This thesis aims to investigate engagement in Option B+ (life-long antiretroviral therapy (ART) for all HIV-positive pregnant and breastfeeding women) among pregnant and postpartum women residing in Agincourt health and demographic surveillance system (HDSS) in north-east South Africa. Paper A describes outcomes for 1325 patients reported as lost to follow-up (LTFU) in routine clinic records showing that patients who were pregnant at ART initiation have differing outcomes including a lower risk of death. Paper B reports on the spatial distribution of 315 patient transfers between clinics and identifies risk factors for these transfers, showing that pregnant and postpartum women are more likely to transfer their treatment, more likely to not report their transfer to their origin clinic, and less likely to transfer to a clinic outside the study area, when compared to adults in the general population. Paper C evaluates the accuracy of mortality estimates provided through routine treatment records in TIER.Net (the national HIV database), by comparing them to data generated through the record review and patient tracing. The results suggest that TIER.Net underestimates mortality and overestimates programme attrition, with important consequences for treatment programme policy, planning, and monitoring and evaluation, and estimation of the UNAIDS 90-90-90 targets. Paper D draws on in-depth interview data with women living with HIV to explore the intersection between early infant diagnosis (EID) and maternal engagement in care, finding that continued engagement in care promotes utilisation of EID, but also that the EID result can undermine maternal engagement in care. Paper E documents the challenges faced by patient tracing programmes and the lessons learnt during the ascertainment of patient outcomes and gives recommendations for future practice. The collective findings of this PhD research provide estimates of HIV patient outcomes after LTFU from prevention of mother-to-child transmission of HIV (PMTCT) programmes and generate new insights into some of the underlying causes. This research demonstrates the feasibility of using multiple linked data sources to improve the ascertainment of these outcomes.