An explainable artificial intelligence model for the prediction of childbirth development from psychological indicators of anxiety

Identifying and analyzing psychosocial factors associated with antepartum anxiety is a crucial challenge to determining the degree of influence these factors have on the development of the labor process. We designed a longitudinal study with 235 pregnant women from two different health areas where several self-report questionnaires were used to estimate antepartum anxiety. These questionnaires are STAI (State-Trait Anxiety Inventory), ISRA (Inventario de Situaciones y Respuestas de Ansiedad), ASI (Anxiety Sensitivity Index), and a custom Survey of factors of concern in childbirth. In total, we obtain 140 dependent variables to predict two target variables: type of delivery (Eutocic and Dystocic) and labor lasting. A machine learning model using a Multi-Layer Perceptron Classifier is trained for the first target variable, obtaining a global accuracy of .81 and a recall for class dystocic of .88. For the forecasting of the labor lasting we train a Gradient Boosting Regressor that achieves a mean average error of 2.03 hours. Next, we apply eXplainable Artificial Intelligence (XAI) techniques to analyze the most relevant variables for predicting delivery outcomes. For the prediction of the type of delivery, ISRA-S, ASI, and STAI-E are the input variables with the highest impact on the classification. Complementarily, the number of previous children, ASI, and ISRA-C are identified as the input variables with the highest impact on the labor lasting