The postpartum period is a time in which mothers are at an increased risk of experiencing mental health problems (Clapp et al., 2016; Langan Martin et al., 2016; San Martin Porter et al., 2023; Xu et al., 2014). Existing literature also suggests poor postpartum mental health is associated with negative short- and long-term effects for both mother and child (Chithiramohan and Eslick, 2023; Murray, 1992; Murray et al., 2010; Murray et al., 2011; Sanger et al., 2015; Stein et al., 2014). These women are more likely to experience poorer physical health and impaired postpartum maternal-infant bonding, in addition to perceiving impaired relationships with their partners (Da Costa et al., 2006; Lilja et al., 2012; Tietz et al., 2014). Their offspring have been observed to exhibit increased rates of behavioural difficulties in early childhood, have poorer academic performance and an increased risk of depression during adolescence (Murray, 1992; Murray et al., 2011; Naicker et al., 2012).

Mental health problems during the postpartum period may require inpatient care when severe. Postpartum psychiatric admissions (PPAs) are the third most common admission reason for inpatient admission among mothers in the six weeks following childbirth, with literature suggesting that psychiatric admission rates may remain elevated above pre-pregnancy levels for up to two years post-childbirth (Clapp et al., 2016; Langan Martin et al., 2016). The most frequent diagnoses among PPAs are psychotic disorders, followed by depression and bipolar disorder (Langan Martin et al., 2016). While joint admission of mother and baby is generally recommended during PPAs, it is not always feasible, resulting in concerns for negative clinical and parenting outcomes resulting from the separation of mother and newborn (Gillham and Wittkowski, 2015).

Given the negative impacts of poor postpartum maternal mental health and PPAs, identification of prenatal risk factors may be instrumental in informing prevention and intervention programs prior to delivery. Unsurprisingly, one of the strongest risk factors is mental health problems during pregnancy, which can be identified at a population level with appropriate screening tools such as the Edinburgh Postnatal Depression Scale (EPDS) (Ghaedrahmati et al., 2017). The EPDS is a self-reported and validated 10-item questionnaire commonly used to screen for symptoms of perinatal depression and anxiety (Cox et al., 1987). In Australia, clinical practice guidelines recommend its use at least once during both the antenatal and postnatal periods (Highet, 2023). While antenatal screening with the EPDS does not directly predict the risk of PPAs, it does provide valuable insights, which may aid in identifying mothers who may benefit from intervention programs to prevent the onset of postpartum mental health problems and their associated harm.

Despite the benefits of early intervention, predicting health outcomes, especially those that are rare or uncommon, is difficult. This is due in part to complex, and in some cases potentially unknown, interactions between a myriad of variables, and the potential for non-events to far exceed events, leading to issues such as sparse data bias which poses methodological challenges for traditional statistical techniques. Additionally, prediction is difficult due to the requirement of considerably large and representative samples for prediction models to have acceptable generalisability and accuracy (Greenland et al., 2016; Ngiam and Khor, 2019). For this reason, across many healthcare settings, there has been growing interest in applying machine learning methods to routinely collected population-based healthcare data to identify patients at risk of selected outcomes (Alowais et al., 2023). However, in the context of predicting postpartum mental health problems, such attempts have been limited and often focused exclusively on the diagnoses of postpartum depression as an outcome, constructing models based on relatively few risk factors or using non-population level data (Amit et al., 2021; Betts et al., 2020; Hochman et al., 2021; Matsumura et al., 2025; Qi et al., 2025; Wakefield and Frasch, 2023; Wang et al., 2019; Zhang et al., 2021). Moreover, none of the previous machine learning prediction models in the context of postpartum psychiatric admissions have included a universal measure of antenatal depression such as the Edinburgh Postnatal Depression Scale (EPDS), which may improve model performance.

This study will build upon our team's previous work in which Betts et al. (2020) used machine learning methods to predict postpartum psychiatric admissions among mothers with a prenatal mental health diagnosis. Data will include the full population of pregnant mothers, for which information on prenatal depression has been recently made available by the adoption of universal screening with the Edinburgh Postnatal Depression Scale (EPDS) during pregnancy. Thus, we aim to construct the first prediction model for postpartum psychiatric admissions applicable to the full population and, in doing so, investigate the importance of universal screening for depressive symptoms in pregnancy with the EPDS.