Logo: to the web site of Uppsala University

Postpartum depression and anxiety are pervasive and debilitating conditions, yet current risk identification is limited to self-report measures. This thesis examines whether noninvasive physiological measures collected during pregnancy improve prediction of postpartum depression (PPD) and anxiety beyond psychosocial screening. Across two prospective Uppsala cohorts (BASIC; 3PAD), heart rate variability (HRV), prepulse inhibition (PPI) of the acoustic startle response, and task-evoked pupil dilation (PD) were assessed in late pregnancy (BASIC) and at both early and late pregnancy (3PAD). Outcome measures were symptoms of depression and anxiety at six weeks postpartum. In Study I, elastic net models combining HRV with psychosocial variables discriminated PPD (AUC 0.93) and anxiety (AUC 0.83), but inclusion of HRV did not significantly improve accuracy beyond psychosocial predictors. In Study II, logistic regression analyses indicated that reduced PPI at 86 dB predicted de novo PPD among women without antenatal depression (AUC 0.81), with slight gains when covariates were added (AUC 0.91), suggesting potential utility for identifying cases missed by symptom-based screening. In Study III, random-forest models showed HRV from two gestational timepoints provided modest but reliable improvements in predicting PPD and state anxiety, with low frequency/high frequency ratio the most informative index. HRV indices added no incremental value for trait anxiety. In Study IV, elastic-net models including PD achieved good discrimination (AUCs 0.88-0.91) but did not outperform psychosocial predictors. Nevertheless, outcome-specific profiles were observed: PPD was linked to blunted early engagement and reduced sustained effort, state anxiety showed heightened early engagement followed by dampened responses across multiple indices, and trait anxiety involved opposing changes in mobilization speed across pregnancy. These patterns suggest that pupil-linked arousal dynamics may differentiate depression and anxiety outcomes, even if they add little predictive power. Overall, antenatal physiological measures were shown to complement psychosocial screening and provide mechanistic insight into perinatal mood vulnerability. While certain indices showed promise, predictive gains were modest and inconsistent, indicating that these measures are not yet ready for clinical application. Larger, externally validated studies are needed to establish their robustness and to determine whether they can be integrated into multiparametric, tiered risk algorithms to support prevention in perinatal care.