Carotid near-occlusion is often overlooked when CT angiography is assessed in routine practice

Carotid near-occlusion is a variant of severe carotid stenosis. In contrast to conventional ≥ 50% carotid stenosis, near-occlusion has a reduced artery size (“collapse”) distal to the stenosis [1‐3]. The collapse is a physiological response to reduction in flow. When the flow reduction is severe, the distal artery has a threadlike appearance (near-occlusion with full collapse) (Fig. 1); with less severe flow reduction, the distal artery shows a “normal-appearing” but small distal artery (near-occlusion without full collapse) [1‐3] (Fig. 2). Near-occlusions need to be distinguished from conventional ≥ 50% stenosis because guidelines recommend medical therapy alone for symptomatic near-occlusion but revascularization for conventional ≥ 50% stenosis [4, 5]. Because near-occlusions show partially collapsed internal carotid arteries (ICA), NASCET prescribed that percent carotid stenosis not be calculated in cases of near-occlusion as this would produce fallacious stenosis assessments [1‐3, 6].

Near-occlusion can be shown by computed tomography angiography (CTA) [1, 2, 7, 8]. However, literature review suggests that many clinicians remain unaware of the existence of near-occlusion without full collapse as a lesser near-occlusion, thus recognizing only near-occlusion with full collapse [2]. A review of several cerebral angiography textbooks yielded no reference to near-occlusions with a “normal-appearing” albeit small distal artery [9‐13]. Similarly, no description drawing attention to the near-occlusion diagnosis as including more than just full near-occlusion was found [9‐13]. Hence, it is plausible that incomplete knowledge, lack of attention to detail, and inaccurate interpretation or synthesis of CTA findings might cause many near-occlusions to be missed for diagnosis in routine practice. A reasonable way to assess this is to compare routine practice grading of CTA with expert grading, but no study has performed such a comparison.

The purpose of this study was to assess the sensitivity and specificity of CTA for carotid near-occlusion diagnosis, interpreted in routine clinical practice compared with expert assessments.

We evaluated a consecutive sample of CTAs from all patients aged ≥ 18 years performed at or transferred to the Radiology Department at the University Hospital of Northern Sweden, Umeå, Sweden, between 2010 and 2014 (Fig. 3). CTAs were performed using several machines and protocols locally and at all referring sites, none adding delayed phase imaging. All 4403 exams (4042 patients) were re-evaluated by one observer (E.J.) and all cases of suspected near-occlusion or near-occlusion mimics were re-evaluated in a blinded fashion by a second observer (A.F.) with disagreements settled by consensus discussion. All evaluations were performed blinded to original reports; none of which was reported then by E.J. or A.F.. Conventional angiography was rarely used in the clinical setting; only CTAs were assessed in this study. The study was approved by the ethical review board in Umeå with need for informed consent waived due to the observational nature of the study.

In the routine practice grading analysis, 384 CTA exams from 379 consecutive patients were included, but 1 image report was missing, resulting in 383 exams with data available (baseline data presented in Table 1). There were three bilateral near-occlusions, with ICAs ranging 0.8–2.8 mm and clearly smaller than the ECA bilaterally.

Sensitivity of near-occlusion (including those with and without full collapsed) in the original report was 20% (95% CI 12–28%). Specificity of near-occlusion was 99% (Fig. 4). Detected near-occlusions had more prominent ICA asymmetry than near-occlusions mistaken for conventional stenosis; near-occlusions with full collapse were more often detected than near-occlusion without full collapse (Table 2). The vast majority (81%) of missed near-occlusions were called conventional stenosis. Near-occlusion sensitivity ranged 0–58% among the nine participating neuroradiologists (Fig. 5). Two neuroradiologists with > 50% sensitivity were the most experienced, one recently completed training. Among 29 general radiologists who reviewed 1–3 exams with near-occlusion (34 exams in total), three correctly diagnosed a single near-occlusion (Fig. 5).

Of the 104 near-occlusion cases, 21 (20%) were detected, 33 (32%) included no near-occlusion term or mention of distal ICA, and 15 (14%) were mistaken for occlusion; remaining 35 (34%) had some mention or indication of near-occlusion in the report (Table 3). Of 104 local near-occlusion cases, clinicians perceived the degree of stenosis to be near-occlusion in 18% (n = 19) and varied with diagnosis in the CTA report, but also by ultrasound findings (Table 3). Chance of clinicians perceiving the near-occlusion diagnosis tended to be higher among the 21 cases with near-occlusion as sole diagnosis (43%) than the 35 cases with mention or indication of near-occlusion (23%; p = 0.14, χ²).

In the national comparison, the 13 participants produced 182 imaging reports; 104 were near-occlusion cases. Near-occlusion sensitivity was 8% (95% CI 2–13%) and specificity was 100% (Fig. 6). Sensitivity ranged from 0 to 75% between observers. Near-occlusion with and without full collapse was similarly often detected (10% and 6%, respectively; p = 0.71, χ²).

Combining the local and national CTA comparison, 16% (8/51) of radiologists that assessed at least one near-occlusion case detected any near-occlusion, higher for neuroradiologists (36%) compared with general radiologists (8%).

Inter-rater reliability between the two expert reviewers for near-occlusion diagnosis on CTA was good: kappa 0.80; intra-rater reliability on 49 randomly selected cases was kappa 0.75 and 0.88 for each observer. The inter-rater reliability between the two readers of CTA reports was good (kappa 0.88, 95% CI 0.79–0.97).

Exploring diagnosing near-occlusion was based on fulfilling 4 of 4 Bartlett criteria, 93% sensitive (87/94) and 97% specific (257/264) compared with expert diagnosis. Of the 94 cases with 4 of 4 Bartlett criteria, local imaging reports diagnosed conventional stenosis or thrombosis in 64 (68%), near-occlusion in 18 (19%), and occlusion in 12 (13%). However, 25 exams (7%, 10 near-occlusions and 15 conventional ≥ 50% stenosis) were excluded from this assessment due to calcified stenosis (n = 11), contralateral occlusion (n = 12), or both (n = 2). All ten false negatives missed a single feature, all well explained by interpretive approach such as bilateral near-occlusion and stenosis proximal to bulb causing ECA collapse. Requiring ≥ 3 Bartlett criteria (including 3 of 3 in cases with a single missing feature) for near-occlusion diagnosis resulted in 100% sensitivity (102/102) and 87% specificity (244/279) compared with expert diagnosis.

The main finding of this study is that near-occlusion is often overlooked when CTA is assessed in routine practice. Near-occlusion sensitivity ranged between 0 and 75% between radiologists, with only 16% of radiologists finding any near-occlusion.

With a paucity of studies reporting grading accuracy of carotid near-occlusion, comparison with similar findings is difficult. Near-occlusion with full collapse is perceived to be the only type of near-occlusion by many [1, 2]. However, since 94% of near-occlusions in the major trial reports were near-occlusions without full collapse [3], it is reasonable to include near-occlusion without full collapse in the definition of near-occlusion. Mistaking near-occlusion without full collapse for conventional stenosis will result in calculating a percent stenosis ratio which is an underestimate of the true stenosis because the distal internal carotid artery lumen is decreased [6]. Given the omission of near-occlusions without full collapse in textbooks [9‐13] and much of the literature [1, 2], it is not surprising that radiologists under-recognize it. Indeed, a small distal artery was noted, but not correctly assessed, in many of the near-occlusion cases in this study. Therefore, increasing awareness of near-occlusion without collapse is a global educational priority. Although using 4 of 4 Bartlett criteria was reasonably sensitive (90%) and specific (98%) compared with interpretive approach, the interpretive approach is seemingly superior as causes for not meeting certain thresholds are assessable and cases with missing features can be assessed. Requiring ≥ 3 criteria led to a drop in specificity to 90%. Indeed, superiority of an interpretive approach was concluded in the Bartlett study [8] underscoring the need for an educational effort focused on the interpretive approach, but learners might initially benefit from the Bartlett criteria. A lack of awareness and/or lack of attention to detail leads to symptomatic near-occlusions misdiagnosed as conventional stenosis which, by current guidelines [4, 5], would result in unnecessary treatments. Recent non-randomized evidence has challenged these guidelines [2, 15‐22], why further studies and trials might prove superiority of other management strategies. Improved diagnostics may enhance future clinical studies and allow for implementation of findings.

The main analysis of near-occlusion sensitivity was based on original reports finding a 20% sensitivity. This finding was similar (19% sensitivity) when the more easily reproducible 4 of 4 Bartlett criteria approach was used. As each case was assessed by different observers in the original report analysis, it was impossible to separate case and observer dependency. The additional national sample analysis revealed substantial difference in sensitivity (0–75%) despite reviewing the same cases, indicating that variation in sensitivity was caused by difference in observer proficiency. However, near-occlusions with full collapse and smaller distal diameter were more often detected in the original analysis, so the chance of a near-occlusion to be correctly diagnosed was also dependent on case details. Detection of at least one near-occlusion was more common for neuroradiologists than general radiologists, but some neuroradiologists missed all near-occlusions, even those with full collapse. Also, the national sample revealed that the low sensitivity of near-occlusion was not limited to a single center.

Some reports included both a near-occlusion diagnosis and a percentage grade as if they were both correct at the same time. Near-occlusions should not be graded as percent stenosis as deriving percent stenosis is fallacious; these are mutually exclusive [1‐3, 22]. We only accepted a near-occlusion term or synonym if it was used as to describe severe stenosis associated with small distal artery, not when describing severe stenosis alone. As near-occlusion is a management altering diagnosis, only correct use of the term should reasonably be acceptable. Many reports mentioned small distal artery; some also associated this with the stenosis. Hence, in these cases, a combination of incorrect terminology and information synthesis caused the missed near-occlusion diagnosis. It is unclear how often the observer saw but did not report a small distal ICA among the cases without mention of small distal ICA; hence, for these cases, it is unclear how often failure to spot the small distal ICA failure to understand the small distal ICA caused the missed near-occlusion diagnosis.

Unfortunately, delayed imaging was not used by all, and conventional angiography was rarely used during the study period (though conventional angiography might not always have a delayed series to show a slowly filling diminutive distal cervical ICA). If used, delayed images might prevent some near-occlusions to be mistaken for occlusions in the original reports. There were possibly cases of near-occlusion only separable from occlusion with delayed imaging, mistaken for occlusion by both the experts and clinical observers. However, interpretation education needs priority as many mistook no visible contrast in severe stenosis lumen and clearly patent arteries beyond the stenosis as occlusion with retrograde filling. With CTA aiming at arterial phase, it is near impossible for contrast to reach from skull base down to bulb region in this short time, rather a technical limitation as partial volume effect is more reasonable to presume. However, the vast majority (81%) of missed near-occlusions were called conventional stenosis—not occlusion.

As the near-occlusion experts conducting the study did not participate in the clinical or diagnostic work during the time of the study, observation bias was reduced. Although we have sampled a consecutive cohort from a single hospital and added scattered observers from a single country, these findings may or may not be representative in other locales, depending on attention to detail of near-occlusions. However, we suspect these results are not specific to Swedish general radiologists or neuroradiologists. As the CTA findings varied with interpreter expertise, CTA findings are likely to vary greatly between centers.

In conclusion, near-occlusion seems to be systematically overlooked in clinical routine practice, presumably due to lack of attention to detail and limited application of documented grading criteria when assessing CTA.