Incremental prognostic value of coronary artery calcium progression within a large community-benefit calcium score registry

Abstract

Background

Coronary artery calcium (CAC) scoring is a robust cardiovascular risk marker, yet the prognostic value of serial assessment and the relative importance of incident calcium versus progression from elevated baseline CAC remains incompletely defined.

Methods

We analyzed 4166 CLARIFY registry participants (NCT04075162) who underwent ≥2 CAC scans ≥11 months apart (mean interval 4.5 years). Defining CAC progression (ΔCAC>10 Agatston units, AU) identified 1982 progressors and 2184 non-progressors. Annualized absolute, square-root, and log-transformed progression were evaluated for association with major adverse cardiovascular events (MACE: myocardial infarction, stroke, mortality, revascularization, and heart failure). Incremental prognostic value was assessed by adding annualized progression to the base MESA-CAC model incorporating baseline CAC, demographics, and traditional risk factors.

Results

Over a median 5.9 year follow-up, progressors had higher MACE incidence, though risk gradients differed by baseline CAC. Incident CAC from a baseline zero score was not associated with increased MACE (all p > 0.60). However, progression of established CAC was associated with nearly double 5-point MACE (HR 1.96 [1.38–2.80]; p < 0.001) which persisted across baseline CAC strata of 1–100, 101–200, and >200 AU (HR 1.74–2.65; all p < 0.01). Annualized progression independently predicted events, most strongly using log-transformed metrics (overall cohort HR 1.27 [1.15–1.41]; p < 0.0001). Associations persisted across smokers, diabetics and statin-users. Adding annualized progression to the base MESA-CAC model improved net risk reclassification (NRI ∼0.10).

Conclusions

Serial CAC imaging may provide incremental prognostic value in individuals with pre-existing atherosclerosis. While incident CAC after a baseline zero score may confer limited short-term risk, it may present a window for preventive intervention.

1. Introduction

Non-contrast computed tomography (CT) coronary artery calcium (CAC) scoring has become a guideline-directed cornerstone of adjunctive patient-centered decision-making in asymptomatic individuals. There is a clear established gradient of risk between Agatston CAC score and major adverse cardiovascular events (MACE) [[1], [2], [3], [4], [5], [6], [7]]. Whereas a CAC score of zero confers substantial negative predictive value (i.e., the “Power of Zero”), atherosclerosis is dynamic, with progression often occurring in many patients due to poor risk-factor control [[8], [9], [10]]. CAC progression over time and the clinical utility of serial CAC screening remain topics of intense interest. As an indicator of risk, CAC progression may help guide longitudinal preventive strategies as well as provide a predictive marker for cognitive outcomes [11].

Current international guidelines offer varying recommendations for serial scanning. The American College of Cardiology/American Heart Association (ACC/AHA) and European Society of Cardiology (ESC) both suggest a “warranty period” [12] for those with a baseline score of zero, recommending a repeat scan in 5–10 years, particularly if statin therapy was not initiated [13,14]. The Canadian Cardiovascular Society (CCS) suggests tighter 5-year rescanning for patients aged 40–75 if statins continue to be withheld [15]. Despite these established intervals for patients with an initial CAC of zero (CAC₁=0), global guidelines lack a consensus for monitoring the velocity of progression in patients with established plaque (CAC₁>0) [6].

Prior studies in large cohorts, such as the Multi-Ethnic Study of Atherosclerosis (MESA) and the Heinz Nixdorf studies, have demonstrated that CAC progression provides independent, incremental prognostic value over underlying traditional risk factors [[16], [17], [18], [19], [20]]. Despite this and the varying international guidelines noted above, serial scanning has not been widely adopted in clinical practice, in part due to a lack of standardized screening protocols, an ill-defined optimal time interval between scans, uncertainty regarding how progression should alter therapeutic decision-making, and the economic implications of repeated testing [21]. Importantly, the risk associated with those who progress from a baseline of zero (CAC₁=0, CAC₂>0) may differ from those who progress from an elevated baseline (CAC₁>0, CAC₂>CAC₁), particularly given the negative predictive value of a zero CAC score and its enduring legacy of protection that may extend to >10 years in some patients [3,12,22,23]. One study focusing on all-cause mortality suggested that only those with established baseline CAC derive significant prognostic information from serial scanning [16], whereas an analysis of the MESA cohort suggested that CAC₁=0 progressors may have a modest elevation in MACE risk over a long follow-up period of 7.6 years [17].

Understanding the risk gradients associated with calcium progression and the velocity of progression could help inform preventive strategies and establish standardized guidelines for serial CAC screening. Accordingly, the purpose of this study was to assess: (1) the incremental prognostic value of serial CAC scanning, (2) if the prognostic significance of progression differs between those with and without baseline plaque, and (3) whether the excess risk associated with progression persists across traditional high-risk subgroups. Through these investigations, we aimed to clarify which patient populations would derive the greatest clinical benefit from serial coronary calcium assessment.

2. Methods

2.1. Study population and selection

We identified 4166 individuals selected from the Community Benefit of No-charge Calcium Score Screening (CLARIFY) Registry (Clinicaltrials.gov, NCT04075162) at University Hospitals Cleveland Medical Center (UH) [24,25] who underwent at least two CT CAC scans, between 2014 and 2023 (Central Illustration). For comparison of baseline characteristics, we also identified 105,608 patients within CLARIFY who underwent only a single CT CAC scan. Given the retrospective nature of the study and use of de-identified patient data, the UH Institutional Review Board (IRB #STUDY20190995) waived informed consent. CAC scans (both primary and subsequent) were ordered by treating healthcare providers based on clinical observations; there was no institutional policy or protocol in place regarding the exact indications for ordering the first and especially any subsequent CAC scans. All CAC scores, including those from repeat scanning, were reported to referring clinicians per standard clinical care. In 2018, based on guideline recommendations, repeat scanning was allowed in ∼5 years for individuals with CAC₁=0 when statin therapy was deferred. Shorter or individualized intervals for higher risk phenotypes were recommended without firm guidance and generally required consultation with a cardiologist [1]. To maximize the available cohort size while maintaining biological relevance, a minimum inter-scan interval of 11 months was required for inclusion, similar to prior studies [16]. In those with more than two scans, the first and second scans only were considered.

Central Illustration: Study design and CAC progression clinical trajectories. (Top) Overview of (left) retrospective enrollment and study population and (right) data extraction, MACE event definition, CAC progression and non-progression definitions. (Bottom) Example axial CT CAC images of the proximal left anterior descending artery (LAD) for patients within three distinct CAC trajectories: (left) Non-Progression, (middle) Zero Baseline Progression, and (right) Non-Zero Baseline Progression. The example for Baseline Non-Zero Progression (CAC 1757→5131) is an extreme example for visualization purposes and not representative of the entire subgroup. Each scan acquisition year and CAC₁ or CAC₂ values are reported at top left (yellow text). The aortic root (Ao) is also labeled for anatomic orientation. Observed risk for each non-progression or progression trajectory are provided as a range of MACE rates (events per 1000 person-years, PY) across definitions. Subsequent clinical decision-making is proposed.

2.2. Data extraction and clinical endpoints

Clinical data including Agatston scores, scan dates, medication use, demographics and laboratory values were extracted from the electronic health record (Central Illustration). Reported demographics, labs, and medications (Table 1) reflect data at the time of the baseline CAC scan. Participants were followed for incident MACE, defined using nested composites 3-point (3P) MACE (myocardial infarction [MI], stroke or all-cause mortality), 4-point (4P) MACE (3P plus coronary revascularization), and 5-point (5P) MACE (4P plus heart failure).

Table 1.

Baseline characteristics and cardiovascular risk profile of the study cohort. Mild, moderate, and severe progression are defined based on absolute change in CAC score in Agatston units (AU) between scans as 10–50 AU, 50–150 AU, and >150 AU, respectively. Medication and laboratory data are reported at time of the first CAC scan.

Characteristic	Overall (n = 4166)	Non-Progressors (n = 2184)	Mild Progression (n = 1197)	Moderate Progression (n = 369)	Severe Progression (n = 416)	p-value
Age at CAC1, years	58.42 ± 8.37	56.60 ± 8.36	59.92 ± 8.04	60.67 ± 7.80	61.70 ± 7.52	<0.001
Sex						<0.001
Female	2079 (50%)	1305 (60%)	540 (45%)	126 (34%)	108 (26%)
Male	2087 (50%)	879 (40%)	657 (55%)	243 (66%)	308 (74%)
Race						0.30
Caucasian	3743 (90%)	1937 (89%)	1093 (91%)	334 (91%)	379 (91%)
Black	241 (5.8%)	143 (6.5%)	57 (4.8%)	19 (5.1%)	22 (5.3%)
Other	80 (1.9%)	50 (2.3%)	20 (1.7%)	4 (1.1%)	6 (1.4%)
Declined	102 (2.4%)	54 (2.5%)	27 (2.3%)	12 (3.3%)	9 (2.2%)
Ethnicity						—
Hispanic/Latino	46 (1.1%)	24 (1.1%)	15 (1.3%)	2 (0.5%)	5 (1.2%)
Not Hispanic	3885 (93%)	2048 (94%)	1109 (93%)	342 (93%)	386 (93%)
Declined	235 (5.6%)	112 (5.1%)	73 (6.1%)	25 (6.8%)	25 (6.0%)
BMI, kg/m²	29.26 ± 5.64	28.83 ± 5.67	29.56 ± 5.54	29.81 ± 5.59	30.20 ± 5.60	<0.001
Systolic BP, mmHg	126.84 ± 14.23	125.58 ± 14.15	127.28 ± 13.93	129.73 ± 14.59	129.62 ± 14.40	<0.001
eGFR, mL/min/1.73m²	87.19 ± 16.62	87.74 ± 16.63	86.92 ± 16.13	86.19 ± 16.65	85.92 ± 17.80	0.20
HDL, mg/dL	57.62 ± 21.04	58.87 ± 21.29	56.93 ± 20.23	54.61 ± 21.27	55.71 ± 21.32	<0.001
Total Cholesterol, mg/dL	211.58 ± 42.63	214.78 ± 41.45	211.21 ± 43.30	206.31 ± 41.25	200.50 ± 45.72	<0.001
LDL, mg/dL	126.88 ± 38.82	129.14 ± 38.22	127.04 ± 39.07	124.25 ± 37.37	116.87 ± 40.89	<0.001
Triglycerides, mg/dL	135.58 ± 89.51	133.22 ± 84.93	136.85 ± 99.35	137.95 ± 85.50	142.21 ± 86.38	0.034
HbA1c, %	5.77 ± 0.80	5.73 ± 0.76	5.78 ± 0.77	5.85 ± 0.92	5.93 ± 0.94	<0.001
Smoking	1036 (25%)	455 (21%)	308 (26%)	134 (36%)	139 (33%)
Diabetes	469 (11%)	187 (9%)	138 (12%)	57 (15%)	87 (21%)
Hypertension	2036 (49%)	938 (43%)	613 (51%)	216 (59%)	269 (65%)
Medication Use
Statins	2361 (57%)	939 (43%)	779 (65%)	292 (79%)	351 (84%)
PCSK9i	68 (2%)	26 (1%)	23 (2%)	6 (2%)	13 (3%)
Beta-Blockers	958 (23%)	446 (20%)	284 (24%)	95 (26%)	133 (32%)
CCB	853 (20%)	393 (18%)	254 (21%)	85 (23%)	121 (29%)
ACEi	949 (23%)	420 (19%)	291 (24%)	105 (28%)	133 (32%)
ARB	826 (20%)	364 (17%)	258 (22%)	84 (23%)	120 (29%)
ARNi	154 (4%)	65 (3%)	51 (4%)	14 (4%)	24 (6%)
Diuretics	1029 (25%)	472 (22%)	310 (26%)	112 (30%)	135 (32%)
MRA	155 (4%)	88 (4%)	36 (3%)	13 (4%)	18 (4%)

ACEi=angiotensin-converting enzyme inhibitors; ARB=angiotensin II receptor blocker; ARNi=angiotensin receptor-neprilysin inhibitor; CAC₁=first CAC score; CAC₂=second CAC score; CCB=calcium channel blockers; MACE=major adverse cardiovascular events; MRA=mineralocorticoid receptor antagonists; PCSK9i=proprotein convertase subtilisin/kexin type 9 inhibitor.

2.3. Defining CAC progression and non-progression

To account for inherent variability in CT image acquisition and processing, an absolute difference (∆CAC CAC₂—CAC₁) of ≤10 Agatston units (AU) between the two scans (regardless of time interval) was considered non-progression. Those patients identified as having ∆CAC<0 AU were manually reviewed and considered non-progression as each was found to be a spurious result due to a measurement or segmentation error. Patients were then categorized into three progression groups based on their absolute ∆CAC between scans: mild (10–50 AU), moderate (51–150 AU), and severe (>150 AU) for visualization of event incidence. To further characterize the velocity of disease, annualized progression rates (annualized absolute ∆CAC, Eq. (1) below) were calculated, and categorized into intervals of <0, 0–9, 10–99, 100–199, and ≥200 AU/year. Note that values < 0 are considered as non-progression in our analyses as explained above. Finally, for participants starting with a baseline CAC of 0, the incident CAC rate, expressed as events per 100 person-years, was calculated by dividing the number of incident CAC cases by the sum of all inter-scan intervals. Overall, three distinct CAC trajectories were defined: Non-progression (∆CAC≤10), Zero Baseline CAC Progression (CAC₁=0, ∆CAC>10), and Non-Zero Baseline CAC Progression (CAC₁>0, ∆CAC>10). Representative CT CAC images are provided (Central Illustration). Subgroups stratified by absolute baseline CAC of 1–100, 101–200, and >200 AU were also interrogated to compare additional subtypes of progressors.

2.4. Mathematical transformations to reduce skewness in CAC data

To assess rate of change normalized by time between scans (∆t) for use in survival models, we calculated the following annualized metrics. For log transformation, a constant of 25 was added to all scores to permit the inclusion of participants with a baseline CAC of zero (i.e., the “MESA” method), as used previously [16,18].

Eq. (1): Annualized absolute difference:

$$\frac{CA{C}_2-CA{C}_1}{\Delta t}$$ (1)

Eq. (2): Annualized square-root transformation:

$$\frac{\sqrt{CA{C}_2}-\sqrt{CA{C}_1}}{\Delta t}$$ (2)

Eq. (3): Annualized log-transformation:

$$\frac{ln\left(CA{C}_2+25\right)-ln\left(CA{C}_1+25\right)}{\Delta t}$$ (3)

2.5. Statistical analysis

Cox proportional hazard models were used to estimate hazard ratios (HR) for 3P, 4P, and 5P MACE. Initial models were adjusted for age, sex, race, and time between scans. To assess the independent prognostic value of progression, a base MESA-CAC model was constructed including the first CAC score (i.e., excluding the second CAC score), age, sex, race, and traditional risk factors including smoking, diabetes, and dyslipidemia [17]. This was compared against extended models incorporating one of the three progression metrics above. The incremental performance of these metrics for 5-year MACE was quantified using integrated discrimination improvement (IDI) and net reclassification improvement (NRI) with perturbation resampling. For categorical NRI analyses, predicted 5-year risk categories were defined as <5%, 5–10%, 10–20%, and ≥20%. Finally, subgroup analyses for traditional risk factors of smoking and diabetes as well as statin-use were also performed. All data analysis was performed using R.

Data availability

The clinical datasets analyzed for this study are not publicly available due to stipulations of the UH IRB protocol limiting de-identified patient data sharing, though they may be made available through institutional review on reasonable request.

3. Results

3.1. CAC progressors are older and have higher statin utilization

A total of 4166 participants (mean age 58 years, 50% male, 90% Caucasian, and 5.8% Black) from the CLARIFY registry were included in the analysis (Table 1). For the whole cohort, the median follow-up and mean time between scans was 5.9 years and 4.5 years, respectively. The median follow-up for those who progressed from Zero Baseline CAC and from Non-Zero Baseline CAC was 6.36 years and 5.9 years, respectively. Participants were further categorized by absolute annual CAC change into four groups (Table 1): non-progressor (n = 2184; 52%), mild (n = 1197; 29%), moderate (n = 369; 9%), and severe (n = 416; 10%). As the severity of CAC progression increased, participants were significantly more likely to be older, male, and have a higher prevalence of hypertension and diabetes (all p < 0.001). Low-density lipoprotein cholesterol (LDL-C) levels, which represent a composite of both statin treated and untreated participants, were lower in the severe progression group compared to non-progressors (116.9 ± 40.9 vs. 129.1 ± 38.2 mg/dL; p < 0.001). Concurrently, the relative use of lipid-lowering statins reported at baseline CAC scanning was higher in severe progressors compared to non-progressors (84% vs. 43%). Comparing baseline characteristics (Supplemental Table 1) of the serial scanning cohort (n = 4166) to the much larger general screening cohort of CLARIFY who possess only a single CAC score (n = 105,608), a higher prevalence of diabetes (9–21% vs. 5–15%) and more frequent statin utilization (43–84% vs. 26–58%) is noted.

The distribution of annualized CAC progression (Fig. 1A) showed the highest frequency of participants residing in the 0–10 AU/year range, which then tapers off with increasing annualized change. Average annualized CAC change exhibited a steady increase with age, reaching a peak at ∼75 years (Fig. 1B). Beyond this age, the progression rate declined with an increase in data variability, likely reflecting a smaller sample size of patients over 75 years who had two CAC scans.

Fig. 1.

Annual CAC progression characteristics by age, frequency, and baseline CAC status. (A) Histogram illustrating absolute annual CAC progression distribution, (B) average annual change in CAC by age at time of first CAC scan, and (C) yearly incidence rates (% per 100 person-years) of new CAC for patients starting with Zero Baseline CAC, (D) box plots showing annual CAC progression for categories of (left) all progressors regardless of baseline CAC, (middle) Zero Baseline Progression and (right) Non-Zero Baseline Progression. AU/year=Agatston units per year.

3.2. CAC incidence and progression patterns

For those with established plaque at baseline, annualized progression varied by demographic subgroup (Supplemental Table 2). Distinct sex-based differences were observed in both the frequency and magnitude of progression. Men demonstrated a higher median velocity than women (17.3 vs. 9.8 AU/year). Black men exhibited the highest median annual progression of any subgroup (24.2 AU/year, IQR 5.1–51.4). Hispanic males had a lower median annual change of 15.0 AU/year (IQR 2.6–43.3) compared to 17.1 AU/year in Non-Hispanic males. In terms of categorical progression, the majority of participants (52.4%) experienced an annual change between 10 and 99 AU/year. “High velocity” progression (≥100 AU/year) was observed in 5.5% of the cohort.

Among participants in the Zero Baseline CAC cohort (Supplementary Table 3, n = 1927), 517 (26.8%) developed incident CAC during 8932 person-years of follow-up; the overall annual incidence rate was 5.79 per 100 person-years (95% CI 5.31–6.31). Yearly CAC incidence rates (Fig. 1C) increased with age, ranging from 4.57 in participants aged <50 to 10.20 per 100 person-years in those aged 70–80 (p for trend <0.001). The distribution of annual CAC incidence among Zero Baseline vs Non-Zero Baseline Progressors is presented (Fig. 1D).

3.3. MACE risk is most elevated in those with progression of existing CAC

Cumulative incidence of 3P, 4P, and 5P MACE was significantly higher among progressors compared to non-progressors (p < 0.01 for all; Fig. 2, top). When progressors were stratified by severity of absolute progression between scans (mild: 1–50 AU; moderate: 51–150 AU; severe: >150 AU), an incremental increase in the cumulative incidence for events was observed across all MACE definitions (Fig. 2, bottom). Notably, event curves diverged around 5–6 years across MACE definitions for progressors vs. non-progressors.

Fig. 2.

Cumulative incidence of MACE among progressors vs. non-progressors. (Top) Cumulative incidence of 3P, 4P and 5P MACE of all progressors (red) vs non-progressors (blue). Pair-wise comparison p-values are reported. (Bottom) Cumulative incidence of MACE for non-progressors (blue) vs progressors categorized as mild (green, 10–50 AU), moderate (orange, 50–150 AU), and severe (red, >150 AU) absolute progression between scans. Number at risk for each group by year is provided.

After adjusting for age, sex, and time between scans, the risk of an event was significantly different depending upon the baseline CAC score (Table 2). Participants with Zero Baseline CAC Progression (CAC₁=0, ∆CAC>10) did not show a statistically significant increase in risk for 3P MACE (0.80 [0.39–1.67]; p = 0.60). The risk for 4P MACE (0.94 [0.48–1.84]; p = 0.80) and 5P MACE (0.99 [0.54–1.81]; p > 0.90) were also no different. Similarly, those who had a baseline score that remained stable (Non-Zero Baseline Non-Progression; CAC₁>0 and ∆CAC≤10) showed no significant increase in risk across any MACE definition (Table 2). The lack of association between MACE and Zero Baseline Progression was also observed when interrogating annualized progression metrics (Table 3). Notably, the time between scans was a significant covariate across all models (HR ∼0.73; p < 0.001).

Table 2.

MACE prediction for non-progression and progression cohorts adjusted for age, sex, and time between scans.

Variablea	3P MACE HR (95% CI), Rate/1000 PYb	p	4P MACE HR (95% CI), Rate/1000 PY	p	5P MACE HR (95% CI), Rate/1000 PY	p
Categoryc (n)d
Non-Zero Baseline & Progression (2018)e	1.67 (1.11–2.52), 7.49	0.014	1.83 (1.23–2.72), 8.61	0.003	1.96 (1.38–2.80), 11.22	<0.001
CAC1 1–100 (1208)	1.47 (0.94–2.30), 6.12	0.093	1.54 (1.00–2.38), 6.71	0.051	1.74 (1.18–2.55), 9.09	0.005
CAC1 101–200 (332)	1.86 (1.01–3.40), 8.31	0.045	2.46 (1.43–4.24), 11.46	0.001	2.32 (1.41–3.83), 13.04	<0.001
CAC1 >200 (478)	2.28 (1.35–3.86), 10.81	0.002	2.39 (1.44–3.98), 11.96	<0.001	2.65 (1.69–4.15), 16.16	<0.001
Non-Zero Baseline & Non-Progression (18)f	2.21 (0.30–16.2), 11.61	0.40	2.19 (0.30–16.0), 11.61	0.40	1.75 (0.24–12.7), 11.61	0.60
Zero Baseline & Progression (517)g	0.80 (0.39–1.67), 2.75	0.60	0.94 (0.48–1.84), 3.36	0.80	0.99 (0.54–1.81), 4.28	>0.90
Covariates
Age at CAC1	1.02 (1.00–1.05)	0.034	1.03 (1.01–1.05)	0.011	1.03 (1.01–1.05)	0.003
Sex
Maleh	0.86 (0.60–1.23)	0.40	0.89 (0.63–1.25)	0.50	0.85 (0.63–1.15)	0.30
Time Between Scans	0.73 (0.66–0.81)	<0.001	0.74 (0.67–0.81)	<0.001	0.72 (0.67–0.79)	<0.001

^aSignificant values (p < 0.05) are denoted in boldface font.

^bPerson-years (PY) are calculated as the sum of follow-up time to first event or censoring.

^cZero Baseline & Non-Progression cohort (CAC₁ CAC₂=0) was used as a reference.

^dn=number of patients in cohort.

^eCAC₁ > 0; ∆CAC > 10; includes all Non-Zero Baseline Progressors. Baseline CAC (i.e., CAC₁) subcategories are presented in the rows below.

^fCAC₁ > 0; ∆CAC ≤ 10.

^gCAC₁ = 0; ∆CAC > 10.

^hFemale sex was used as a reference.

Table 3.

MACE risk for patients with Zero Baseline CAC Progression (CAC₁=0, ∆CAC>10).

Outcome	Progression Metric	HR (95% CI)	p-value
3P MACE	Annualized absolute difference	0.47 (0.10–2.21)	0.3369
	Annualized square-root transformed difference	0.88 (0.43–1.80)	0.7308
	Annualized ln(CAC+25) difference	0.76 (0.36–1.62)	0.4807
4P MACE	Annualized absolute difference	0.38 (0.06–2.39)	0.2993
	Annualized square-root transformed difference	0.80 (0.39–1.65)	0.5535
	Annualized ln(CAC+25) difference	0.68 (0.31–1.49)	0.3321
5P MACE	Annualized absolute difference	0.96 (0.51–1.81)	0.9033
	Annualized square-root transformed difference	1.14 (0.72–1.80)	0.5905
	Annualized ln(CAC+25) difference	1.04 (0.64–1.70)	0.8630

In contrast, participants with established coronary calcification (Non-Zero Baseline Progression) exhibited significantly higher MACE event rates compared to those with CAC₁=0 who did not progress (Table 2). The strongest association was observed for 5P MACE (1.96 [1.38–2.80]; p < 0.001), followed by 4P MACE (1.83 [1.23–2.72]; p = 0.003) and 3P MACE (1.67 [1.11–2.52]; p = 0.014). Notably, the average inter-scan interval for Non-Zero Baseline Progressors was shorter than those with CAC₁=0 (4.4 years vs 5.2 years). Adjustment for baseline CAC in Non-Zero Baseline Progressors retained predictive power for MACE (HR 1.66–1.77 for 4P and 5P; p < 0.05 and HR 1.51; p = 0.054 for 3P MACE). Stratifying progressors into clinically relevant groups based on absolute baseline CAC burden of 1–100, 101–200 and >200 AU (Table 2) demonstrated a robust, step-wise increase in the hazard for 5P MACE (HR 1.74–2.65, all p < 0.05). After adjustment for diabetes and smoking (Supplemental Table 4), the magnitude of risk was attenuated though the overall gradient of risk persisted. Notably, progression from a baseline CAC of 100–200 AU or >200 AU remained predictive of 4P MACE whereas the 5P MACE composite remained robust across baseline CAC strata (1.59–2.21, all p < 0.05).

3.4. Incremental prognostic value of annualized progression metrics

In models adjusted for baseline CAC and traditional risk factors (age, sex, race, smoking, diabetes, and dyslipidemia), annualized absolute, square-root, and log-transformed metrics were consistently associated with increased MACE (Fig. 3A). For the whole patient cohort including progressors and non-progressors, the annualized log-transformed (ln[CAC+25]) metric had the highest predictive power across MACE definitions, though was most strongly associated with the 5P composite (HR 1.27 [1.15–1.41]; p < 0.0001).

Fig. 3.

Association between annualized CAC progression metrics and MACE. Forest plots illustrating models predicting 3P (red), 4P (blue), and 5P (green) MACE using annualized log-transformed (ln[CAC+25]), annualized square-root transformed, and annualized absolute CAC difference for (A) the total patient cohort including non-progressors and all progressors and (B) progressors with CAC₁>0 (Non-Zero Baseline) only. All models are adjusted for baseline CAC, age, sex, race, and traditional risk factors including diabetes, smoking, and dyslipidemia.

Similar findings were observed in the Non-Zero Baseline Progression cohort (Fig. 3B), where annualized log-transformed CAC remained a significant predictor of 5P MACE (HR 1.30 [1.16–1.45]; p < 0.0001). In this group, the annualized absolute difference was significant for 4P and 5P MACE (HR 1.13–1.15; all p < 0.05) but did not reach significance for 3P MACE (HR 1.11; p = 0.1053).

All metrics of annualized CAC progression above provided incremental improvement over the base MESA-CAC model (comprised of baseline CAC, age, sex, race, diabetes, dyslipidemia, and smoking status) for predicting 5-year MACE. Annualized log-transformed CAC demonstrated the most robust improvement in discrimination (IDI: 0.010), followed by square-root transformation (IDI 0.006) and absolute change (IDI 0.002). Net reclassification improvement (NRI) was similar across metrics, ranging from 0.097 to 0.106.

To further contextualize the clinical impact of these findings, we evaluated the movement of participants across established 5-year ACC/AHA cardiovascular risk thresholds (<5%, 5–10%, 10–20%, and ≥20%) using a risk reclassification matrix (Table 4). The addition of the annualized log-transformed progression to the base MESA-CAC model reclassified 247 participants (5.9% of the total cohort) into different risk categories. When stratified by event status, 10.3% of participants who experienced a MACE event were reclassified into a higher risk category compared with 7.7% who moved downward, resulting in a net +2.6 percentage points shift among event patients. Among non-event patients, 2.2% and 3.1% reclassified downward and upward, respectively, or a + 0.9 percentage point net movement.

Table 4.

Reclassification matrix for predicted 5-year MACE when adding annualized log-transformed CAC progression to the base MESA-CAC model.

	Base MESA-CAC Model + Annualized Log-Transformed CAC Progressionb
Base MESA-CAC Modela	<5%	5–10%	10–20%	≥20%	Row Total
<5%	3601 (96.9%)c	110 (3.0%)	4 (0.1%)	2 (0.1%)	3717
5–10%	77 (22.1%)	249 (71.6%)	21 (6.0%)	1 (0.3%)	348
10–20%	0 (0.0%)	17 (23.9%)	49 (69.0%)	5 (7.0%)	71
≥20%	0 (0.0%)	0 (0.0%)	10 (33.3%)	20 (66.7%)	30
Column Total	3678	376	84	28	4166

^aBase MESA-CAC Model = baseline CAC, age, sex, race, diabetes, dyslipidemia, and smoking status. Rows report risk categories assigned by the MESA-CAC model.

^bThe combined model integrates the MESA-CAC model with annualized log-transformed CAC progression (Eq. (3) of the Methods). Columns report risk categories assigned by the combined model.

^cValues presented as n (row%); Diagonal = no reclassification. Upper triangle = upward reclassification (to higher risk); lower triangle = downward reclassification (to lower risk).

3.5. Prognostic value of annualized CAC progression persists across traditional risk factor subgroups

Stratified analyses by traditional risk factors of diabetes and smoking as well as baseline statin use were performed (Supplemental Table 5). Annualized CAC progression remained a significant predictor of MACE. The most pronounced predictive value was observed among smokers, where annualized log-transformed progression was associated with the highest hazard for 5P MACE (1.43 [1.21–1.69]; p < 0.0001). Similarly, in patients with diabetes, log-transformed progression was a strong and significant predictor of events (HR 1.37 [1.13–1.65]; p = 0.0012). In a subgroup analysis of progressors using statins, absolute and square root transformed annualized CAC progression metrics remained significantly associated with 3-, 4- and 5P MACE (HR 1.2–1.3; all p < 0.05; Supplemental Table 5).

4. Discussion

The utility of serial CAC screening has been a subject of clinical interest and debate for over a decade [21]. Our findings demonstrate that while baseline CAC should remain the primary attribute for risk assessment, the rate of CAC progression may provide additional prognostic insight, particularly when baseline CAC is already established [16]. For patients with Zero Baseline CAC, incident CAC on the second scan did not translate into a significant increase in MACE risk within our study window (median follow-up of 6.3 years for this cohort). Interpreting these results alongside historical studies [16–18] supports the “Power of Zero” and its warranty period, suggesting that subsequent progression is of modest if any short-term risk, and likely requires longer follow-up to observe clinical events. In contrast, for those with Non-Zero Baseline CAC, any further progression was associated with an increase in MACE risk. For the whole cohort, the clinical utility of serial scanning was further supported by a net improvement in reclassification (NRI ∼0.10) when testing annualized progression metrics over the base MESA-CAC model consisting of baseline CAC, demographics and traditional risk factors alone. Further, mapping these changes across established AHA/ACC risk thresholds demonstrated that adding log-transformed progression to the MESA-CAC model yielded a net upward reclassification of +2.6 percentage points among patients who experienced MACE. CAC progression continued to robustly predict MACE risk in Non-Zero Baseline Progressors albeit the association was attenuated after correcting for traditional risk factors. The association between annualized progression and MACE was particularly pronounced among smokers and patients with diabetes, where progression showed the strongest relationship with subsequent events. Among statin users who demonstrated CAC progression, multiple measures of annualized CAC progression continued to be significantly associated with MACE.

Current international guidelines for serial CAC scanning focus on those with a Zero Baseline CAC and often suggest a 5–10 year warranty period before re-scanning to re-evaluate statin initiation. However, there are currently no specific guidelines for assessing those with a Non-Zero Baseline CAC [6]. In this study, our results suggest serial scanning in those with a Non-Zero Baseline CAC may provide additional prognostic insight. Specifically, in subgroups exhibiting rapid accumulation, such as the median 24.2 AU/year progression observed in Black men, serial scanning could serve as a tool for transitioning from moderate to high-intensity statin therapy or justifying aggressive addition of non-statin lipid-lowering agents such as ezetimibe or Proprotein Convertase Subtilisin/Kexin type 9 (PCSK9) inhibitors. Given the limitations of our retrospective analyses, this is hypothesis-generating and would need careful investigation in prospective clinical trials.

In our cohort, the severe absolute progression group had the lowest mean LDL-C and highest statin utilization (reported at the time of the baseline CAC scan) yet also the highest incidence of MACE. An increase in CAC on serial scans [26] has been suggested as a plaque stabilization effect of lipid-lowering therapy [27] which may confound this observation. When considering progressors only, annualized metrics also remained predictive in statin users. Overall, these results need to be interpreted with caution given (1) CAC scores were reported to referring clinicians and thus the presence of statin therapy likely represents more of a marker of high baseline risk, (2) statin use is much higher in this cohort relative to registry participants who possess only one CAC scan and (3) though statin use is listed, there is no confirmation of statin adherence. Further, statin use itself is likely a marker of elevated baseline risk and may not be a causative or protective factor for progression. Specific longitudinal tracking of therapeutic use and/or intensification is not easily captured in our registry but would be an area of future investigation.

Interestingly, the incremental prognostic value of Non-Zero Baseline CAC progression was maintained across all MACE definitions, though the strongest predictive performance was observed for the 5P MACE composite which includes all heart failure events. Although fundamentally a surrogate for atherosclerosis, CAC progression’s robust association with 5P MACE may capture the broader chronic sequelae of advanced coronary disease, including ischemic heart failure. Recent studies also suggest that CAC progression may be a marker of overall neurovascular health [11].

Though used as a covariate to adjust models, the time between scans did not have a significant association with events in prior studies (HR 0.95 [0.89–1.01]; p = 0.13) [16]. However, in our cohort, the scanning interval had an inverse association with the risk of MACE across definitions (HR ∼ 0.73, p < 0.01). Further, the average inter-scan interval for Non-Zero Baseline Progressors was shorter than those with a Zero Baseline CAC (4.4 years vs 5.2 years). While a shorter interval combined with substantial CAC change could theoretically reflect higher progression velocity, this observational finding should be interpreted cautiously. As repeat CAC testing was clinician-directed, earlier repeat imaging may preferentially occur in individuals perceived to be at higher cardiovascular risk. Thus, the observed association between shorter scan intervals and increased event risk may reflect underlying patient risk or clinical decision-making rather than a direct effect of progression velocity. Nevertheless, a divergence point was seen at 5–6 years on cumulative incidence curves (Fig. 2). A slightly shorter scanning interval of 4–5 years could potentially provide necessary lead time for risk stratification while allowing for intensification of therapies before the onset of clinical events, but further study is needed.

Recent work also suggests that changes in CAC density may provide complementary information [28]. Whereas our findings demonstrate Agatston CAC progression, which is based on CAC volume, is associated with increased cardiovascular risk, increases in CAC density have been associated with lower MACE risk, potentially reflecting plaque stabilization [28]. Incorporating both volumetric and density-based measures of CAC progression may therefore further refine risk stratification in future studies.

We acknowledge that our study has multiple limitations. As a retrospective analysis of the CLARIFY registry, there is an inherent risk of selection bias, as the decision to perform a second CAC scan was made by individual clinicians and not dictated by a randomized protocol. Given the widely understood “Power of Zero”, this may have rendered a biased, higher risk cohort than those who would typically obtain only one CT CAC. As suggested by a higher prevalence of traditional risk factors and statin utilization relative to the much larger general screening population within CLARIFY possessing only one CAC score, our findings may apply to a self-selected subgroup of individuals and not a general screening population. Secondly, the study population was predominantly Caucasian, which may limit generalizability. Finally, the Agatston score only measures calcified plaque and does not account for non-calcified or “vulnerable” plaque, which can be visualized by coronary CT angiography (CCTA). Future directions may include interrogating serial CCTA images to assess for changes in their non-calcified plaque burden or pericoronary adipose composition. Additionally, radiomics-based analysis of other anatomic compartments captured on the existing CT CAC scans of this cohort, including epicardial, periaortic, and hepatic adipose tissue, could further strengthen risk prediction models.

5. Conclusion

In a large cohort of 4166 participants from the CLARIFY registry, serial CAC imaging provided incremental prognostic value, though its utility was dependent on the baseline presence of CAC. Whereas Zero Baseline CAC Progression did not render an increase in risk in the near-term, the progression of established plaque was a robust and independent predictor of increased MACE risk.

These findings indicate that serial CAC screening may offer additional prognostic insight in patients with pre-existing atherosclerosis, regardless of underlying traditional risk factors. Additionally, the low near-term risk for patients progressing from a zero baseline CAC score may signal early-stage atherosclerosis and provide a window of opportunity to initiate therapeutic interventions. Ultimately, carefully designed prospective clinical trials are required to better define the role of longitudinal monitoring in patients with established CAC.

Declaration of AI-assisted technologies use

Generative AI technology was utilized to visualize the initial layout and flow of graphical methods of the Central Illustration (top panel) which was then adapted manually. All other figures, including the bottom panel of the Central Illustration were generated manually.

Funding

This work received no funding.

Ethical review statement

This study includes patients from the Community Benefit of No-charge Calcium Score Screening (CLARIFY) Registry (Clinicaltrials.gov, NCT04075162). Given the retrospective nature of the study and use of de-identified patient data, the University Hospitals (UH) Institutional Review Board (IRB #STUDY20190995) waived informed consent.

Data availability statement

The clinical datasets analyzed for this study are not publicly available due to stipulations of the UH IRB protocol limiting de-identified patient data sharing, though they may be made available through institutional review on reasonable request.

CRediT authorship contribution statement

Michael D. Glidden: Writing – review & editing, Writing – original draft, Visualization, Methodology, Formal analysis, Data curation, Conceptualization. Santosh K. Sirasapalli: Writing – review & editing, Visualization, Validation, Software, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Mark Yoder: Data curation. Jean-Eudes Dazard: Methodology, Data curation. Zhuo Chen: Methodology, Data curation. Sai Rahul Ponnana: Methodology, Formal analysis, Data curation. Kanimozhi Sivanantham: Methodology, Data curation. Neda Shafiabadi Hassani: Methodology. Tong Zhang: Methodology, Data curation. Gokul Parameswaran: Methodology, Data curation. Robert A. Okyere: Methodology. Joseph K. Amoah: Methodology. Salil V. Deo: Writing – review & editing, Supervision, Methodology, Formal analysis, Data curation. Ian J. Neeland: Methodology. Robert Gilkeson: Methodology. Imran Rashid: Methodology. Sadeer Al-Kindi: Methodology. Daniel I. Simon: Methodology. Sanjay Rajagopalan: Writing – review & editing, Validation, Supervision, Project administration, Methodology, Formal analysis, Conceptualization.

Declaration of competing interest

The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Sadeer Al-Kindi, MD is an Executive Editor of the American Journal of Preventive Cardiology. If there are other authors, they declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.