Background
Echocardiography is the imaging modality of choice for the evaluation of coronary artery (CA) abnormalities in Kawasaki disease. Small series have established high specificity and sensitivity for detecting abnormalities, yet visualization rates of individual CA segments and factors associated with success are unknown.
Methods
In the Pediatric Heart Network’s randomized trial of primary steroid treatment for Kawasaki disease, echocardiograms were interpreted locally and by a core laboratory. Univariate and multivariate predictors of CA visualization by the local lab as determined by the core lab were explored, and agreement of CA size measured locally and by the core lab was assessed.
Results
A total of 589 echocardiograms from 199 patients were obtained over 27 months. Visualization rates for the left main, proximal and distal left anterior descending, and proximal right CAs ranged from 91% to 98% but were lower for the distal right (65%), circumflex (86%), and posterior descending (54%) CAs. For the distal right and circumflex CAs, visualization rates improved over the course of the study ( P < .05). In multivariate analysis, local center, CA segment, and time from study start to echocardiography were independent predictors of visualization (all P values < .001). For segments for which visualization rates varied by center, higher percentage visualization was associated with larger center volume ( P = .001). Routine sedation use was also associated with higher visualization rates.
Conclusions
Successful CA visualization in Kawasaki disease is associated with the segment being evaluated and is influenced by center volume and sedation use. Increased visualization rates over time suggest a learning curve and underscore the value of core lab oversight in pediatric multicenter trials.
Kawasaki disease (KD) is a systemic vasculitis of unknown etiology that affects predominantly infants and young children. Inflammatory cells cause segmental destruction of the elastica interna of medium-sized, muscular extraparenchymal arteries, particularly the coronary arteries (CAs) in locations similar to those found in atherosclerotic CA disease. Almost all morbidity in KD after the first week of illness is related to CA aneurysms, which develop in up to 25% of untreated children and <5% of those appropriately treated.
Echocardiography is the mainstay of diagnosis and long-term surveillance of CA aneurysms in children with confirmed KD. Additionally, because there is no specific diagnostic test for KD and not all clinical features may be present, particularly in young infants, echocardiography has a key role in the evaluation of children with suspected or incomplete KD in the 2004 American Heart Association recommendations. The sensitivity and specificity of echocardiography for the detection of proximal CA dilation and aneurysms were high in single-center series compared with the gold standard of angiography. Despite the reliance on echocardiography in the management of KD, however, few studies have examined rates of visualization of CA segments, with or without aneurysms or dilation, across clinical centers or the patient characteristics and institutional factors that correlate with successful CA imaging.
We sought to describe the frequency of successful visualization of individual CA segments by both local and core laboratory assessments, to define potential factors associated with imaging success, to describe how that frequency changes over the course of a multicenter trial, and to compare visualization rates between local and core lab interpreters. Finally, we sought to compare measurements obtained by the individual sites with those obtained in the same examination by the core lab. These goals were pursued using the database of the Pediatric Heart Network (PHN) randomized trial of pulse steroids in primary treatment of KD.
Methods
Subjects
PHN investigators evaluated the efficacy of corticosteroids for the primary treatment of KD in a multicenter, randomized, double-blind, placebo-controlled trial. Subjects were recruited for the trial between December 2002 and December 2004 from eight clinical centers in North America (seven in the United States and one in Canada). Informed consent was obtained from each subject’s parent or guardian before participation. Patients meeting the criteria for acute KD were enrolled between days 4 and 10 of illness and randomly assigned to receive a single dose of intravenous methylprednisolone or placebo in addition to conventional therapy with intravenous immunoglobulin and aspirin. The primary trial outcome was maximum CA dimension Z score adjusted for body surface area (BSA) of either the proximal left anterior descending (LAD) segment or right CA (RCA). No difference between treatment groups was found. The study was conducted in accordance with the guidelines of the PHN’s Data and Safety Monitoring Board and of each center’s institutional review board.
Echocardiographic Methods
Echocardiograms were obtained at baseline and again at 7.8 ± 1.8 days (median, 8.0 days; allowable window, 3–14 days) and 36.5 ± 4.3 days (median, 36.0 days; allowable window, 21–49 days) after randomization. Sedation was given according to local practice. The cardiac ultrasound examinations were performed using transducers with the highest frequency possible.
Echocardiographers at the participating centers acquired the studies according to a uniform, predetermined protocol, which included display of seven CA segments. The left main CA (LMCA), left circumflex and posterior descending (PD) segments were treated as single segments. The LAD was divided into proximal and distal segments. The proximal LAD was defined as the portion from the bifurcation of the LMCA into the circumflex CA and LAD to the point at which the artery crossed the plane of the pulmonary valve. The distal LAD was defined as the segment beyond the plane of the pulmonary valve. The RCA was also divided into proximal and distal segments. The proximal RCA was defined as the portion of the artery between the ostium and the lateral border of the tricuspid valve. The distal RCA was measured at the mid segment of the atrioventricular groove in apical imaging planes. Online measurements were made of the internal luminal diameters of arterial segments using electronic calipers during image acquisition, and the largest measurement obtained from all views was recorded for the trial. Measurements were made from internal surface to internal surface and excluded points of branching, as these normally exceed adjacent vessel diameter. Figure 1 demonstrates the measurements made of each CA segment in a patient with significant dilation of all of CA segments.
Studies were recorded in a dynamic video format, either digitally or on standard 0.5-inch super VHS tape, and submitted to the core lab for interpretation by a single observer (S.D.C.), blinded to subject identification, treatment, and study visit. Video examinations were converted to digital imaging sets to allow calibrated measurements. Dimensions of the LMCA, proximal LAD, and proximal RCA were adjusted for BSA and expressed in standard deviation units ( Z scores) to assess their sizes compared with the normal population. The core lab provided training materials for the local centers for image standardization at study launch and periodic additional site-specific feedback on image quality.
Statistical Analyses
The primary outcomes for this analysis were the visualization rates of the seven CA segments. Candidate factors associated with successful visualization included time from study onset to echocardiography, artery segment, center, study visit, age at echocardiography, BSA, body mass index (BMI), use of sedation, and type of imaging media (digital vs videotape). Time was modeled as days from date of study launch to date of echocardiographic acquisition as a continuous variable, as well as in three 9-month time blocks, to determine if rates of visualization improved with time over the study period. Univariate and multivariate associations between visualization rate and potential predictors (including patient factors as well as source of reading, local vs central) were estimated using a generalized linear mixed model (binomial link), which incorporates a random effect for subject to account for multiple echocardiograms obtained from the same subject. If the estimate of the common variance of the random effect was zero, so that within-subject correlation could be ignored, we used simple logistic regression with a fixed effect only. P values < .05 were considered statistically significant. An intraclass correlation coefficient was used to estimate core lab versus local center agreement on CA dimensions. Confidence intervals for the intraclass correlation coefficients derived from longitudinal data were obtained by bootstrapping with 1,000 samples. The κ statistic was used to estimate core lab versus local center agreement with respect to whether a CA Z score was above normal ( Z > 2.0). All analyses were performed using SAS version 9 (SAS Institute, Inc., Cary NC).
Results
We enrolled 199 patients (62% male; mean age, 3.3 ± 2.2 years; mean BSA, 0.6 ± 0.2 m 2 ), in whom 589 echocardiograms were obtained. Two echocardiograms were assessed by the core lab to be unacceptable, that is, none of the primary measurements (LMCA, proximal LAD, or proximal RCA) could be made. Ten additional studies were obtained outside of protocol windows and were excluded from analysis of association between visualization and study visit portions of these results. One patient found to have a single RCA was excluded from analysis of visualization rate of the LMCA.
The overall rates of visualization of coronary segments as determined by core lab analysis are presented in Table 1 . The LMCA, proximal LAD, and proximal RCA each were visualized on almost all studies (≥98%). Visualization rates for the distal RCA, circumflex, and PD arteries were lower (65%, 86%, and 54%, respectively).
| Visualized (%) | ||||||
|---|---|---|---|---|---|---|
| CA segment | Overall ∗ | 0–9 months | 10–18 months | 19–27 months | Categorical P | Continuous P |
| n | 589 | 254 | 225 | 110 | ||
| LMCA | 583 (99%) | 98 | 99 | 100 | .80 | .18 |
| Proximal LAD | 579 (98%) | 98 | 99 | 99 | .55 | .15 |
| Distal LAD | 538 (91%) | 91 | 91 | 93 | .86 | .44 |
| Proximal RCA | 576 (98%) | 98 | 98 | 97 | .84 | .81 |
| Distal RCA | 385 (65%) | 59 | 72 | 67 | .08 | .03 |
| Circumflex | 509 (86%) | 82 | 89 | 91 | .12 | .03 |
| PD | 316 (54%) | 48 | 61 | 52 | .05 | .12 |
∗ Counts indicate number of echocardiograms, not number of subjects.
Univariate Analyses
We evaluated the rate of CA visualization by the core lab as a function of the time since study launch, in terms of both tertile of study period (categorical) and the number of days since study onset (continuous) variables ( Table 1 ). The distal RCA and circumflex CA were visualized more often in echocardiograms performed later in the study period. Visualization of the PD improved from the first 9-month block of the study to the second but did not reach significance in the continuous model.
Visualization rates as determined by the core lab stratified by center are shown in Table 2 . Rates for three of the most reliably identified segments (the LMCA, proximal LAD, and proximal RCA) were similar among the clinical sites. The four least commonly visualized distal coronary segments demonstrated significant variation among the eight sites, even when the least successful center was excluded.
| CA segment | Site 1 | Site 2 | Site 3 | Site 4 | Site 5 | Site 6 | Site 7 | Site 8 | P |
|---|---|---|---|---|---|---|---|---|---|
| n ∗ | 93 | 154 | 80 | 71 | 30 | 51 | 81 | 29 | |
| LMCA (%) | 100 | 98 | 100 | 97 | 100 | 100 | 100 | 97 | .99 |
| Proximal LAD (%) | 100 | 100 | 100 | 93 | 100 | 96 | 99 | 93 | .89 |
| Distal LAD (%) | 93 | 98 | 98 | 82 | 93 | 84 | 93 | 66 | <.001 |
| Proximal RCA (%) | 98 | 100 | 99 | 92 | 100 | 96 | 100 | 93 | .59 |
| Distal RCA (%) | 67 | 92 | 36 | 28 | 77 | 45 | 94 | 38 | <.001 |
| Circumflex (%) | 90 | 92 | 85 | 68 | 90 | 80 | 95 | 79 | .005 |
| PD (%) | 51 | 67 | 48 | 39 | 47 | 41 | 78 | 7 | <.001 |
∗ Counts indicate number of echocardiograms, not number of subjects.
In an effort to define center characteristics that would account for this difference in visualization rates, we examined whether patient sedation was associated with improved visualization. In all, 299 echocardiograms were obtained in patients aged <3 years, and 72% of these were obtained with sedation. The sedation strategy used most frequently was chloral hydrate (81% of sedated studies). The next most frequent medication, midazolam, was used in 16% of sedated exams, usually in combination with other medications (ketamine, nalbuphine hydrochloride, pentobarbital, and chloral hydrate). Overall sedation use in this subset of young patients varied significantly among the eight local centers, ranging from 6% to 100%. Visualization of the distal LAD, distal RCA, and PD was more often successful in echocardiograms obtained with sedation than in those without sedation ( Table 3 ). We also investigated whether time to randomization of first patient (with a longer time indicating a greater delay in implementing the imaging protocol after centralized training) or center volume affected visualization rates. The number of days from study onset to first patient randomization ranged from 10 to 42 (median, 16 days). Centers that randomized their first patient at or earlier than the median of 16 days were more likely to visualize the distal RCA and PD than centers randomizing later ( Table 4 ). Local center volume was categorized by total number of echocardiograms submitted for analysis: high was defined as >90 studies (two centers), medium as 70 to 90 studies (three centers), and low as <70 studies (three centers). For the distal LAD, distal RCA, and PD the odds of visualization at high-volume centers were three to five times higher than the odds at low-volume centers ( Table 4 ).
| Visualization rate | ||||
|---|---|---|---|---|
| CA segment | Sedation (%) | No Sedation (%) | OR (95% CI) for visualization, sedation vs no sedation | P |
| LMCA | 99 | 99 | 2.57 (0.16–42.40) | .51 |
| Proximal LAD | 99 | 95 | 5.39 (0.96–30.31) | .06 |
| Distal LAD | 94 | 82 | 3.76 (1.50–9.40) | .005 |
| Proximal RCA | 99 | 96 | 2.35 (0.41–13.56) | .34 |
| Distal RCA | 80 | 42 | 5.59 (2.74–11.41) | <.001 |
| Circumflex | 90 | 81 | 2.03 (0.83–4.96) | .12 |
| PD | 63 | 34 | 3.16 (1.58–6.33) | .001 |
| Time to first randomization following training: early vs late ∗ | Local center volume: high vs low † | |||
|---|---|---|---|---|
| CA segment | OR (95% CI) | P | OR (95% CI) | P |
| Distal LAD | 1.61 (9.83–3.14) | .16 | 5.07 (2.10–12.30) | .002 |
| Distal RCA | 2.30 (1.33–4.00) | .003 | 4.93 (2.41–10.10) | <.001 |
| Circumflex | 1.76 (0.96–3.22) | .07 | 2.18 (0.95–5.00) | .59 |
| PD | 1.67 (1.03–2.71) | .039 | 3.41 (1.79–6.48) | .001 |
∗ Early: time to first randomization following training ≤ 16 days; late: time to first randomization following training > 16 days.
† High center volume: >90 echocardiograms submitted for analysis; low center volume: <70 echocardiograms submitted for analysis.
The percentage of visualized CA segments did not vary according to whether the patient’s echocardiogram was performed at baseline, 1 week, or 5 weeks after randomization. Similarly, patient BSA, BMI, and age at echocardiography did not affect rates of visualization, except for lower visualization of the distal RCA in patients with higher BSAs ( P = .009) and with higher BMIs ( P = .002). Two of the eight sites submitted images in both videotape and Digital Imaging and Communications in Medicine (DICOM) formats; the other sites used only videotape. A total of 64 examinations (11%) were in DICOM format. Visualization of the distal RCA only was better for DICOM examinations than for those on videotape: 81% versus 63%, respectively (odds ratio, 2.63; 95% confidence interval, 1.06–6.53; P = .037).
Multivariate Model
Using a mixed logistic model, we explored independent factors associated with visualization. Starting with all stated candidate predictors (time from the onset of study enrollment to echocardiography, artery segment, local center, study visit, age at echocardiography, and BMI and BSA at echocardiography), variables were eliminated sequentially on the basis of the highest P values. The first model included 4,123 observations: seven segments in all 589 echocardiograms. Variability in visualization rates was most strongly and independently associated with local center and the particular CA segment being imaged, with sedation having a moderate association after accounting for local center and CA segment (odds ratio, 1.41; P = .047; Table 5 ). The second model included only those patients aged <3 years. A total of 2,093 observations were performed in seven CA segments in 299 echocardiograms. In this model, local center, CA segment, and sedation (odds ratio, 1.89; P = .024) all remained independently associated with visualization; increasing time from trial start to echocardiographic acquisition was also significant in this model ( Table 5 ).
