The COVID-19 pandemic, beginning in early 2020, profoundly impacted Italy, causing widespread psychological distress. The ensuing lockdown measures and uncertainty surrounding the virus significantly disrupted healthcare services, including prenatal and postnatal care. Pregnant and postpartum women, already a vulnerable population, experienced heightened mental health risks due to the added stressors of the pandemic. Previous research suggested significant psychological distress, including depression, anxiety, and stress, among pregnant women during this period. However, a more comprehensive understanding of the emotional experiences and coping strategies of this population was needed. This study aimed to provide a deeper analysis of the emotional landscape of pregnant and postpartum women in Italy during the pandemic's initial months, building upon previous findings and leveraging AI-powered natural language processing to analyze a large dataset.

Existing literature documented the heightened mental health risks for pregnant and postpartum women during the COVID-19 pandemic. Studies showed increased rates of depression and anxiety, linked to factors such as fear of infection, isolation, and disruption of healthcare services. Previous research by the authors had provided preliminary insights into the emotional experiences of pregnant women in Italy during the early lockdown. However, the current study aimed to build upon this work through a larger-scale analysis employing advanced AI techniques to gain deeper insight into the emotional complexities and coping strategies adopted by pregnant and postpartum women during this period. The impact of misinformation and contradictory information from media sources was also reviewed, highlighting the importance of accurate and empathetic communication during a pandemic.

This study employed a mixed-methods approach using secondary analysis of the COVID-ASSESS dataset, a national cross-sectional survey administered in Italy during the first wave of the COVID-19 pandemic (March–May 2020). The survey included socio-demographic information, psychological assessments, and open-ended questions capturing women's emotional experiences. The dataset included 1774 women (1136 pregnant, 638 postpartum). AI-powered natural language processing (NLP) was utilized for analysis. GPT-3.5-turbo, a large language model, was employed for emotion classification (anger, anticipation, joy, trust, fear, surprise, sadness, disgust) using a zero-shot learning approach. GPT-4 was used for thematic analysis of open-ended responses, focusing on communication with healthcare professionals (HCPs), media, peers, and general reflections. The GPT-4 analysis involved tokenization, frequency analysis, semantic analysis, and scoring of themes, with textual citations provided as evidence. A thematic map was created to synthesize the relationships between themes. A sensitivity analysis compared GPT-3.5-turbo and GPT-4 performance, showing no significant difference in emotion classification accuracy.

The analysis revealed a significant shift in the emotional landscape of pregnant and postpartum women from pre-pandemic to pandemic periods. Before the pandemic, trust, anticipation, and joy were dominant emotions. During the pandemic, sadness and fear became most prevalent, with substantial decreases in trust and joy. Thematic analysis of communication with HCPs revealed five key themes: fear and anxiety (score 8/10), uncertainty and confusion (7/10), emotional support and reassurance (6/10), professionalism and competence (5/10), and distance and detachment (4/10). Analysis of media communication identified fear and anxiety (10/10), confusion and contradictions (8/10), sensationalism and alarmism (7/10), and misinformation and inaccuracy (6/10) as prominent themes. Communication with peers revealed themes of remote communication (8.5/10), adaptation and coping strategies (7.5/10), emotional support (7/10), information sharing (6/10), adaptation to the new normal (5.5/10), and fear and anxiety (5.5/10). Open-ended questions highlighted fear and anxiety (9/10), loneliness and isolation (8/10), gratitude and hope (6/10), and disappointment and sadness (5/10).

The findings confirm the significant impact of the COVID-19 pandemic on the mental health of pregnant and postpartum women. The shift from positive emotions pre-pandemic to dominant fear and anxiety during the pandemic underscores the need for targeted interventions and support. The themes emerging from the analysis of communication highlight the importance of clear, consistent, empathetic, and accurate information dissemination from HCPs and media. The reliance on remote communication and social support networks among peers suggests the importance of leveraging these resources to mitigate the negative impacts of isolation. The presence of both fear and hope reflects the complex emotional responses during a crisis. The study's findings align with previous research on the psychological effects of pandemics and the vulnerability of pregnant and postpartum women. The use of AI in this study enables large-scale analysis, increasing the generalizability of findings.

This study provides comprehensive insights into the emotional experiences of pregnant and postpartum women during the COVID-19 pandemic in Italy. The findings highlight the profound psychological impact of the pandemic, emphasizing the critical need for targeted mental health support and effective communication strategies. Future research could explore the long-term effects of the pandemic on perinatal mental health and evaluate the effectiveness of specific interventions aimed at improving communication and providing emotional support during similar crises. The integration of AI in qualitative data analysis offers a powerful tool for future research in this area.

While the use of AI enhanced the efficiency and scalability of the analysis, limitations exist. The zero-shot learning approach of GPT-3.5 may not have captured all the nuances of emotional expression as accurately as a fine-tuned model. The interpretability of AI-generated results can be challenging. The reliance on self-reported data from a voluntary sample may introduce biases. Future research could benefit from a more comprehensive evaluation of AI-based qualitative analysis methods and incorporating diverse methodologies to validate findings.