Feasibility and reproducibility of speckle tracking echocardiography in routine assessment of the fetal heart in a low-risk population: a commentary letter

To the Editor,

We read with great interest the recent study by Sacco et al. [1] regarding the reproducibility of speckle-tracking echocardiography (STE) in routine fetal cardiac assessments performed by non-experts in STE, in an NHS clinical setting. The authors reported that intraobserver and interobserver variability ranged from poor to moderate, concluding that a high level of expertise is crucial for achieving reproducible results in clinical practice. We would like to acknowledge the authors for their valuable work, but we have several points to address regarding the methodology and interpretation of the results.

1. Image quality and exclusion criteria: The quality of image acquisition is crucial for accurate STE assessments [2], [3], [4]. Sacco et al. describe a protocol involving a 2D four-chamber apical view with clear visualization of valve insertions, endocardial borders, and apex. While this is essential for 2D STE analysis, other technical factors – such as frame rate, depth, zoom, and the absence of outflow tracts – also influence image quality [2]. Of the 450 scans, 158 (35 %) were excluded due to insufficient quality. Given that the four-chamber view is standard in fetal cardiac screening and that the scans were conducted by trained ultrasound professionals, this exclusion rate is notable. Although the operators were described as non-experts, this likely refers to their STE experience rather than their sonographic competence. A more refined imaging protocol and greater sonographic experience could have reduced exclusions. This relatively high exclusion rate may affect reproducibility and warrants further discussion.

2. Learning curve and number of Pilot images: De Vet et al. demonstrated a learning curve for STE assessments, recommending at least 100 measurements to achieve proficiency. They evaluated three trainees with varying fetal echocardiography experience – extensive (OB/GYN trainee specializing in maternal-fetal medicine), limited (OB/GYN trainee), and none (medical student) – by comparing their measurements to an expert’s. After 100 assessments, all showed moderate-to-good agreement, with the maternal-fetal medicine trainee achieving the highest consistency with expert measurements [5]. In Sacco et al.’s study, ultrasound scans were conducted in the main ultrasound department by physicians with advanced training in fetal sonography. The 2D STE analysis was performed by operators identified as an obstetric doctor with advanced training in fetal sonography and, an OB/GYN trainee specializing in maternal-fetal medicine. The operators analyzed only 20 images in a pilot phase to familiarize themselves with the STE software [1]. Given this limited exposure, the accuracy of the reproducibility findings may have been influenced by the operators’ familiarity with the technique. We suggest that a more extensive pilot phase might have contributed to improved reproducibility of the results.

3. Missing parameters in STE analysis: The researchers chose to assess a comprehensive set of cardiac parameters, including ejection fraction (EF), cardiac output (CO), stroke volume (SV), and fractional area change (FAC). While we agree that these parameters are important, we believe that global longitudinal strain (GLS) and global longitudinal strain rate (GLSR) should also have been included in the analysis. GLS and GLSR are considered the most reliable parameters for fetal heart assessments, particularly given the small size of the fetal heart [6]. Also, the reproducibility of GLS and GLSR is not (yet) fully understood. Comparing the results with other papers measuring the same parameters would have been easier if the same measurements were provided.

4. Reproducibility and selection of cardiac cycles: As the authors mention rightly, STE analysis requires a precise selection of cardiac cycles. In Sacco et al.’s study, operators independently analyzed the same three-second cine loop but had the freedom to select the cardiac cycle for evaluation. We fully support the authors’ decision to make the study as “real-world” as possible. However, for intraobserver reliability, the operators performed two analyses on each cine loop. Were the same three-second cine loops selected in both analyses and was the same cardiac cycle within the cine loop selected? Additionally, what was the time interval between the two evaluations? Variation between two evaluations can lead to differences in the outcomes [7].

5. Comparison with other studies and statistical methods: In the discussion, Sacco et al. compare their reproducibility with those from other studies [1]. However, we believe that reproducibility should be compared with studies employing the same methodology, specifically those that derived variables from the same cardiac cycle within a single cine loop. For instance, Crispi et al. assessed the inter- and intraobserver reliability of GLS [8]. Huntley et al. investigated the inter- and intraobserver reliability of both the global sphericity index and GLS [9] whereas Enzensberger et al. focused on GLS reliability [10]. In contrast, Sacco et al. evaluated the inter- and intraobserver reliability of FAC, EF, SV, CO [1]. Notably, the variables assessed in this study differ from those examined in the above-mentioned studies. It remains unclear whether the same methodology was applied to assess cardiac measurements, limiting direct comparability.

Furthermore, the authors suggest that variations in software packages, including algorithms and parameter settings, may contribute to discrepancies in results. While we agree with this, it does not explain the poor to moderate inter- and intraobserver reliability, given that the same ultrasound device and software were used in their study.

6. Gestational age and clustering of data: Sacco et al. included 393 ultrasound scans from 200 participants, implying multiple scans per participant at different time points during pregnancy. However, it is unclear whether the authors accounted for the longitudinal nature of the data in their analysis [1]. Additionally, it is known that GLS decreases with advancing gestational age [11]. Therefore, we recommend assessing reproducibility in separate gestational age categories, as this could have an impact on reproducibility scores and is essential for clinical implementation. Also, the current literature shows variations in reproducibility across different gestational ages. Additionally, reproducibility decreases toward the end of pregnancy, possibly due to suboptimal imaging conditions [12].

7. Statistical methodology: CCC vs. ICC: The authors used Lin’s Concordance Correlation Coefficient (CCC) to assess intra- and interobserver reliability. While CCC is widely used for comparing two measurement techniques, we believe the Intraclass Correlation Coefficient (ICC) would be more appropriate for this study, because of the following reasons: ICC is commonly used in studies involving multiple raters and repeated measures [13]. Since two independent researchers analyzed multiple measurements per patient with the same technique and software, ICC would better assess the consistency of measurements. Many of the studies cited in the discussion utilized ICC to assess reproducibility and none of them used the CCC [8], [9], [10]. The ICC and CCC outcomes are not directly comparable, as they assess different aspects of reliability. ICC evaluates the consistency of measurements [13] whereas CCC quantifies the degree of agreement between two methods or the presence of systematic bias between two measurements [14].

Conclusions: The reproducibility of STE during pregnancy, particularly when performed by non-experts, is an ongoing concern. Until today, it is not well documented, and Sacco et al.’s study is the first to address this subject. While we fully support the testing of STE analysis prior to clinical implementation, we believe that further standardization and a more detailed methodological approach are required to ensure consistency and reliability. Given the current findings, we feel it may be premature to advocate for its widespread use without further investigation and methodological refinement.

We greatly appreciate the authors’ contribution to this important area of research and look forward to their response.