Association of Intima‐Media Thickness Measured at the Common Carotid Artery With Incident Carotid Plaque: Individual Participant Data Meta‐Analysis of 20 Prospective Studies

Data were sought from the Proof‐ATHERO consortium; a detailed description of this collaboration has been published elsewhere.³³ For inclusion in the current analysis, participants were required to have data pertaining to (1) baseline CCA‐IMT and (2) carotid plaque status (yes versus no) at baseline and at least at one visit during follow‐up. The baseline visit was defined as the first visit, at which carotid plaque status was available, and follow‐up as subsequent visits. We excluded participants with a baseline history of CVD (defined as coronary heart disease or stroke) or preexisting carotid plaque at baseline from the analysis. Furthermore, to avoid overfitting and convergence issues of statistical models, we excluded studies that recorded <20 events of incident carotid plaque. Moreover, we searched the literature for additional prospective studies on the association of baseline CCA‐IMT with incident carotid plaque in individuals free of carotid plaque at baseline that were published until December 1, 2022. We used the search terms (“intima‐media thickness” [all fields] OR “IMT” [all fields] OR “intima media thickness” [all fields] AND “plaque” [all fields] AND “incident” [all fields] OR “prospective” [all fields]) in PubMed and TS=(“intima‐media thickness” OR “IMT” OR “intima media thickness”) AND TS=(“plaque” AND [“incident” OR “prospective”]) in Web of Science.

Details on the study‐specific definitions of CCA‐IMT and carotid plaque are provided in Table S2 and have been described previously.³³ In quantifying CCA‐IMT, we gave preference to mean CCA‐IMT values or, alternatively, used maximum CCA‐IMT. When studies provided cIMT measurements at several locations of the CCA (ie, near and far wall, left and right side, and different insonation angles), we used the arithmetic mean of all available values. When measuring cIMT, most studies focused on a 10‐mm long segment at the distal part of the CCA (Table S2 and Figure S1). Incident carotid plaque was defined as the development of first‐ever plaque during follow‐up in any segment of the carotid artery (ie, left or right CCA, carotid bifurcation, or internal carotid artery). Fourteen studies (70%) defined carotid plaque as focal thickening, and some others relied on different thresholds of cIMT (Table S2).

Statistical analyses were conducted according to a predefined analysis plan. We calculated odds ratios (ORs) for incident plaque using a 2‐stage approach. We first estimated ORs within each study separately, and then combined study‐specific ORs using random‐effects meta‐analysis using the method of moments procedure of DerSimonian and Laird. Between‐studies heterogeneity was quantified with the I² statistics.³⁴ We conducted complete‐case analyses, if not stated otherwise.

In the primary analysis, we used logistic regression models to estimate ORs for incident plaque per SD higher level of CCA‐IMT, defining the SD of CCA‐IMT within each study separately. The CCA‐IMT distribution was checked for normality by visually inspecting quantile‐quantile plots. We report ORs (1) adjusted for age, sex, and trial arm; and (2) further adjusted progressively for ethnicity, smoking, history of diabetes, body mass index, systolic blood pressure, low‐density lipoprotein cholesterol, high‐density lipoprotein cholesterol, lipid‐lowering medication, antihypertensive medication, estimated glomerular filtration rate, and hs‐CRP (high‐sensitivity C‐reactive protein). We also conducted analyses that expressed ORs per 0.1‐mm higher level of baseline CCA‐IMT. To inspect the shape of association between baseline CCA‐IMT and incident plaque, we calculated ORs across study‐specific CCA‐IMT quintiles, pooled them using multivariate random‐effects meta‐analysis,³⁵ plotted them against the mean CCA‐IMT value within each quintile, and added the best‐fitting line through the OR estimates. We evaluated log linearity of the association between baseline CCA‐IMT and incident carotid plaque by visually inspecting whether OR estimates lie on the corresponding best‐fitting lines. In this analysis, we used floating absolute risks³⁶ to calculate 95% CIs for quintile groups (including the reference group), thereby enabling head‐to‐head comparisons between effect sizes of any 2 of the quintiles.

We also investigated effect modification with formal tests of interaction across clinically relevant predefined variables (ie, age, sex, lipid‐lowering medication, and low‐density lipoprotein cholesterol at baseline and development of CVD during follow‐up). We used meta‐regression³⁷ to test for differences by selected study‐level characteristics (ie, study type and type of CCA‐IMT measurement). In subgroup analyses, we applied Bonferroni correction³⁸ to account for multiple testing (ie, P values ≤0.0071 [0.05/7 tests] were deemed as statistically significant). In addition, we investigated whether ORs varied by median duration of follow‐up using meta‐regression.³⁷ Moreover, we conducted sex‐specific analyses and estimated pooled ORs separately for women and men.

Finally, we conducted sensitivity analyses that: (1) took into account the time to plaque development by use of Cox regression (after ensuring that the proportional hazards assumption was met on the basis of Schoenfeld residuals and the graphical inspection of log[−log] plots), estimating the date of carotid plaque development as the visit at which carotid plaque had first been detected or, alternatively, as the midpoint between this and the preceding visit; (2) used long‐term average CCA‐IMT values (“usual levels”) estimated with regression calibration³⁹ on the basis of repeated CCA‐IMT measurements over time; (3) used within‐study multiple imputation of missing values suggested by Burgess et al⁴⁰ (ie, imputed sporadically missing values in each study separately [80 data sets] before applying the Rubin rule and then combining study‐specific effect sizes with random‐effects meta‐analysis); (4) omitted participants with a large CCA‐IMT value (>1.5 mm), which could be indicative of undetected carotid plaque; and (5) omitted studies that had defined carotid plaque as CCA‐IMT above a specific threshold rather than as focal thickening. We additionally conducted a separate sensitivity analysis that compared the association of baseline CCA‐IMT with carotid plaque development at the same side of the neck (ie, right CCA‐IMT with right carotid plaque and left CCA‐IMT with left carotid plaque) and at the opposite side of the neck (ie, right CCA‐IMT with left carotid plaque and left CCA‐IMT with right carotid plaque).

In addition, we meta‐analyzed the results of the studies from the Proof‐ATHERO consortium with the studies we found in the literature for which we were not able to obtain individual participant data. We focused on the Proof‐ATHERO studies included in our multivariable‐adjusted meta‐analysis to enhance the comparability to the studies from the literature. Again, we meta‐analyzed ORs for incident carotid plaque per SD higher baseline CCA‐IMT using random‐effects meta‐analysis.

All statistical tests were 2‐sided, and we deemed P≤0.05 as statistically significant, unless specified otherwise. Statistical analyses were conducted using Stata version 15.1 (StataCorp).

The derivation of the study sample contributing to the present study is outlined in Figure 1. Of the 74 studies involved in the Proof‐ATHERO consortium, we excluded 48 that did not record incident carotid plaque. After further excluding participants who did not meet the prespecified inclusion criteria and excluding studies that recorded <20 incident carotid plaque events, a total of 20 studies involving 21 494 participants remained for analysis.^{15, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59}

Table 1 and Table S3 summarize key characteristics of the studies and participants we analyzed. Twelve studies recruited participants from the general population, 6 recruited participants from high‐risk populations (ie, individuals with baseline coronary atherosclerosis, renal disease, or other vascular risk factors), and 2 were clinical trials (involving individuals on hemodialysis and with heterozygous familial hypercholesterolemia). The pooled mean age at baseline was 56 years (SD, 9 years); 55% of the participants were women. The overall mean of baseline CCA‐IMT values was 0.71 mm (SD, 0.17 mm), with 15 studies reporting mean CCA‐IMT values and 5 studies reporting maximum CCA‐IMT values. Over a median follow‐up of 5.9 years (5th–95th percentile, 1.9–19.0 years), 8278 participants (39%) developed first‐ever carotid plaque.

Figure 2 depicts development of carotid plaque across quintiles of baseline CCA‐IMT. In the first, second, third, fourth, and fifth quintile, 1293 (28.9%), 1419 (33.1%), 1614 (36.8%), 1737 (41.7%), and 2215 (53.0%) individuals developed incident carotid plaque, respectively. The odds appeared to increase log‐linearly across CCA‐IMT quintiles when adjusting for age, sex, and trial arm as well as in the multivariable‐adjusted model.

The pooled OR for first‐ever carotid plaque development, adjusted for age, sex, and trial arm, was 1.40 (95% CI, 1.31–1.50; I²=63.9%) per SD higher level of baseline CCA‐IMT (for study‐specific estimates, see Figure S2). The corresponding OR per 0.1‐mm higher baseline level of CCA‐IMT was 1.30 (95% CI, 1.23–1.38; I²=71.8%). As shown in Table 2, the association was slightly weakened when the OR was further adjusted for potential confounding variables. In a model further adjusted for ethnicity, smoking, diabetes, body mass index, systolic blood pressure, low‐ and high‐density lipoprotein cholesterol, and lipid‐lowering and antihypertensive medication, the OR per SD higher baseline CCA‐IMT was 1.34 (95% CI, 1.24–1.45; I²=59.4%; 14 studies; 16 297 participants; 6381 incident carotid plaques). The ORs were virtually identical when further adjusted for estimated glomerular filtration rate or log‐transformed hs‐CRP values. In subgroup analyses (Figure 3), there was no evidence for effect modification by age, sex, intake of lipid‐lowering medication, low‐density lipoprotein cholesterol, development of CVD during follow‐up, type of study, and type of CCA‐IMT measure, when we considered a multiplicity‐adjusted threshold for statistical significance (all P>0.0071). In addition, we found no statistically significant difference in ORs by median duration of follow‐up, as demonstrated in Figure S3 (P=0.804). As shown in Table S4, results were also similar in sex‐specific analyses. The age‐ and trial arm–adjusted OR for incident carotid plaque per SD higher baseline CCA‐IMT was 1.38 (95% CI, 1.24–1.53; I²=69.0%; 18 studies; 11 756 participants; 4228 incident carotid plaques) in women and 1.39 (95% CI, 1.31–1.46; I²=10.8%; 18 studies; 8980 participants; 3611 incident carotid plaques) in men.

In sensitivity analyses, we observed similar ORs when we multiplied imputed missing values, excluded individuals with CCA‐IMT values >1.5 mm, or restricted analyses to studies that defined carotid plaque as focal thickening (Figure 4A). Stronger associations were observed when we considered long‐term averages (“usual levels”) of CCA‐IMT values, which we estimated on the basis of repeated CCA‐IMT measurements taken at a median of 2 occasions (range, 2–9 occasions). Median time between 2 consecutive CCA‐IMT measurements was 3.0 years (interquartile range, 2.3–5.4 years). The OR per SD higher “usual” CCA‐IMT was 1.71 (95% CI, 1.54–1.89; I²=63.9%) when adjusted for age, sex, and trial arm and 1.65 (95% CI, 1.44–1.88; I²=59.4%) in the multivariable‐adjusted model. When we used Cox regression and estimated the dates of plaque development as the visit at which plaque had first been detected, the hazard ratio (HR) for incident plaque per SD higher baseline CCA‐IMT was 1.24 (95% CI, 1.17–1.30; I²=74.4%) when adjusted for age, sex, and trial arm and 1.16 (95% CI, 1.09–1.24; I²=74.8%) in the multivariable‐adjusted model. When we estimated dates of plaque development as the midpoint between the visit at which plaque had first been detected and the preceding visit, HRs for incident plaque per SD higher baseline CCA‐IMT were 1.28 (95% CI, 1.22–1.33; I²=64.8%) when adjusted for age, sex, and trial arm and 1.22 (95% CI, 1.16–1.29; I²=65.7%) in the multivariable‐adjusted model. Finally, side‐specific analyses revealed somewhat stronger associations for an ipsilateral development than a contralateral development of carotid plaque (Figure 4B).

We identified 5 studies from the literature to supplement our multivariable individual participant data meta‐analysis (Figure S4).^{14, 17, 27, 28, 29} The pooled OR for carotid plaque per SD higher baseline CCA‐IMT based on data from these 5 studies was 1.28 (95% CI, 1.14–1.43; I²=20.1%; 5 studies; 3736 participants). When meta‐analyzing ORs of the studies from the Proof‐ATHERO consortium that were included in the multivariable‐adjusted meta‐analysis with aggregated data of these 5 studies, the pooled OR for incident carotid plaque per SD higher baseline CCA‐IMT was 1.33 (95% CI, 1.24–1.42; I²=54.1%; 18 studies; 19 295 participants).

We have previously investigated the relationship between CCA‐IMT and development of carotid plaque in a literature‐based meta‐analysis that considered 7 general population studies with a total of 9341 participants and 1288 events of carotid plaque.³⁰ In this analysis preceding the current study, we had observed a pooled relative risk for incident carotid plaque of 1.78 (95% CI, 1.53–2.07), when comparing individuals in the top quartile of baseline CCA‐IMT with individuals in the bottom quartile. Although this effect size is comparable to the effect size in the current study (see results across quintiles in Figure 2), a key strength of the current analysis is that it included 6 times more incident outcomes and could therefore quantify the association more precisely (in addition to other advantages related to the individual participant data access). We were also able to include hitherto unpublished findings from 15 studies and extended the analysis to high‐risk populations and clinical trials. When meta‐analyzing the studies contributing to the Proof‐ATHERO consortium with the aggregated data of the additional studies we found in the literature,^{14, 17, 27, 28, 29} we identified a multivariable‐adjusted OR for incident carotid plaque of 1.33 (95% CI, 1.24–1.42; I²=54.1%) per SD higher CCA‐IMT, which is nearly the same result as in the present multivariable‐adjusted primary analysis. Our findings are also in line with results from other studies in the literature that analyzed the association of CCA‐IMT with carotid plaque differently. The Tromsø study, for instance, observed a positive association between baseline cIMT and a higher number of plaques at follow‐up.¹⁹ The SHIP (Study of Health in Pomerania) reported that individuals with elevated CCA‐IMT had a higher risk for developing additional plaques in previously unaffected arterial segments.¹⁸ In contrast to these studies and our report, the Reykjavik Risk Evaluation for Infarct Estimates study found no statistically significant association between CCA‐IMT and formation of a new plaque.²⁰

Measurement of cIMT and carotid plaque is generally performed noninvasively with high‐resolution B‐mode ultrasound. cIMT is defined as the so‐called double‐line pattern, representing the distance between the lumen‐intima and the media‐adventitia interfaces.⁶⁰ The 2011 Mannheim cIMT and plaque consensus recommends cIMT to be measured at the far wall of the CCA in an area free of carotid plaque.⁶¹ In 2008, the American Society of Echocardiography also recommended measuring CCA‐IMT at the far wall of the carotid artery in their Consensus Statement but, contrarily, to include sections with carotid plaque.⁶² In the studies contributing to the present report, there were some differences in how CCA‐IMT was assessed (Table S2 and Figure S1). cIMT was often measured at different sections of the CCA, at the left and/or right side of the neck, and at the near and/or far wall of the CCA. To reduce variability and include a broad range of information from the entire CCA, we averaged all the available measurements to obtain an overall CCA‐IMT value. Moreover, in a meta‐regression analysis (Figure 3), we observed that the association was similarly strong in studies reporting mean CCA‐IMT and studies reporting maximum CCA‐IMT.

Besides different definitions of CCA‐IMT, studies also varied in terms of carotid plaque assessment (Table S2). The Mannheim cIMT and plaque consensus defines carotid plaque as focal thickening of at least 0.5 mm or 50% of its surrounding area or as cIMT >1.5 mm.⁶¹ Similarly, the American Society of Echocardiography recommends defining carotid plaque as “(1) any focal thickening thought to be atherosclerotic in origin and encroaching into the lumen of any segment of the carotid artery (protuberant‐type plaque) or (2) in the case of diffuse vessel wall atherosclerosis, when carotid intima‐media thickness measures ≥1.5 mm in any segment of the carotid artery (diffuse‐type plaque).”⁶³ Although most of the studies contributing to our analysis defined carotid plaque as focal structure, some others defined it as cIMT above a predefined threshold. The latter may be problematic in the present analysis because cIMT is assumed to thicken progressively over time, and a direct association between elevated baseline cIMT and carotid plaque development in those studies would therefore be a logical consequence. Reassuringly, though, when we excluded these studies in a sensitivity analysis, the effect size pooled across the remaining studies was similar as in the primary analysis (OR, 1.38 [95% CI, 1.29–1.47]). Another potential challenge is that early stage of plaque development may sometimes be misclassified as elevated cIMT.⁶¹ Therefore, we conducted a sensitivity analysis that omitted individuals with CCA‐IMT >1.5 mm, which could be indicative for carotid plaque. Again, this analysis yielded results comparable to the primary analysis, with an overall OR for carotid plaque development of 1.41 (95% CI, 1.32–1.50) per SD higher level of baseline CCA‐IMT. Although we did not observe significant effect modification by differences in ultrasound protocols, discrepancies in definitions of cIMT and carotid plaque are suboptimal and standardizations of measurement techniques would be an essential approach to obtain adequate comparisons.⁶¹

As atherosclerosis often develops over years without symptoms or detection, early identification of vulnerable individuals is the key to prevent its clinical sequelae. Current evidence shows that increased cIMT relates to unfavorable levels of risk factors,^{64, 65, 66} presence of atherosclerosis elsewhere in the arterial system,⁴ and the risk of future CVD events.^{6, 7} We have previously shown in an analysis of 119 clinical trials that different types of interventions reduce progression of cIMT and that the greater reductions in cIMT progression are associated with greater reductions in CVD risk, endorsing its usefulness as a surrogate marker.⁶⁷ Leading on from this, we now provide further evidence for the role of cIMT as a risk marker for atherosclerotic disease, which may help to identify individuals at risk of developing advanced atherosclerotic lesions earlier.

The present analysis has several strengths. First, we analyzed data of the Proof‐ATHERO consortium, the worldwide largest consortium with data on repeated assessments of atherosclerosis and CVD, and included 20 different studies with >21 000 individuals. Thus, a major strength of the current analysis is its large sample size, which allows estimating effect sizes with adequate precision. Second, we included data from studies in a variety of clinical settings, thereby enhancing the generalizability of our findings to various populations. Third, we excluded individuals with a history of CVD, reducing the potential influence of subsequent drug treatments or frequent medical checks on the development of carotid plaques. Fourth, access to individual participant data allowed us to harmonize outcomes, exposures, and levels of adjustment, and perform various participant‐level sensitivity analyses. Fifth, in a sensitivity analysis, we capitalized on the serial CCA‐IMT measurements available in our studies and estimated ORs for incident carotid plaque based on long‐term averages of CCA‐IMT rather than a single baseline measurement, thereby taking into account within‐person variation of CCA‐IMT during follow‐up. Sixth, because we had access to participant‐level data, we were able to study the shape of association between CCA‐IMT and development of carotid plaque across fifths of baseline CCA‐IMT. Seventh, compared with our previous literature‐based meta‐analysis, individual participant data meta‐analysis enabled a sophisticated analysis of effect sizes across participant‐level subgroups. Our study also has limitations. First, there were differences in how the individual studies defined and measured CCA‐IMT and carotid plaque. Second, because carotid plaque status was only available at the study visits and not in between, there was uncertainty about the exact time point of plaque development. For this reason, we prespecified to use logistic regression in our primary analysis. When we used Cox regression based on the estimated time to plaque development, HRs were highly significant, although numerically lower than ORs, as expected when the rare disease assumption is not met (39% developed the outcome plaque). Third, because the present meta‐analysis focused on plaque status, we cannot draw any conclusions about the relationship of CCA‐IMT with plaque quality, size, or architecture, which are more detailed measures to quantify and characterize carotid plaque.⁸ Fourth, our analysis includes long‐term follow‐up studies, with baseline examinations typically taking place in 1990s to early 2000s,³³ and ultrasound devices have improved significantly since then. Consequently, it can be assumed that it would nowadays be possible to obtain ultrasound images with higher resolution, which would also enable us to identify plaques of smaller dimension. Furthermore, although recent guidelines also suggest measuring 3‐dimensional carotid plaque by applying modern ultrasound techniques,⁶³ we were only able to analyze 2‐dimensional carotid plaque data because of the unavailability in our long‐term follow‐up studies. Also, we only investigated ultrasound‐based markers measured in the carotid arteries. Further research is needed on whether our results also hold for other vascular beds (eg, the femoral arteries).⁸ Finally, plaque data on specific carotid arterial segments (ie, CCA, carotid bifurcation, and internal carotid artery) were sparse and could not be considered in our analysis, which prevented us from investigating the association between CCA‐IMT and segment‐specific development of carotid plaque.