Reproducibility of echocardiographic indices of left atrial size in dogs with subclinical myxomatous mitral valve disease

Abstract Background Reliability of echocardiographic measurements of left atrial (LA) size, an important marker of disease severity, has not been reported in dogs with myxomatous mitral valve disease (MMVD). Objectives To define and compare reliability of left atrial dimension/diameter (LAD), LAD indexed to aortic valve diameter (LAD/AoD), left atrium‐to‐aortic root ratio (LA/Ao), left atrial volume acquired from a right parasternal long‐axis (LAVRPLx), and left apical view (LAVLAP) in dogs with subclinical MMVD. Animals Nine dogs with subclinical MMVD. Methods Prospective reproducibility study. Dogs underwent 12 echocardiographic examinations by 2 operators on the mornings and afternoons of 3 nonconsecutive days within 1 week. Reliability (measurement variability) was quantified using coefficients of variation (CV) and 95% repeatability/reproducibility coefficients (95% RC). A mixed‐model analysis of variance (ANOVA) was used to determine if time of day, day, and operator were significant sources of variability for each index. Results Linear measurements (LAD, LAD/AoD, and LA/Ao) exhibited less within‐day, between‐day, and interoperator variability (CVs, 3.9%‐12.5%) than did volume estimate measurements (LAVRPLx and LAVLAP; CVs, 11.8%‐17.9%). Of the linear measurements, LA/Ao exhibited greater variability (CVs, 9.9%‐12.5%) compared to LAD and LAD/AoD (CVs, 3.9%‐4.9%). Operator was a significant (P = .005) source of variability for LA/Ao. Conclusions and Clinical Importance Compared to other linear measurements, LA/Ao was the least reproducible and most dependent on operator. The 95% RC for each LA size index are provided to help identify clinically relevant changes (beyond intraoperator or interoperator variability) during serial echocardiographic examinations of dogs with subclinical MMVD.


| INTRODUCTION
Myxomatous mitral valve disease (MMVD) is the most common cardiac disease and most frequent cause of heart failure in dogs. Progressive mitral valve regurgitation results in enlargement of the left atrium (LA) and ventricle. Assessment of LA size represents an especially important marker of disease severity. Echocardiographic indices of LA size have been shown to predict risk for heart failure, 1,2 guide monitoring and treatment during the subclinical period, [3][4][5][6] and impact prognosis in dogs with MMVD. 3,4,[7][8][9][10] Quantitative LA size assessment most commonly is performed using linear measurements derived from 2-dimensional echocardiography, 11,12 but volume estimates using 2-dimensional and 3-dimensional echocardiography also have been evaluated for clinical use. 7,[13][14][15][16][17][18][19][20][21][22] Linear measurements are simple and efficient, but are crude surrogates of chamber volume. Conversely, volume estimate measurements that utilize planimetry or 3-dimensional echocardiography are more representative of true chamber volume, but are less efficient and can be more challenging to measure. Thus, they might be less precise and reproducible. Additionally, 3-dimensional echocardiography technology currently is hindered by accessibility, transducer size limitations, and cost.
The most commonly reported measurement of LA size is the LAto-aortic root ratio (LA/Ao), acquired from a single right parasternal short-axis image. It provides a convenient body size-independent measurement of LA size and is typically the representative echocardiographic index of LA size reported in studies of dogs with MMVD. [4][5][6]8 However, potential limitations of this method include incorporating a pulmonary vein in the LA measurement, 23 lack of consistency of timing of the measurements within the cardiac cycle, 24,25 and defining the path of aortic root measurement relative to valve sinuses.
One alternative linear measurement of LA size is the left atrial diameter/dimension (LAD) acquired from a standard right parasternal long-axis 4-chamber image. It can be normalized to body size using an allometric equation 21,25,26 or indexed to the aortic valve diameter (AoD) acquired from a separate standard right parasternal long-axis image of the left ventricular outflow view. 21,27 Doing so avoids incorporating a pulmonary vein, avoids measuring a sinus of Valsalva, and standardizes timing of the measurements for LAD (just before opening of the mitral valve) and AoD (during early to midsystole between the hinge points of the maximally opened aortic valve cusps).
The ideal LA size measurement used for clinical practice should be accurate, precise, and reproducible. Studies assessing accuracy (ie, comparing to a recognized gold standard such as magnetic resonance imaging) are challenging and often absent. This absence underscores the importance of studies assessing precision and reproducibility, which are of great importance and have clinical relevance. 28,29 Delineating true change in indices of LA size caused by disease progression or regression rather than change related to measurement variation, physiological variability, or both is clinically valuable. Also, assessing the reproducibility of LA size measurement bears relevance to future clinical studies and trials that involve serial echocardiographic examinations, multiple sonographers, or both. For example, measurements with higher reproducibility (less measurement variation) provide greater statistical power to detect differences between groups, which decreases sample size and cost. 30 To our knowledge, reproducibility of several linear and volume estimate measurements of LA size have not been evaluated in dogs with clinically stable MMVD. Therefore, our objective was to evaluate and compare both intraoperator and interoperator reliability (ie, magnitude of error and variability between measurements) of several linear and volume estimate measurements of LA size in dogs with subclinical MMVD.

| MATERIALS AND METHODS
All procedures in this study were approved by the Institutional Animal Care and Use Committee at the University of California, Davis (protocol #: 20438). All dog owners provided written, informed consent before enrollment.

| Animals
Dogs diagnosed with subclinical MMVD and having a complete echocardiographic examination by the clinical cardiology service at our hospital were prospectively recruited for the study. To be eligible for inclusion, dogs had to be ≥6 years of age, have a body weight ≤20 kg, have a characteristic left apical systolic murmur grade ≥3 of 6, and be free of clinical signs. Echocardiographic examinations were reviewed for the presence of mitral regurgitation using color Doppler imaging and valve thickening or irregularity, leaflet prolapse or both. Dogs were excluded if they were affected with any other cardiovascular disease, if they were receiving medications known to affect the cardiovascular system, if their temperament was not conducive for multiple echocardiographic examinations, or if they had previous or current radiographic and clinical evidence of heart failure.

| Experimental design
After thoracic radiographs to confirm dogs were not in left heart fail-

| Echocardiographic measurements
The value recorded for each measurement consisted of the average of 3, usually consecutive, cardiac cycles. No attempts to standardize heart rate or respiratory rate were made during image acquisition or measurements. For all cardiac chamber measurements, the bloodtissue interface (ie, inner edge-to-inner edge) measurement technique was utilized. To avoid precision bias, investigators did not specifically discuss how to perform the measurements. Instead, investigators performed all measurements as they would on clinical patients, using

| Statistical analysis
Statistical analyses were performed using computer software (MedCalc Statistical Software, MedCalc Software bvba, Ostend, Belgium and R statistical computing, R package version 0.84.1, Vienna, Austria). Reliability of the repeated echocardiographic studies was quantified using coefficients of variation (CV) and 95% repeatability/ reproducibility coefficients (95% RC), both of which utilize the withinsubject SD (wSD). The wSD was calculated as the square root of the within-subject variance (mean square error), which was determined by 1-way ANOVA with dogs as the grouping variable. When multiple ANOVAs were utilized, the within-subject variance was averaged to provide a representative value for calculations of CV and 95% RC. Coefficients of variation were determined using the wSD method and were calculated as: (wSD Ä overall mean) × 100. The 95% RC were calculated as 1.96 × √2 × wSD. 30 To determine if operator, time of day, or day were significant sources of variation for each LA size measurement, mixed-model ANOVAs were performed using dog as a random effect and operator, time of day, and day as fixed effects. Spearman rank correlation coefficients (r s ), intraclass correlation coefficients (ICC), and mean (SD) bias (using Bland-Altman's method) also were determined for interoperator reproducibility assessments. For ICC calculations, a 2-way single measures mixed effect model (all subjects are measured by the same observers) for absolute agreement was selected. 33 Statistical significance was set at P < .05. Within-day, between-day, and interoperator reliability are summarized in Table 1. All CVs were <20%, and except for between-day and interoperator reliability of LAV LAP , all CVs were <15%. The linear measurements LAD and LAD/AoD had the least variability and had CVs <5%. The CVs for LA/Ao ranged from 9.9% to 12.5%, with highest CV for interoperator variability. The volume estimate measurements of LA size (LAV RPLx and LAV LAP ) had the highest variability, with CVs ranging from 11.8% to 17.9%. The 95% RC that represent variation from within-day, between-day and interoperator variability also are presented in Table 1. In general, 95% RC that represent interoperator variability were slightly higher than intraoperator variability (within-day and between-day).
The mixed-model ANOVA showed that operator was a significant (P = .005) source of variability for LA/Ao. Significant sources of variability (time of day, day, or operator) were not identified for LAD, LAD/AoD, LAV RPLx , and LAV LAP (P ≥ .1). Scatter plots illustrating interobserver reproducibility assessments for all LA size indices are presented in Figure 1. Spearman (r s ) and ICC and mean (SD) bias for the interobserver reproducibility analysis of the LA size indices are presented in Table 2. Relatively strong correlations and agreement T A B L E 1 Reliability of left atrial size indices from repeated echocardiographic examinations of 9 dogs with subclinical myxomatous mitral valve disease

| DISCUSSION
Our results showed that linear measurements exhibited less vari- The terminology and statistical methods used to quantify measurement variation of echocardiographic measurements can be overwhelming, confusing, and even misleading. We reviewed current and relevant echocardiography literature to help clarify terminology and guide our interpretation and statistical methods. 28,29,34 Reproducibility is a broad term that represents the variation of the same measurement made on a subject under changing conditions such as different operators, locations, instrumentation, environments, or time frames. 29,30 Repeatability represents variation in repeated measurements made on the same subject under identical conditions. 29 It largely represents a test-retest procedure within a short time frame (eg, a few minutes), ideally under similar hemodynamic conditions. 29 Reliability represents the magnitude of error or variability between measurements. 29 Reproducibility, repeatability, and reliability can be assessed by many statistical tests that evaluate correlation, association, bias, agreement, and magnitude of error between measurements, and each statistical test has advantages and disadvantages and might be misleading depending on the type of assessment and data gathered.  To more comprehensively evaluate interoperator variability and avoid conformity bias, more operators from multiple, independent institutions should be involved. Therefore, our results do not necessarily represent all echocardiographers. This design feature is an unavoidable shortcoming of all reproducibility studies.
In conclusion, our study suggests linear measurements of LA size are more reproducible compared to volume estimate measurements.