR) have been comprehensively described and validated in previous study.³⁴ Detailed information can be found in the supplemental information. A CT-FFR value of ≤0.80 indicated hemodynamic significance of coronary artery stenosis. Lesion-specific ischemia was present when CT-FFR ≤0.80 was measured 2 cm distal to the lesion.⁶ ΔCT-FFR was calculated by subtracting the CT-FFR distal to the lesion from the CT-FFR proximal to the lesion.⁶

Quantification and statistical analysis

Statistical analyses were performed using SPSS version 22.0 (IBM SPSS Inc.) and R version 4.2.2 (R Foundation). Descriptive statistics, between-group comparisons, and intra-/inter-observer agreement details are provided in the supplemental information. The specific test statistics (e.g., chi-square value, U-value) of derivation cohort for each comparison are provided in the Table S8. Variables with P-values <0.05 in comparisons between culprit/non-high-risk lesions and their precursors were included in multivariate analyses. Logistic regression calculated odds ratios (ORs) and 95% confidence intervals (95% CI) for short-term ACS risk stratification group. Multivariable competing risk Cox regression (some patients in the control group died during follow-up) estimated hazard ratios (HRs) and 95% CI for long-term ACS risk stratification group. Given the sample size limitations in this study, population health data were used to simulate a propensity score in multivariable regression models 2–3, based on the results of previous literature. The propensity score was then converted into log odds using the formula: log [PS/(1 - PS)]. These log odds were subsequently included as a covariate in the regression analysis for model adjustment³⁵. Receiver operating characteristic (ROC) analysis and time-dependent areas under the curves (AUCs) were employed to assess the predictive performance of CCTA features, while DeLong’s test was used to compare differences in AUC values. Model performance was validated using multiple approaches: the calibration curve was applied to assess the agreement between predicted and observed probabilities; the Hosmer-Lemeshow test was used to evaluate goodness-of-fit; decision curve analysis (DCA) quantified clinical utility across different risk thresholds; while the net reclassification improvement (NRI) and integrated discrimination improvement (IDI) were employed to measure improvements in risk stratification and predictive discrimination. A P-value of <0.05 was considered statistically significant.

Additional resources

Cohort 4 in this study is a prospective multicenter nested case-control cohort derived from the following registered clinical trials: Chinese CT-FFR Study 2: ClinicalTrials.gov ID ChiCTR2000031410 and China CT-FFR Study 3: ClinicalTrials.gov ID ChiCTR2100053219 (https://www.chictr.org.cn/showproj.html?proj=135328).

Published: November 3, 2025