We believe that the report by Peters et al. in this issue of JASE makes a significant contribution to our understanding of how to perform carotid intima-media thickness (IMT) measurements in clinical trials and possibly in clinical practice.
In their report, the investigators describe the level of data completeness achieved when acquiring carotid IMT data using a clinical trial protocol applied in the Measuring Effects on Intima-Media Thickness: An Evaluation of Rosuvastatin (METEOR) study. The investigators conclude in favor of complex protocols for IMT acquisition and measurements to achieve greater precision and to improve over what can be achieved with less complex imaging protocols.
Peters et al. ‘s data, as presented, give us an opportunity for an open discussion of what carotid IMT protocols can and cannot achieve. Although the investigators’ findings address the very focused and technical aspects of an IMT clinical trial, we believe that there is room to extrapolate some of their findings to the clinical realm. We also offer a plausible alternate interpretation of their results. Our conclusions are different from those reached by the investigators: we believe that the data they present should not discourage the use of a relatively simple IMT scan protocol and that the benefits of multiple-angle IMT protocols do not necessarily outweigh their costs.
In METEOR, each side of the carotid bifurcation was scanned from five different angles of interrogation. The method of segmenting the carotid artery near the bifurcation conforms to the three-level protocol used in the Atherosclerosis Risk in Communities (ARIC) study. Sampling from five angles was repeated in the common carotid artery, the carotid bifurcation, and the proximal internal carotid artery on each side. The number of images generated at each level and on each side was five, with a possible maximum of 30 images for each enrollee. Measurements were attempted for all these segments. The investigators then conducted an analysis of the number of images of sufficient quality to measure IMT. The data completeness achieved by such a comprehensive approach would seem to set a new standard for what is needed in IMT imaging protocols, whether they are used for clinical trials or for clinical risk assessment. This concept is echoed by Peters et al. ‘s conclusion: “Extensive ultrasound protocols are required to obtain the highest precision to observe a treatment effect and to fully cover the degree of atherosclerotic burden.” But is this really the case?
Limitations Due to Enrollee Selection
The results of the present study should be taken with a bit of caution. This is not a study in which all candidates were enrolled and imaged, and then the resultant IMT image quality was evaluated. A methods report describing the enrollment phase of the METEOR study indicates that 5,751 individuals were screened. Peters et al. indicate that 1.5% of screened individuals were not included in the METEOR study, because of incomplete data or the inability to estimate whether IMT was within the required range of 1.2 to 3.5 mm. The 1.5% of 5,751 candidate enrollees corresponds to 86 individuals. We believe that these 86 should have been accounted for when Peters et al. investigated the factors associated with incomplete IMT data. Are the investigators really evaluating the performance of IMT measurements in the 982 selected individuals with proven successful imaging (their Table 1), or should the denominator not be larger and include these 86 enrollees with unsuccessful IMT imaging? Although it is likely that a high body mass index was also a cause for these “failures,” it is also possible that some other risk factor or factors were associated with poor image quality in this subset of individuals. Inclusion of these 86 individuals would have increased the denominator of enrollees with ultrasound imaging from 982 to ≥1,068. This adds between 8% and 9% to the number of potential METEOR enrollees with incomplete IMT data before randomization. This number would likely have decreased the investigators’ reported overall completeness rates. On the basis of these numbers, we estimate that inclusion of these 86 individuals in the investigators’ Table 4 would have decrease the IMT completeness rate from 84% to somewhere between 73% and 77%. Furthermore, the data presented do not indicate to the reader how often individuals with reliable IMT ultrasound studies were excluded from the cohort simply on the basis of too high (≥3.5 mm) or too low (≤1.2 mm) IMT values. We reviewed this report and previous METEOR publications and were unable to find the reasons for selecting METEOR participants on the basis of IMT values between the 1.2-mm and 3.5-mm IMT cut points. Could different selection cut points have affected the IMT completeness rates? Would the rates have been different if all comers had been included? Would that have affected the ability to detect differences between the treatment and placebo arms? These questions are currently unanswered.
What is clear, however, is that the exclusion of individuals during screening examinations could have modified estimates of IMT data completeness and the association between completeness and risk factors.
The Precision of Estimates of Intima-Media Thickness Progression
The “one-view” carotid IMT imaging protocols adopted by the Atorvastatin Versus Simvastatin on Atherosclerosis Progression and Arterial Biology for the Investigation of the Treatment Effects of Reducing Cholesterol trials have shown the ability to detect significant differences in IMT progression between treated and control arms.
Peters et al. present their data in support of the hypothesis that a more complex acquisition and analysis protocol will improve the precision of IMT measurements. Is this the case? The data used to reach this conclusion are found in their Figure 3 and more importantly in their Table 5. Because precision is a decrease in variance, one would expect that the variance of IMT measurements should be lower for a five-angle protocol (the current protocol) compared with a much simpler one-angle protocol. The investigators’ Table 5 shows that the variances for measurements with a one-angle protocol are comparable, in fact less than or equal to those achieved with the five-angle protocol. Looking at the proposed IMT outcome variable, shown as the last rows of the top and bottom subtables, the standard error for the one-view protocol is 0.0032 mm/y (variance, 0.00001024 mm 2 /y 2 ), qualitatively similar to the standard error for the five-angle protocol at 0.0036 mm/y (variance, 0.0000130 mm 2 /y 2 ). The precision of the one-angle protocol is not significantly different from that of the five-angle protocol.
The magnitude of IMT rate-of-change in the treatment arm for the one-angle protocol is −0.0026 ± 0.0013 mm/y, similar to −0.0014 ± 0.0014 mm/y for the five-angle protocol. In fact, the observed decrease in IMT is subjectively greater for the one-angle protocol than for the five-angle protocol, a finding that would favor the hypothesis of a greater treatment effect. Looking at the placebo arm, the one-angle protocol gives a progression rate of 0.0147 ± 0.0021 mm/y, compared with 0.0131 ± 0.0022 mm/y for the five-angle protocol. One can argue that the IMT results generated by these two protocols are not statistically different. Subjectively, assuming similar bias in estimation, the observed IMT rate-of-change data might favor a one-angle protocol because of a greater IMT treatment effect and slightly more rapid IMT progression in controls.
The Precision of Estimates of Intima-Media Thickness Progression
The “one-view” carotid IMT imaging protocols adopted by the Atorvastatin Versus Simvastatin on Atherosclerosis Progression and Arterial Biology for the Investigation of the Treatment Effects of Reducing Cholesterol trials have shown the ability to detect significant differences in IMT progression between treated and control arms.
Peters et al. present their data in support of the hypothesis that a more complex acquisition and analysis protocol will improve the precision of IMT measurements. Is this the case? The data used to reach this conclusion are found in their Figure 3 and more importantly in their Table 5. Because precision is a decrease in variance, one would expect that the variance of IMT measurements should be lower for a five-angle protocol (the current protocol) compared with a much simpler one-angle protocol. The investigators’ Table 5 shows that the variances for measurements with a one-angle protocol are comparable, in fact less than or equal to those achieved with the five-angle protocol. Looking at the proposed IMT outcome variable, shown as the last rows of the top and bottom subtables, the standard error for the one-view protocol is 0.0032 mm/y (variance, 0.00001024 mm 2 /y 2 ), qualitatively similar to the standard error for the five-angle protocol at 0.0036 mm/y (variance, 0.0000130 mm 2 /y 2 ). The precision of the one-angle protocol is not significantly different from that of the five-angle protocol.
The magnitude of IMT rate-of-change in the treatment arm for the one-angle protocol is −0.0026 ± 0.0013 mm/y, similar to −0.0014 ± 0.0014 mm/y for the five-angle protocol. In fact, the observed decrease in IMT is subjectively greater for the one-angle protocol than for the five-angle protocol, a finding that would favor the hypothesis of a greater treatment effect. Looking at the placebo arm, the one-angle protocol gives a progression rate of 0.0147 ± 0.0021 mm/y, compared with 0.0131 ± 0.0022 mm/y for the five-angle protocol. One can argue that the IMT results generated by these two protocols are not statistically different. Subjectively, assuming similar bias in estimation, the observed IMT rate-of-change data might favor a one-angle protocol because of a greater IMT treatment effect and slightly more rapid IMT progression in controls.
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
