Electroencephalographic signatures during pregnancy and their role on parenting-related measures: a scoping review protocol

Inclusion criteria

Participants

Considering the broad nature of the primary research question, any empirical study encompassing a group of pregnant women who completed EEG recordings at any stage of gestation will be included. No sociodemographic, obstetric, or clinical exclusion criteria will be employed. Non-human studies will be excluded.

Concept

Considering the defined research questions, three main concepts need to be closely addressed:

Pregnancy will be considered from the moment of reported conception until the end of gestation. Longitudinal studies contrasting assessments during pregnancy with measurements attained before conception and/or postpartum will also be included.

Only EEG-neural correlates related to pregnancy will be assessed. Several classification frameworks have been proposed for EEG signals. Here, we will use two different levels of classification to synthesise the literature depending on whether the focus is on spontaneous or stimulus-driven brain dynamics. For the first level, we will classify the studies based on the type of EEG recordings (Zhang et al., 2023). That is, (a) resting-state EEG (i.e., to study the brain’s intrinsic electrical activity when the subject is not engaged in any specific task), and (b) task-based EEG (i.e., to investigate brain responses during specific activities or after exposure to a given set of stimuli). When it comes to analysing these recordings, several EEG metrics can be employed depending on the research focus. Therefore, for the second level of classification, we will divide the EEG metrics depending on the analysis: (a) Power spectrum analyses, which allow for studying the frequency components of a signal and understanding how power is distributed across different frequencies (Dressler et al., 2004); (b) Time-frequency analyses that enable to measure the temporal dynamic changes of the neural activity across frequencies (Morales & Bowers, 2022); (c) Event-related potentials that allow for analysing the neural correlates of mental processes by time-locking amplitude changes in EEG to specific task events (Sur & Sinha, 2009); (d) Connectivity analyses, which give the means to assess correlations in electrical activity recorded at different electrode sites. These analyses help to understand how different brain regions communicate with each other and are putatively organised in specific neural networks (Zhang et al., 2023); (e) Source localization analyses which are used to estimate/infer the brain location where the electrical signals are generated. This second classification is essential for categorising the current state of evidence based on the neural phenomena of interest (Stancin et al., 2021).

A broad conceptualization of parenting-related measures will be employed for the purpose of this review. After screening for recent systematic reviews on the subject, parenting-related measures will be defined as any indicator that captures beliefs, attitudes, expectations, and behaviors related to parenting and/or transition to motherhood. This will include parental bonding (also known as parental attachment, parent-fetal bond, or parent-child bond), mother-infant interaction (infant- and adult-initiated interactive behaviours as well as dyadic interaction measures), parenting style, positive and/or negative parenting measures, parental mentalizing or reflective functioning, parental self-efficacy, parental adaptation, parenting competence, parenting self-esteem, parental involvement, parental mastery, motherhood expectations, or other analogous measures (Best et al., 2023; Georg et al., 2023; Savage et al., 2019; Tesson et al., 2021; Beeck et al., 2023; Wittkowski et al., 2020).

Parenting-related measures can be attained through self-report by the pregnant participant, informant-report (e.g., partner, other family members, healthcare professional) and/or observational-based coding tools. Importantly, as some of these variables are more likely to occur before birth (e.g., motherhood expectations), while others after birth (e.g., mother-infant interaction), specification regarding the assessment moment will be provided.

Context

To provide a comprehensive overview of EEG-based research in female pregnant participants, this review will include samples recruited from the community (general population) and/or from healthcare settings (e.g., hospital-based, primary healthcare, home care). Studies will not be limited to a specific geographical location or social context.

Types of sources

Both quantitative and mixed-method studies, though only the quantitative data will be considered from mixed-method studies. More specifically, the review will encompass observational studies including prospective or retrospective cohort studies, cross-sectional studies, and case-control studies (i.e., pregnant vs non-pregnant) as well as experimental and quasi-experimental studies such as randomised controlled trials and non-randomized controlled trials (i.e., pre and post treatment). Qualitative studies (e.g., non-quantitative EEG analyses), case reports, case series, opinion articles, and letters to the editor will be excluded. Grey literature (e.g., conference abstracts, Ph. D. and master’s thesis, preprints) will be considered if they meet the eligibility criteria. Finally, systematic reviews and meta-analyses directly associated with the topic will be included only for the full-text citation checking step (see search strategy section). Only records available in English, Portuguese, and/or Spanish will be deemed eligible for inclusion.

Search strategy

The search strategy will follow the three-step method, as recommended by the JBI (Aromataris et al., 2014):

As a final step, we will contact five experts in the field of pregnancy, identified through Expertscape.com by email. We will request their input regarding the inclusion of any potentially relevant study that may have been overlooked.

Study/Source of evidence selection

We will use asreview lab (https://asreview.nl/), an open-source machine learning-aided pipeline applied to systematic reviews (Boetje & van de Schoot 2024). The process is an active learning cycle that trains the model while records are classified as relevant or irrelevant. As records get labelled, the learning algorithm improves leading to reordering the screening queue where we prioritise those most likely to be relevant. Ultimately, this method significantly reduces the time required for screening large datasets, improves transparency in the decision-making process, and minimises the risk of bias, making it a reliable and robust review process (Khalil et al., 2022; Roth and Wermer-Colan, 2023).

Following the search strategy implementation, duplicates will be removed using EndNote and the remaining references will be imported into asreview lab. We will apply the Oracle mode with default settings to screen the database (Ferdinands et al., 2020; Van De Schoot et al., 2021). This default mode utilizes TF-IDF for feature extraction technique, a naive Bayes classifier, a dynamic resampling strategy, and certainty-based sampling for querying relevant records (Van De Schoot et al., 2021).

This algorithm will be employed after an expert researcher (CC) inputs 5 relevant and 5 irrelevant articles. The 5 eligible articles will ideally cover the different EEG metrics and approaches (time-frequence, resting-state, etc), while the 5 non-eligible studies will address different exclusion criteria (case reports, animal studies, etc).

The number of screened articles will be defined by the knee method (van Haastrecht., 2022). This method calculates the ratio between the slope before and after a critical inflection point in the gain curve, represented by the percentage of screened abstracts (x-axis) and the percentage of identified expected relevant abstracts (y-axis). As a stopping rule for the screening process, a slope ratio > 6 will be applied (Cormack & Grossman, 2016). Since this method is not directly implemented in asreview lab, the knee calculation will be computed at every 5% of screening citations.

The relevant/irrelevant classification in asreview lab will be based on a standardized flowchart for eligibility criteria (see Appendix 3). First, titles and abstracts will be screened, considering the following criteria: (a) Language: English, Portuguese, or Spanish; (b) Type of article: exclude if it is a case report/case series, opinion, or letter; (c) Population of interest: if it includes a group of pregnant participants; and (d) Outcome of interest: if it includes an EEG assessment of the pregnant group. After the screening of all records and disagreements resolved, potentially relevant sources will be retrieved from asreview lab so we can manually assess the full text in detail. At this point, any studies identified from the reference lists of the systematic review and meta-analyses that meet the eligibility criteria will also be included. Articles will exceed the full text if: (a) includes a quantitative EGG metric; and (b) the full text is available. Both title/abstract and full-text screening will be conducted by two independent reviewers (SA and PB) with any disagreements resolved through discussion or by a third reviewer (TOP). To avoid the risk of discarding eligible articles, the reviewers will follow a more liberal approach during the tile/abstract screening. That is, only studies that do not meet any criterion described below will be excluded; otherwise, they will be screened in the full-text step where a final decision will be made.

As recommended by the JBI guidelines (Peters et al., 2020), a pilot test will be conducted before starting each stage of the screening process to assess the flowchart implementation. The pilot will proceed as follows:

Two reviewers (SA and CC) will extract the information from the eligible records using a data extraction matrix on Microsoft Excel (see Appendix 4). The data extraction tables will be developed based on the items detailed below. Initially, two reviewers will pilot the current data extraction matrices using a subset of five eligible studies covering different types of EEG analysis. This approach aims to better assess the effectiveness of the data extraction matrices. Any discrepancies will be resolved by consensus or by referring to a third reviewer (TOP) and any required modifications to the data extraction matrix will be performed. After this, the reviewers will proceed with data extraction and any additional modifications of the data extraction matrix will be reported in the final version of the scoping review. Unreported information will be asked by email to the authors of the record (each set of authors will be contacted three times, after which the requested data will be coded and reported as missing). The final version of this Excel spreadsheet will be made available on OSF.

Data items

To better organise the data extraction process, multiple charting tables will be developed, each containing specific information while sharing a common identifier (first author and year of publication):

¹ The above classification assumes that each analysis type is independent; however, we are aware that connectivity and source localization analyses can be integrated within ERP, frequency, and frequency-time domain analyses. Therefore, if studies incorporate combined analyses, we will introduce new categories within the classification to account for them.

Data analysis and presentation

The search results and the eligibility process will be reported in the scoping review and presented in a PRISMA flow diagram (Tricco et al., 2018). The extracted data will be displayed in the form of tables and graphs according to the RQs. We will address each RQ in a synthesis paragraph using the theoretical framework proposed in the Concept Section pending possible post-hoc decisions that are likely to come from the specificities of included studies.

A first narrative summary will be conducted to provide an overview of the evidence (RQ 1), focusing on the number of studies, year of publication, and country of the study. Along with this description, we will include a line chart to represent the growing interest in the topic over time and a world map to visualise the geographical distribution of the research samples. This will allow us to underscore the importance of a transcultural approach in this field of research, by identifying whether there is an overrepresentation of studies from specific regions (i.e., western countries) and regions that may be underrepresented (i.e., non-western countries and low-income countries).

To answer RQ 1.1, RQ 1.2, and RQ 1.3, which focus on mapping the three main types of study results, we will provide separate tables for each of them (Appendix 5). Each of these tables will include key sample characteristics, data collection methods specificities, EEG metrics, and study findings. Concerning the synthesis paragraphs, fRQ1 we will categorise and quantify the studies based on whether reported: (1) Resting-state vs task-based approach; and (2) whether the stimuli are: cognitive vs emotional; social vs. non-social; infant vs non-infant.

For RQ2, the summary will categorise the studies based on the five timepoints assessments: before conception, 1^st trimester, 2^nd trimester, 3^rd trimester of pregnancy, and postpartum period (six weeks after birth). We will report the percentage of studies that focused on each of these specific timepoints.

For RQ3 we will report the number of studies that used a self-report or an observational-based methodology to assess parenting-related measures, and whether the self-reports are answered by the pregnant participants or by others (e.g., partner, healthcare professionals).

Due to the large amount of information needed to answer RQ 1.4 and R. Q 1.5, only a quantitative summary for each topic will be included in the main text. Tables will be placed as supplementary material. More concretely, the summary answering RQ 1.4 will present the distribution of sample sizes across studies, the percentage of studies: (a) focusing on early or late stages of pregnancy; (b) reporting pregnancy complications, and (c) addressing maternal mental health. The Supplementary table (Appendix 6) will cover sociodemographic characteristics of the sample (e.g., age, educational level, number of children), obstetric information (e.g., weeks of pregnancy, any obstetric or fetal complication) and clinical information (e.g., present or past health medical diagnosis or treatment).

RQ 1.5 encompasses a wide range of relevant EEG methodological information. To address this, two distinct tables will be used, following the approach adopted in a previous scoping review (Hu et al., 2024). Accordingly, Tables 4 and 5 (Appendix 7) will include EEG recording characteristics and EEG preprocessing information respectively. In the quantitative summary, we will detail the technical aspects of EEG by quantifying the average of the duration of EEG recording. We will analyse the percentage of the different main preprocessing steps implemented, that is including filtering (high and/or low-pass filters), artifact rejection methods, and baseline correction procedures. The summary will also include the frequency of each of the data analysis approaches used.