Test-retest reliability of new and conventional echocardiographic parameters of left ventricular systolic function

Abstract

Background: Reliability of left ventricular function measurements depends on actual biological conditions, repeated registrations and their analyses.

Objective: To investigate test-retest reliability of speckle-tracking-derived strain measurements and its determinants compared to the conventional parameters, such as ejection fraction (EF), LV volumes and mitral annular plane systolic excursion (MAPSE).

Methods: In 30 patients with a wide range of left ventricular function (mean EF 46.4 ± 16.4%, range 14-73%), standard echo views were acquired independently in a blinded fashion by two different echocardiographers in immediate sequence and analyzed off-line by two independent readers, creating 4 data sets per patient. Test-retest reliability of studied parameters was calculated using the smallest detectable change (SDC) and a total, inter-acquisition and inter-reader intra-class correlation coefficient (ICC).

Results: The smallest detectable change normalized to the mean absolute value of the measured parameter (SDCrel) was lowest for MAPSE (10.7%). SDCrel for EF was similar to GLS (14.2 and 14.7%, respectively), while SDCrel for CS was much higher (35.6%). The intra-class correlation coefficient was excellent (> 0.9) for all measures of the left ventricular function. Intra-patient inter-acquisition reliability (ICCacq) was significantly better than inter-reader reliability (ICCread) (0.984 vs. 0.950, p = 0.03) only for EF, while no significant difference was observed for any other LV function parameter. Mean intra-subject standard deviations were significantly correlated to the mean values for CS and LV volumes, but not for the other studied parameters.

Conclusions: In a test-retest setting, both with normal and impaired left ventricular function, the smallest relative detectable change of EF, GLS and MAPSE was similar (11-15%), but was much higher for CS (35%). Surprisingly, reliability of GLS was not superior to that of EF. Acquisition and reader to a similar extent influenced the reliability of measurements of all left ventricular function measures except for ejection fraction, where the reliability was more dependent on the reader than on the acquisition.

Keywords: Ejection fraction; Global longitudinal strain; Left ventricular function; Mitral annulus plane systolic excursion; Test–retest reliability.

Fig. 1

Study design. Standard echo sequences were acquired independently by two different, echocardiographers. Both image sets (acquisition 1 and 2) were then analyzed by two independent readers creating 4 data sets per patient (acquisition 1/reading 1, acquisition 1/reading 2, acquisition 2/reading 1 and acquisition 2/reading 2). Intra-class Correlation Coefficient (ICC) for single measures was calculated globally for the 4 image sets and then separately for two different acquisitions analyzed by the same reader (inter-acquisition reliability, ICC_acq) and two different readers analyzing the same acquisition (inter-reader reliability, ICC_read). For estimation of intra-subject variability, the standard deviation of the four measurements of each parameter in each patient was calculated. The mean of these intra-subject standard deviations resulted in the standard error of measurements (SEM) for the whole studied group. The smallest detectable change (SDC) is then calculated as 1.96 × SEM, representing the minimal difference between the measurements that must be overcome to ascertain a true change or difference with a less than 5% chance of error. The smallest relative detectable change (SDC_rel) is defined as the ratio of the SDC to the mean value of the measured parameter

Fig. 2

An example of GLS measurements performed in one of the study patients (2 acquisitions taken during the same examination, analyzed off-line by 2 different readers)

Fig. 3

Bland–Altman plots showing the mean intra-subject relative difference (expressed as percent of the mean within the subject) between acquisition 1 and 2 after averaging of reading 1 and 2 for each acquisition (acquisition-effect) and the mean difference between reading 1 and 2 after averaging of acquisition 1 and 2 for each reading (reading-effect) for GLS (a), CS (b), EF (c), EDV (d), ESV (e) and MAPSE (f) versus the mean value of the respective parameter within the patient. Colored lines indicate bias and limits of agreement (1.96 × SD)

Fig. 3

Bland–Altman plots showing the mean intra-subject relative difference (expressed as percent of the mean within the subject) between acquisition 1 and 2 after averaging of reading 1 and 2 for each acquisition (acquisition-effect) and the mean difference between reading 1 and 2 after averaging of acquisition 1 and 2 for each reading (reading-effect) for GLS (a), CS (b), EF (c), EDV (d), ESV (e) and MAPSE (f) versus the mean value of the respective parameter within the patient. Colored lines indicate bias and limits of agreement (1.96 × SD)

Fig. 4

Correlation between mean intra-subject standard deviations of LV function parameters and their absolute values, indicating that the calculated smallest detectable changes are independent (homoscedastic) of LV function impairment level when assessed by GLS, CS, EF and MAPSE. Intra-subject standard deviations significantly correlated with absolute values of LV volumes