(1165) Using Graph Learning to Model Pathways of New Depressive Symptoms During Pregnancy

We conducted a secondary analysis from the Nulliparous Pregnancy Outcomes Study: Monitoring Mothers-To-Be, a prospective cohort of nulliparous pregnant individuals. We applied a nonparametric graph learning approach - PC-Algorithm using Kernel-based Conditional Independence (PC-KCI) - which allows for the modeling of causal pathways from observational data, correcting for multiple comparisons. PC-KCI was used to model risk of new-onset moderate-to-severe depression (Edinburgh Postnatal Depression Scale (EPDS) > 12 at visit 3) among participants with mild-to-no baseline depression risk (EPDS < 13) at visit 1.

Results:

354/9027 (3.9%) developed new-onset depression risk. Of 141 study-collected early pregnancy risk factors, 26 showed a direct link (p < 0.01) to new-onset depression risk. Figure 1 shows the graphical model output. Using these factors in a generalized additive model yielded an AUROC of 0.84. The strongest psychosocial predictors of depression onset were baseline anxiety and endorsement of specific self-reported “worry” items (e.g., “I am feeling bothered, upset, or worried about changes in my weight and body shape during pregnancy”). Certain acute physical symptoms (e.g., flu-like illness) and life events (e.g., experienced police discrimination) were directly connected to depression onset. The strongest protective factors were having a first-trimester discussion with a physician about the importance of mental health throughout pregnancy, having a planned pregnancy, and having a social support system.

Conclusion:

Using a nonparametric graph learning approach yields a simple, transparent model with good predictive value for new-onset depression in pregnancy. PC-KCI replicated well-known pathways, such as the relationship between anxiety and depression onset. PC-KCI also identified depression pathways, which include external and internal stressors, demographic characteristics, patient-provider interactions, and physical symptoms.