Methodology & Statistics
Prevention is Better Than Cure: Predicting the Onset of Postpartum Depression Using Early Warning Signals
It is estimated that 23,000 women suffer from postpartum depression (PPD) in the Netherlands yearly, making it the most common psychiatric disorder among new mothers. Luckily, research suggests that many mothers respond well to early intervention. In fact, intervening in time can even prevent an onset.
Heterogeneity in PPD calls for a personalized approach: Who should receive prevention, and when? Using state-of-the-art techniques, we will identify causal predictors for postpartum depression and use them to create a deep-learning algorithm for personalized prevention that uses early warning signals to predict an episode before it even occurs.
All of this is achieved in the total of four studies, where in study 1 data from the Brabant studie will be analyzed using Bayesian networks to identify important symptoms and risk factors. In study 2, to validate elicited predictors, we will conduct semi-structured interviews with mothers and experts to investigate recognition and if additional factors have to be considered. For the third study identified predictors will subsequently be used as input for a large experience sampling method (ESM) study where we will intensively monitor (pregnant) mothers. To intensively measure the severity of symptoms and (time-varying) risk factors (building upon study 1-2), 300 mothers will receive two 2-minutes questionnaires per day for 14 consecutive days once a month (starting 32 weeks prenatal until 6 months postnatal). Here, we will rely on the Network Approach to Psychopathology and construct individual networks of interacting PPD symptoms and possible causes for each mother separately. To identify early warning indicators (study 4), mothers will be clinically screened at 4, 12 and 20 weeks (postnatal) with the Mini International Neuropsychiatric Interview (MINI).
Earlier research has shown that statistical indicators (like rising variance) can be used as early warning signals for shifts in systems, like abrupt climate change or an epileptic episode. Here, we innovatively apply this concept to predict if a mother is moving from a mentally healthy to a more depressed state by using the obtained data (study 3) to train a deep-learning algorithm that can predict if a shift is to be expected based on early warning signals. In a second ESM study (200 mothers, similar set-up as study 3), this algorithm will be validated, and regular expert meetings will facilitate clinical interpretation. Importantly, the algorithm can be used in new samples without the need for new training or calibration.
This project is an IOPS project, since we will develop new psychometric methods in the ESM study and apply existing psychometric methods to new situations.
Prof. Dr. H.J.A. van Bakel
Dr. I. Schwabe
Dr. M.I. van den Heuvel
Dr. J. Jongerling
Herbert Simon Research Institute, Tilburg University