Thresholds for Arterial Wall Inflammation Quantified by 18F-FDG PET Imaging

Objectives This study assessed 5 frequently applied arterial 18fluorodeoxyglucose (18F-FDG) uptake metrics in healthy control subjects, those with risk factors and patients with cardiovascular disease (CVD), to derive uptake thresholds in each subject group. Additionally, we tested the reproducibility of these measures and produced recommended sample sizes for interventional drug studies. Background 18F-FDG positron emission tomography (PET) can identify plaque inflammation as a surrogate endpoint for vascular interventional drug trials. However, an overview of 18F-FDG uptake metrics, threshold values, and reproducibility in healthy compared with diseased subjects is not available. Methods 18F-FDG PET/CT of the carotid arteries and ascending aorta was performed in 83 subjects (61 ± 8 years) comprising 3 groups: 25 healthy controls, 23 patients at increased CVD risk, and 35 patients with known CVD. We quantified 18F-FDG uptake across the whole artery, the most-diseased segment, and within all active segments over several pre-defined cutoffs. We report these data with and without background corrections. Finally, we determined measurement reproducibility and recommended sample sizes for future drug studies based on these results. Results All 18F-FDG uptake metrics were significantly different between healthy and diseased subjects for both the carotids and aorta. Thresholds of physiological 18F-FDG uptake were derived from healthy controls using the 90th percentile of their target to background ratio (TBR) value (TBRmax); whole artery TBRmax is 1.84 for the carotids and 2.68 in the aorta. These were exceeded by >52% of risk factor patients and >67% of CVD patients. Reproducibility was excellent in all study groups (intraclass correlation coefficient >0.95). Using carotid TBRmax as a primary endpoint resulted in sample size estimates approximately 20% lower than aorta. Conclusions We report thresholds for physiological 18F-FDG uptake in the arterial wall in healthy subjects, which are exceeded by the majority of CVD patients. This remains true, independent of readout vessel, signal quantification method, or the use of background correction. We also confirm the high reproducibility of 18F-FDG PET measures of inflammation. Nevertheless, because of overlap between subject categories and the relatively small population studied, these data have limited generalizability until substantiated in larger, prospective event-driven studies. (Vascular Inflammation in Patients at Risk for Atherosclerotic Disease; NTR5006)

A therosclerosis is a chronic, low-grade inflammatory disease of the arterial wall that causes myocardial infarction and stroke (1). Despite aggressive primary and secondary prevention strategies, long-term disability and death from cardiovascular disease (CVD) continue to increase (2). Arterial inflammation is strongly related to the risk of atherosclerotic plaque rupture. Quantification of inflammation may improve patient risk stratification and allow new drug therapies to be tested (1).
Noninvasive imaging, in particular with 18 F-fluordeoxyglucose ( 18 F-FDG) positron emission tomography (PET), has been used in this way (3,4). Arterial wall 18 F-FDG uptake mirrors inflammatory activity in atherosclerosis (5-7); inflammatory cells consume large amounts of glucose in comparison with other plaque cells. This results in 18 F-FDG accumulation. In addition, arterial 18 F-FDG uptake is higher in morphologically unstable plaques and predicts future vascular events (8-13). 18 F-FDG PET can assess the efficacy (or futility) of treatments designed to lower plaque inflammation (14)(15)(16)(17)(18)(19)(20)(21)(22)(23)(24)(25)(26)(27). As shown in Online Table 1, the number of vascular intervention trials using 18 F-FDG PET as a surrogate marker of inflammation is growing, with one-half being published in the past 2 years. Several of these studies enriched their study populations by excluding subjects with 18 F-FDG uptake below predefined thresholds. However, a consensus regarding the most appropriate thresholds is lacking (28)(29)(30)(31), primarily because healthy subjects, presumably without pathological arterial inflammation, have not been systematically imaged, and large-scale prospective outcome studies are awaited (32,33). Without these data, it is challenging to enroll patients with sufficient arterial inflammation to need therapy and to avoid randomizing those unlikely to respond.
In this study, we assessed 5 frequently applied arterial 18 F-FDG uptake metrics in 3 distinct groups: healthy control subjects, those with risk factors for CVD, and a group with established CVD. Considering 18 F-FDG uptake in the arterial wall of healthy control subjects as physiological, we determined the 90th percentile for arterial wall inflammatory activity using several commonly reported PET endpoints. Finally, we determined the reproducibility of published measures of 18 F-FDG uptake and derived optimal sample sizes for drug studies based on our results.      Table 2).

METHODS
ACTIVE SEGMENT APPROACH. We also examined the TBR of the most-diseased segment TBR (Online Table 2). In addition, an active segment analysis was performed using several pre-defined cutoffs. Using a Values are mean AE SD, % (n), or median [IQR]. *p value between all groups. †p value between patients at increased CVD risk and patients with known disease. ‡Agatston score.
Whereas the % active slices remained significantly different between groups, the TBR active slices did not (Online Table 3).
In contrast to the carotids, a much larger proportion of the subjects had active aortic walls. With a cutoff of $2.40, 88% of the healthy control subjects had at least 1 active slice; however, the TBR active slices and % active slices were not distinct between groups ( Table 2). With the active definition at $2.60 or $2.80, more than one-half of the healthy control subjects still had active segments (Online Table 3). For the highest cutoff, TBR active slices was significantly different between groups (p ¼ 0.015).
THRESHOLDS. The TBR thresholds based on the tolerance interval in healthy control subjects are listed in Table 3. Based on the 90th percentile of this interval, the threshold for SUV max was 1.85 for the carotids and 2.38 for the aorta. For TBR max , this threshold was set at 1.84 for the carotids and 2.68 for the aorta. Figure 2 illustrates both the SUV max and TBR max values per group, with corresponding thresholds (red dashed lines). For SUV max , 39% to 43% of those at increased CVD risk versus 66% of the CVD patients exceeded these thresholds. For TBR max , these numbers were in general larger; 52% to 57% of those at increased CVD risk and 67% to 74% of CVD patients. In Online Table 4, we also provide the thresholds using the 95th percentile values. SAMPLE SIZES. Based on the TBR max values in the present study, Figure 3 depicts the sample sizes required for an estimated drug effect; ranging from 5% to 20%, as has been observed in previous drug trials (Online Table 1). Carotid TBR as a primary endpoint requires approximately 20% fewer subjects compared with aorta TBR. Of note, sample sizes based on SUV max values necessitate approximately 20% to 45% more subjects compared with TBR max (Online Figure 1).
REPRODUCIBILITY. The intraobserver and interobserver and interscan agreement within 3 weeks was excellent for TBR max as indicated by: 1) ICC values of >0.95 with narrow 95% confidence intervals; and 2) the absence of fixed or proportional bias in the Bland-Altman plots (Online Figure 2). In addition, agreement for all 18 F-FDG metrics was also excellent in healthy control subjects (Online Tables 4 and 5).

DISCUSSION
In the present work, we tested 5 frequently applied approaches to quantify 18 F-FDG uptake in the arterial wall of healthy controls, patients at risk, and patients with known CVD. Whole artery SUV max was significantly different between groups, and 18 F-FDG venous blood background values were similar. As  Abbreviations as in Table 2.
van der Valk et al. As shown in Table 2, differences in background 18 F-FDG activity between groups exist but are not significantly different. Both background correction methods show smaller variations compared with SUV; in patients with established CVD, the carotid SUV SD is 0.37 versus 0.28 to 0.30 after background correction. Consequently, the sample size based on TBR as readout is smaller than SUV. In addition, in drug

A Controls
Scatterplots showing the gradual increase in SUV max and TBR max for the carotids (A and B) and aorta (C and D) in healthy control subjects, patients at CVD risk and patients with known CVD. The red dashed line represents the 90th percentile value in healthy control subjects.
CVD ¼ cardiovascular disease; TBR max ¼ maximum target to background ratio; other abbreviations as in Figure 1.  (39,40) and the incremental value in cardiovascular risk stratification (11). The present study was not designed to investigate the nature of 18 F-FDG vascular uptake, but nevertheless showed that SUVs and TBRs were consistently higher in the aorta compared with the carotids. This is relevant when applying an "index vessel approach" to drug trials because, in w80% of subjects, the index vessel will originate from the aorta (31). This might be suboptimal, as we also demonstrated that aortic TBR as endpoint requires a larger sample size to detect drug efficacy (37). Taking into account that the published drug-induced TBR changes have been relatively small (ranging between 5% and 15%) (Online Table 1), the optimal choice of endpoint vessel is important.
The use of the carotid artery as a readout vessel holds the strongest biological validation linking the 18 F-FDG signal and inflammation to recommend it (5)(6)(7)13,37,38). Therefore, we suggest that if the index vessel approach is not used, the carotid artery is best validated as primary readout vessel, as highlighted by Gholami et al. (31).  Sample sizes required for studies using TBR max as the primary endpoint. These are dependent on the estimated drug effect (ranging between 5% and 20%) and target vessel for imaging (carotid artery or aorta). Abbreviation as in Figure 2.
van der Valk et al. intervention studies revealing small variations during a 3-to 6-month timeframe (19,23 (34,48), substantial variation in patient preparation (e.g., glucose levels, time of fasting), PET image protocol (e.g., time and areas of scanning) and technology (e.g., acquisition, reconstruction), and measurement parameters exists and harmonization is warranted (28)(29)(30)(31). As such, extrapolation of our thresholds is limited to studies using similar imaging and analysis protocols. Third, with respect to the population-based approach with a relative small group size, it must be stressed that clinical characteristics of the studied groups in this study (among others, age, sex, lipid levels) should be taken into account upon extrapolation of our thresholds. Finally, this study was not designed to associate 18 F-FDG uptake with additional structural or functional features of the artery because we used a non-contrast-enhanced CT as part of the PET/CT. Future studies using magnetic resonance imaging should improve such assessments as well as correct for partial volume effects, which is a well-described limitation of PET imaging (31).