Materials and Methods
We conducted a multicenter, retrospective and prospective, descriptive, longitudinal study between 2016 and 2020.

Data collection

Our study focused on three periods:
Our database consists of information from two tertiary centers (the Saint Joseph Hospital in Marseille and the Groupe Hospitalier Sud Réunion de Saint Pierre [GHSR]) and one primary center (an ultrasound center in the Var). Images were retrieved from records via the following:
For each operator, we extracted, from the respective databases, images relating to the LVOT corresponding to 15 examinations in the second and third trimester, for all three periods. For each file that was randomly examined, there were either no images, or one or more images. For the latter, the best image for LVOT examination was selected.
The person responsible for retrieving these images referred to the operators as A (senior expert), B (senior experienced obstetrician), C (senior experienced obstetrician) and D (midwife) for the Saint Joseph hospital; A (senior expert), B (senior experienced obstetrician), C (young obstetrician) and D (midwife) for the GHSR; and A (senior experienced radiologist) and B (midwife) for the Var center. Three files (before 2016, in 2017, and in 2020) were thus created and contained data on the practitioners’ examinations.
For each practitioner, there were 30 files, numbered T2-1 to T2-15 for the 15 examinations in the second trimester and T3-1 to T3-15 for the 15 examinations in the third trimester for each of the three periods.

Data processing

Data were based on seven LVOT quality criteria proposed by Dr. E. Quarello (EQ) in 2019 (11), defined by visualization of the ANNEX 1 .
From the very start of our data collection, we asked all practitioners to carefully read the information and explanatory sheets on the proposed quality criteria (ANNEX 1 ). Three months later, we evaluated the possible implementation of these criteria in their daily practice through the study of images relating to examination of the LVOT from 15 examinations in the second trimester and 15 examinations in the third trimester, for these same sonographers during the first quarter of 2020.
Over the three periods, each image from the 10 practitioners’ examinations was rated for each of the seven quality criteria. The ratings chosen were 0 or 1 depending on the absence or presence of the quality criteria, respectively. The rating was carried out by a resident specialising in gynaecology-obstetrics with a focus on foetal medicine, Maud Regouin (MR). Thus, each image was scored out of seven points.
For all the images from each period, we attempted to identify the position of the back of the foetus according to the following classification (Figure 1 ):
To analyze inter-operator variability, 30 images were then arbitrarily selected from the image bank and an expert (EQ) was asked to grade each of them. These scores were compared to the scores from the trainee doctor (MR) for the same set of images. To analyze intra-operative variability of the junior doctor and expert (EQ), the 30 images were analyzed again after more than 24 hours by the same respective doctors and the scores were compared for the same set of images.

Data protection

All images were collected anonymously in each database. Only data that was strictly necessary and relevant to the research objectives was collected. All images were anonymised by the practitioners by removing first and last names. Access to this database was confidential and restricted to the direct users of the data.

Ethical and regulatory considerations

As this was non-interventional retrospective and prospective research involving non-human data from a study in the health field, a simplified MR004 procedure was granted with an agreement to comply with the National Commission for Information Technology and Civil Liberties (CNIL) (CNIL Reference 2219789 v 0) and registration of the project on the Register of studies without CNIL authorization. Patients were informed individually and in writing that ultrasound data from their records could be used. Each patient was given the possibility of declining the use of their data, by any means possible. In the event of no reply, compliance was assumed.