LIFE-COURSE COMORBIDITY PATTERNS AND INTEGRATED PREDICTION OF POSTPARTUM DEPRESSION, MULTIMORBIDITY, AND SYMPTOM PROGRESSION

Background Postpartum depression (PPD) is common and disabling, yet its longitudinal comorbidity patterns and predictability remain poorly understood. Most studies are limited by short-term designs and focus on isolated disorders rather than comprehensive multimorbidity spanning psychiatric, autoimmune, metabolic, and pain-related conditions across the whole life-course. Methods This study leveraged 8,804 women with delivery records in the All of Us cohort (438 with diagnosed PPD) to characterize multimorbidity trajectories and develop integrated prediction models. Comorbidities were grouped into 38 conditions across five clinical categories, examined cross-sectionally and in monthly time bins from 250 months before to 500 months after delivery. Latent class analysis (LCA) identified pre- and post-delivery multimorbidity profiles and transitions between classes. Machine learning models combining polygenic risk scores (PRS) for depression, obstetric complications, and socioeconomic factors were used to predict PPD, post-delivery comorbidity class membership, and symptom worsening among initially low-burden women. Results Six comorbidities were significantly associated with PPD after FDR correction (q < 0.05), including depression (OR=4.94), anxiety (OR=3.22), post-traumatic stress disorder (PTSD; OR=1.64), premenstrual dysphoric disorder (OR=3.31), celiac disease (OR=4.11), and polycystic ovary syndrome (OR=2.39). Temporal trajectory analyses revealed that 7 of 38 comorbidities showed significantly different prevalence between PPD cases and controls at specific time points, with psychiatric conditions (depression, anxiety and PTSD) clustering around delivery, chronic pain and fatigue emerging 5 months post-delivery (p=3.73 × 10-5 and p=1.22 × 10-4, respectively). LCA identified three distinct comorbidity classes pre- and post-delivery. Pre-delivery, PPD cases were significantly underrepresented in the minimal comorbidity class (55.9% vs. 75.8%, p=7.85 × 10-21) and overrepresented in moderate and high multimorbidity classes. Post-delivery class distributions did not differ by PPD status. Transition analyses showed that previously healthy women with PPD were significantly less likely to remain in the minimal class post-delivery (55.4% vs. 64.6%, p=0.004) and more likely to transition to moderate multimorbidity (37.9% vs. 28.7%, p=0.002). Machine learning models achieved moderate discrimination across all outcomes (AUROCs 0.70–0.73). Top predictors of PPD were age at delivery, depression PRS, and BMI, while socioeconomic disadvantage and obstetric complications drove comorbidity class prediction. Adding PPD diagnosis as a predictor improved symptom worsening prediction (ΔAUROC up to +0.046), but low positive predictive values limit clinical implementation. Conclusions PPD functions as a pivotal life-course event marking elevated risk for multimorbidity accumulation across psychiatric, cardiometabolic, and pain-related trajectories, particularly among previously healthy women. These findings support multimorbidity-informed, life-course screening strategies extending well beyond the traditional postpartum period, and highlight the complementary contributions of genetic liability, obstetric complications, and socioeconomic disadvantage to long-term women's health.