Scientific Abstract
Background: Maternal mental disorders are a leading complication of childbirth and contribute to maternal death. Untreated postpartum psychopathology can harm both maternal and child welfare. Some women experience a traumatic childbirth and develop post-traumatic stress disorder (PTSD) symptoms following delivery (CB-PTSD). No recommended protocol exists to identify women with CB-PTSD. Advancements in computational methods of free text can inform the diagnosis of psychiatric conditions. Whether the narratives of childbirth processed via Machine Learning (ML) can be useful for CB-PTSD screening is unknown. This study examined the utility of written narrative accounts of personal childbirth experience to identify women with CB-PTSD. We developed a model using Natural Language Processing (NLP) and ML algorithms to identify CB-PTSD via classification of birth narratives.
Methods: A total of 1,127 eligible postpartum women provided short written childbirth narrative accounts focusing on the most distressing aspects of their experience. They also completed a PTSD symptom screen to determine CB-PTSD. After exclusion criteria were applied, data from 995 participants were analyzed. An ML-based Sentence-Transformer NLP model represented narratives as vectors serving as inputs for our neural network ML model to identify women with CB-PTSD.
Results: The ML model derived from NLP of childbirth narratives achieved good performance: AUC 0.75, F1-score 0.76, sensitivity 0.8, and specificity 0.70. Moreover, women with CB-PTSD generated longer narratives and used more negative emotional expressions and death-related words in describing their childbirth experience than those with no CB-PTSD.
Conclusions: This study provides proof of concept that personal childbirth narrative accounts generated in the early postpartum period and analyzed via advanced computational methods can detect women likely to endorse CB-PTSD. This suggests that birth narratives are promising to serve as the basis of a tool for informing low-cost, non-invasive maternal mental health screening, and more research that utilizes ML to predict early signs of maternal psychiatric morbidity is warranted.
Search posters
SoundCloud Transcript
Childbirth can be a traumatic experience, and a subset of the population fails to recover psychologically: an estimated 6% develop childbirth-associated post-traumatic stress disorder, or CB-PTSD. This translates to approximately 8 million women affected each year worldwide.
Effective means for screening for CB-PTSD are limited.
In this project, we examined whether brief narratives of childbirth could provide important information to forecast endorsement of CB-PTSD.
We collected written birth stories from 995 women that focused on the distressing aspects of childbirth. The average word count in the narratives was around 160 words. The PTSD Checklist for DSM-5 or PCL-5 was used to confirm probable CB-PTSD.
To analyze the words in the childbirth narrative, we represented sentences as dense, low-dimensional vectors termed ‘embeddings’, using a pre-trained Natural Language Processing model.
We developed a machine learning model that uses the output of the Natural Language Processing model to identify CB-PTSD via narrative classification. The developed machine learning model was trained to classify childbirth narratives as markers of endorsement, or no endorsement, of CB-PTSD.
Our machine learning model revealed that as many as 80% of women with a positive CB-PTSD screen were accurately classified based on words in the childbirth narratives, indicating that our model has good Sensitivity.
Women with a negative CB-PTSD screen were also identified by the model with adequate accuracy, and our model’s Specificity was 70%.
Using personal childbirth narratives is a potentially quick, feasible, and inexpensive way to screen for symptoms of traumatic stress following childbirth. It may inform risk assessment for clinicians to improve the welfare of women following birth trauma.