Background
The aim of this study was to evaluate whether the addition of ultrasound carotid intima-media thickness (CIMT) measurements and risk categories of plaque help predict incident stroke and cardiovascular disease (CVD) in older adults.
Methods
Carotid ultrasound studies were recorded in the multicenter Cardiovascular Health Study. CVD was defined as coronary heart disease plus heart failure plus stroke. Ten-year risk prediction Cox proportional-hazards models for stroke and CVD were calculated using Cardiovascular Health Study–specific coefficients for Framingham risk score factors. Categories of CIMT and CIMT plus plaque were added to Framingham risk score prediction models, and categorical net reclassification improvement (NRI) and Harrell’s c-statistic were calculated.
Results
In 4,384 Cardiovascular Health Study participants (61% women, 14% black; mean baseline age, 72 ± 5 years) without CVD at baseline, higher CIMT category and the presence of plaque were both associated with higher incidence rates for stroke and CVD. The addition of CIMT improved the ability of Framingham risk score–type risk models to discriminate cases from noncases of incident stroke and CVD (NRI = 0.062, P = .015, and NRI = 0.027, P < .001, respectively), with no further improvement by adding plaque. For both outcomes, NRI was driven by down-classifying those without incident disease. Although the addition of plaque to CIMT did not result in a significant NRI for either outcome, it was significant among those without incident disease.
Conclusions
In older adults, the addition of CIMT modestly improves 10-year risk prediction for stroke and CVD beyond a traditional risk factor model, mainly by down-classifying risk in those without stroke or CVD; the addition of plaque to CIMT adds no statistical benefit in the overall cohort, although there is evidence of down-classification in those without events.
The Framingham risk score (FRS) and other traditional cardiovascular disease (CVD) risk factors and algorithms have important predictive value for stroke and other CVD end points. Nonetheless, the majority of incident stroke and other CVD events occur in the low- and intermediate-risk groups characterized by these risk factor predictors. Previous reports have documented an association between carotid intima-media thickness (CIMT) and/or plaque with stroke, transient ischemic attacks, and other clinical manifestations of CVD.
Despite what is known regarding the importance of traditional CVD risk factors and measures of subclinical disease such as CIMT and plaque in predicting future stroke and other CVD events, there is a paucity of information regarding the relative prognostic value of adding carotid ultrasound measurement information to traditional risk factors in elderly individuals. Consequently, we evaluated, in a multicenter cohort of older adults without CVD at baseline, whether CIMT measurements and plaque could add incremental value to traditional risk factors in predicting the 10-year risk for incident stroke and CVD.
Methods
Study Population
The Cardiovascular Health Study (CHS) is a population-based prospective study of men and women aged ≥65 years at baseline. The mean age of the study population at baseline was 72.8 ± 5.6 years. The overall study design for CHS has been previously published. Briefly, between 1989 and 1990, CHS enrolled 5,201 participants using Medicare eligibility lists in four communities: Forsyth County, North Carolina; Sacramento County, California; Washington County, Maryland; and Pittsburgh, Pennsylvania. A second cohort of 687 black participants was recruited between 1992 and 1993. Participants included in this analysis had no evidence of coronary heart disease (CHD), heart failure (HF), or stroke at baseline. All participants underwent a baseline clinical examination that included history, physical examination, blood drawing, carotid ultrasound, and other tests.
Carotid Ultrasonography
Carotid arteries were evaluated at baseline using high-resolution B-mode ultrasonography (model SSA-270A ultrasound machine; Toshiba, Tustin, CA). The scanning protocol has been previously described in detail. The protocols for recording carotid ultrasound studies and measuring CIMT were the same for the scans performed in 1989–1990 and 1992–1993. Both examinations used on-site videotapes as well as direct image capture to a Macintosh II computer, with the digital images and videotapes sent to the Ultrasound Reading Center for subsequent review and processing. The CHS protocol was such that after imaging of the common carotid artery (CCA) below the carotid artery bulb, images were acquired—with the ultrasound beam centered on the internal carotid artery (ICA) flow divider—from the anterolateral, lateral, and posterolateral projections. Plaque measurements were made in either the proximal ICA or the bulb, whichever site had the largest wall protrusion. If a protrusion was not seen, imaging was centered on the carotid bulb.
Quantitative measurements of CIMT were performed on one longitudinal image of the CCA and three longitudinal images of the ICA recorded from both the right and left carotid arteries. Measurements were performed on an image that was selected from a sequence of images replayed from a digital playback buffer. Frames that were free of motion (i.e., in which the preceding and following images showed no motion) were selected. Although there was no attempt to select on the basis of the cardiac cycle, subsequent review of the images has shown that this tended to be at end-diastole. A mouse-activated drawing tool was used to trace the boundaries of the lumen-intima and media-adventitia interfaces of the arterial wall. The distance between these two lines corresponded to the combined thickness of the intima and media. Maximal intima-media thickness (IMT) of the CCA and ICA was calculated as the mean of the maximal IMT of the near and far walls from both left and right carotid arteries. The CIMT measure used in these analyses was the average of maximal CCA and ICA IMT as defined above after standardization (i.e., after subtraction of the mean and division by the standard deviation of the measurement). Focal plaque, when present, was included in the maximum IMT measurement. Gender-specific percentile categories were created using cut points at the 25th and 75th percentile of standardized CIMT for each gender.
Plaque was defined on the basis of the presence of the greatest perceived protrusion of the carotid wall (specifically the IMT) in either the carotid bulb or the proximal ICA. Three plaque categories were defined, no plaque, intermediate risk, and high risk, on the basis of lesion surface, echogenicity, and texture characteristics. High-risk plaques were defined as having at least one of the following characteristics: irregular or ulcerated surface, echolucency, or heterogeneous texture. Individuals with no plaque had lesion surfaces specified as smooth, with lesion density and morphology both specified as “no lesion.” Any other combinations of lesion characteristics were defined as intermediate risk.
Data on intersonographer and interreader variability for CCA and ICA far wall thickness and residual lumen have been previously published. The mean ± SD maximal absolute intersonographer difference for the far wall CCA IMT measurement was 0.20 ± 0.26 mm ( R = 0.52) and for the far wall ICA IMT was 0.65 ± 0.69 mm ( R = 0.52). The mean ± SD maximal absolute interreader differences in IMT measurements were lower for the interreader comparisons: 0.09 ± 0.05 mm for the CCA IMT ( R = 0.91) and 0.41 ± 0.57 mm for the ICA IMT ( R = 0.81).
Cardiovascular Event Ascertainment
Methods used to assess CVD events including stroke and CHD in CHS have been reported previously. Briefly, in CHS, potential clinical events were identified through (1) clinic visits and surveillance calls by the field centers, (2) participant-initiated reports, and (3) secondary sources of events, including review of medical records and Medicare hospitalization data. The CHS Events Committee adjudicated CVD events by reviewing all pertinent data, including history, physical examination, chest radiography report, and medication use. CHD was defined as angina as well as nonfatal and fatal myocardial infarction, coronary artery bypass grafting, or angioplasty. CVD was defined as a composite of nonfatal or fatal CHD, HF, or stroke during the follow-up period. Cause of death was adjudicated by the Events Committee. All deaths due to atherosclerotic CHD were captured in the CHS definition, and all deaths due to atherosclerotic CHD or cerebrovascular disease were captured in the CVD definition. Individuals were censored at the earliest of the following: date of death, date of loss to follow-up, or 10 years.
Statistical Analyses
Stata version 12 software (StataCorp LP, College Station, TX) was used for analyses. Baseline characteristics were summarized according to gender-specific categories of CIMT (<25th percentile, 25th to 75th percentile, and >75th percentile). Incidence rates of stroke and CVD were calculated per 1,000 person-years as a function of CIMT percentile categories and plaque (absent, intermediate risk, and high risk). Cox proportional-hazards models were used to determine—after adjusting for traditional FRS factors (age, gender, race, hypertension medications, systolic blood pressure, diabetes mellitus, total and high-density lipoprotein cholesterol, and smoking)—the associations of CIMT categories and plaque categories with incident stroke and CVD within 10 years. Cox models were used to predict 10-year risk for stroke and CVD for a (1) base model using the CHS-specific coefficients for traditional FRS factors, (2) CIMT model adding CIMT categories to the base model, and (3) full model adding plaque categories to the CIMT model. We also examined the full model using plaque (present vs absent) rather than plaque categories.
We assessed calibration by calculating likelihood ratio test P values from the survival-adapted Hosmer-Lemeshow test. In this context, P values < .05 would suggest the model was not well calibrated and that there was a significant difference between expected and observed event rates. Harrell’s c-statistic was calculated for each model, and predicted risks were categorized into <5%, 5% to <10%, 10% to <20%, and ≥20%. The predicted risk categories were used to compute the net reclassification improvement (NRI) statistic for those experiencing events, those without events, and overall. The event NRI was calculated as the proportion of individuals who were reclassified to a higher risk category minus the proportion reclassified to a lower risk category. The nonevent NRI was calculated as the proportion of individuals who were reclassified to a lower risk category minus the proportion reclassified to a higher risk category. The overall NRI was calculated as the sum of the event NRI and the nonevent NRI. The NRI data and difference in c-statistic were used to compare the CIMT and full models with the base model and the full model with the CIMT model. Further analyses examined the NRI statistic comparing the CIMT model with the base model restricted to those with plaque and to those with high-risk plaque.
It has been suggested that it is individuals with intermediate FRS risk in whom the addition of the carotid ultrasound measures (e.g., of CIMT and plaque) to the base model is likely to be most useful clinically. Consequently, clinical NRIs (cNRI) were calculated for the primary analysis by restricting the calculation of the NRI to those who were in the intermediate FRS risk groups (5%–20%) for the initial model. Other than this restriction, cNRI was calculated similarly to the NRI except that reclassification was upward or downward only if movement occurred to the ≥20% category or the 0% to 5% category, respectively; movements within the intermediate categories were not considered to be reclassifications. As a sensitivity analysis, we also examined NRI statistics comparing the addition of plaque categories to the base model and comparing the full model with the base plus plaque categories model. In additional sensitivity analyses, the NRI statistics were calculated comparing the CIMT model with the FRS-type model and comparing the full model with the CIMT model, with three different restrictions. First, we excluded individuals who were taking lipid medications; second, we excluded individuals with prevalent peripheral arterial disease (PAD); and third, we included only ischemic stroke as the stroke outcome and censored those with hemorrhagic or unclassified stroke type. PAD was defined as the presence of either claudication (adjudicated) or an ankle-arm index < 0.9. The NRI statistics for the first two were calculated for both stroke and CVD, while those for ischemic stroke were not calculated for CVD.
Results
Of the 5,888 CHS participants, 1,406 were excluded from the analysis because of the presence of CHD, HF, or stroke at baseline. In addition, 25 were excluded because of missing carotid ultrasound data, and 73 were excluded because of missing data for the clinical covariates. Consequently, the analyses presented included 4,384 CHS participants (61% women, 14% black; mean baseline age, 72 ± 5 years). There were 482 strokes included in this analysis. Of these, 450 were classified as ischemic or nonischemic; 397 (88.2%) were ischemic. Of the 1,510 cases of CVD, considering only time at first event, nine had CHD, CHF, and stroke at the same time; nine had stroke and HF; 19 had stroke and CHD; 247 had HF and CHD; 248 had only HF; 641 had only CHD; and 337 had only stroke. With regard to the stroke outcome, six were lost to follow-up and 1,009 died before a stroke or 10 years of follow-up; for the CVD outcome, five were lost to follow-up and 584 died before a CVD event or 10 years of follow-up.
Demographic and risk factor characteristics in the cohort are presented in Table 1 as a function of CIMT percentile category. As can be seen from Table 1 , higher age, systolic blood pressure, total cholesterol, percentage of participants taking hypertension medications, percentage with diabetes, and percentage who were current smokers were associated with higher CIMT percentile category. The mean ± SD of maximal CCA IMT was 1.066 ± 0.217, and the mean ± SD of maximal ICA IMT was 1.440 ± 0.567; these values were used to standardize these measures. After taking the mean of the standardized maximal CCA IMT and maximal ICA IMT, the 25th and 75th percentiles of the summary CIMT measure were −0.895 and 0.176 for women and −0.583 and 0.721 for men, respectively; these cut points were used to categorize CIMT.
| CIMT range | <25th percentile ( n = 1,095) | 25th to 75th percentiles ( n = 2,194) | >75th percentile ( n = 1,095) |
|---|---|---|---|
| Age (y) | 70.78 ± 4.63 | 72.25 ± 5.17 | 74.46 ± 6.00 |
| Systolic BP (mm Hg) | 129.35 ± 18.70 | 136.33 ± 21.10 | 142.94 ± 22.26 |
| Cholesterol (mg/dL) | 208.82 ± 37.95 | 211.32 ± 37.57 | 217.47 ± 41.84 |
| HDL (mg/dL) | 57.70 ± 16.71 | 55.46 ± 15.86 | 53.39 ± 14.23 |
| African Americans | 112 (10.2%) | 339 (15.5%) | 184 (16.8%) |
| Men | 428 (39.1%) | 857 (39.1%) | 428 (39.1%) |
| Hypertension medication | 329 (30.0%) | 876 (39.9%) | 512 (46.8%) |
| Diabetes | 100 (9.1%) | 287 (13.1%) | 221 (20.2%) |
| Smoking category | |||
| Never smoked | 585 (53.4%) | 1,063 (48.5%) | 432 (39.5%) |
| Former smoker | 415 (37.9%) | 889 (40.5%) | 460 (42.0%) |
| Current smoker | 95 (8.7%) | 242 (11.0%) | 203 (18.5%) |
| Plaque risk category | |||
| None | 675 (61.6%) | 363 (16.5%) | 18 (1.6%) |
| Intermediate risk | 169 (15.4%) | 586 (26.7%) | 183 (16.7%) |
| High risk | 251 (22.9%) | 1,245 (56.7%) | 894 (81.6%) |
Figure 1 presents the incidence rates of stroke per 1,000 person-years as a function of CIMT percentile in those with or without plaque. Note that there is a graded relationship between CIMT percentile and the incidence rate of stroke in both those with and without plaque (the P values comparing >75th percentile of CIMT category to <25th percentile of CIMT are <.001 and .025, respectively). Figure 2 displays the incidence rate for overall CVD as a function of CIMT percentile in those with or without plaque. Similarly, in individuals both with ( P < .001) and without ( P = .022) plaque, there was a significantly higher incidence rate for CVD among participants in the >75th percentile for CIMT compared with those in the <25th percentile for CIMT.
CIMT categories and plaque categories were both associated with incident CVD and incident stroke within 10 years when added individually to the baseline model (all P < .001, data not shown). Hazard ratios for all variables in our prediction models for stroke and CVD outcomes are presented in Supplemental Table 1 (available at www.onlinejase.com ). Analysis of the c-statistics revealed that the addition of CIMT improved the ability of the base FRS model to discriminate cases from noncases of incident stroke (Harrell’s c = 0.711 vs 0.699, P = .01), as well as overall CVD (Harrell’s c = 0.679 vs 0.669, P < .001). There was no significant improvement by adding plaque category to the CIMT model ( P = .464 and P = .609, respectively).
Table 2 presents results for reclassification of 10-year stroke risk related to adding CIMT to the base model. The upper section of the table presents data in participants without incident stroke (noncases), whereas the lower section presents data on participants with incident stroke. In those without incident stroke, there were 381 reclassified into a higher risk category and 541 reclassified into a lower risk category by adding CIMT to the base model. Subtracting 381 from 541 and dividing by the overall number of participants in this table resulted in a calculated NRI of 0.041 for nonevents ( P < .001). Similarly calculated, the NRI in the incident stroke group was 0.021 ( P = .391), and the overall NRI for stroke was 0.062 ( P = .015). Table 3 shows an analysis for reclassification of 10-year overall CVD risk related to adding CIMT to the base model. Note that a similar NRI analysis for overall CVD revealed an overall NRI of 0.027 ( P < .001), also driven by reclassifying those without incident CVD into a lower risk category.
| 10-y risk in model without standardized CIMT | 0%–5% | 5%–10% | 10%–20% | ≥20% |
|---|---|---|---|---|
| 10-y risk in model with standardized CIMT in participants without incident stroke (noncases) | ||||
| 0%–<5% | 307 | 19 † | 0 | 0 |
| 5%–<10% | 129 ∗ | 1,370 | 195 † | 0 |
| 10%–<20% | 0 | 278 ∗ | 874 | 167 † |
| ≥20% | 0 | 0 | 134 ∗ | 429 |
| NRI nonevents = 0.041 | 541 reclassified into lower risk | 381 reclassified into higher risk | ||
| 10-y risk in model with standardized CIMT in participants with incident stroke (cases) | ||||
| 0%–<5% | 8 | 1 † | 0 | 0 |
| 5%–<10% | 4 ∗ | 84 | 23 † | 0 |
| 10%–<20% | 0 | 23 ∗ | 142 | 49 † |
| ≥20% | 0 | 0 | 36 ∗ | 112 |
| NRI events = 0.021 | 63 reclassified into lower risk | 73 reclassified into higher risk | ||
∗ Reclassified into lower risk.
| 10-y risk in model without standardized CIMT | 0%–5% | 5%–10% | 10%–20% | ≥20% |
|---|---|---|---|---|
| 10-y risk in model with standardized CIMT in participants without CVD (noncases) | ||||
| 0%–<5% | 0 | 0 † | 0 | 0 |
| 5%–<10% | 0 ∗ | 1 | 0 † | 0 |
| 10%–<20% | 0 | 2 ∗ | 369 | 67 † |
| ≥20% | 0 | 0 | 180 ∗ | 2,255 |
| NRI nonevents = 0.040 | 182 reclassified into lower risk | 67 reclassified into higher risk | ||
| 10-y risk in model with standardized CIMT in participants with CVD (cases) | ||||
| 0%–<5% | 0 | 0 † | 0 | 0 |
| 5%–<10% | 0 ∗ | 0 | 0 † | 0 |
| 10%–<20% | 0 | 1 ∗ | 52 | 13 † |
| ≥20% | 0 | 0 | 31 ∗ | 1,413 |
| NRI events = −0.013 | 32 reclassified into lower risk | 13 reclassified into higher risk | ||
∗ Reclassified into lower risk.
Additional NRI analyses revealed that addition of plaque category to the CIMT model did not significantly improve the prediction of incident stroke or CVD ( P = .184 and P = .307, respectively, data not shown). Calculations of cNRI revealed that adding CIMT to the base model was significant for stroke (cNRI = 0.126, P < .001), and adding plaque category to the CIMT model also resulted in a significant cNRI of 0.086 ( P < .001) for stroke (see Supplemental Table 2 ; available at www.onlinejase.com ). Calculations of the cNRI for overall CVD revealed that neither the addition of CIMT to the base model (cNRI = 0.044, P = .44) nor the addition of plaque category to the CIMT model (cNRI = 0.013, P = .73) resulted in a statistically significant cNRI.
In a secondary analysis, in participants with any plaque, the addition of CIMT to the base model improved prediction for both stroke (NRI = 0.065, P = .013) and overall CVD events (NRI = 0.019, P < .001) ( Table 4 ). Even in participants with high-risk plaque, there was a modest incremental benefit of adding CIMT to the base model for CVD (NRI = 0.014, P = .016) and for stroke (NRI = 0.058, P = .039).
| 10-y risk in model with no CIMT | 0%–5% | 5%–10% | 10%–20% | ≥20% |
|---|---|---|---|---|
| 10-y risk in model with standardized CIMT in participants with plaque and without disease (noncases) | ||||
| 0%–<5% | 0 | 0 † | 0 | 0 |
| 5%–<10% | 0 ∗ | 0 | 0 † | 0 |
| 10%–<20% | 0 | 1 ∗ | 116 | 12 † |
| ≥20% | 0 | 0 | 64 ∗ | 1,869 |
| NRI nonevents = 0.026 | 65 reclassified into lower risk | 12 reclassified into higher risk | ||
| 10-y risk in model with standardized CIMT in participants with plaque and disease (cases) | ||||
| 0%–<5% | 0 | 0 † | 0 | 0 |
| 5%–<10% | 0 ∗ | 0 | 0 † | 0 |
| 10%–<20% | 0 | 0 ∗ | 19 | 3 † |
| ≥20% | 0 | 0 | 11 ∗ | 1,233 |
| NRI events = −0.006 | 11 reclassified into lower risk | 3 reclassified into higher risk | ||
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
