Data analysis

Differences in baseline characteristics among participants with established HIV compared to those newly diagnosed with HIV during the index pregnancy were described with proportions and tested using the χ² test. Multiple imputation was used to estimate relationship status for the 97/650 (14.9%) women in the cohort who had missing data.

We used a multivariable logistic regression to assess factors associated with LTFU during the first month after the first ANC visit. LTFU during the pregnancy and postpartum periods were modeled using multivariable parametric survival analyses, assuming an underlying generalized gamma distribution. Only those participants who were retained during the first month of ANC and had not delivered were included in the pregnancy survival analysis (484/650; Figure 1). Similarly, only those who were retained until delivery were included in the postpartum survival analysis (495/650; Figure 1). A composite event of LTFU and death was used. Women were considered retained in care and administratively censored from the analyses if a transfer was documented or if they stopped taking ART and continue with clinic visits. We used a generalized gamma distribution to derive relative times to event, rather than typical relative hazards produced using a Cox regression. This approach allowed for the modeling of non-proportional hazards (26). The modeling parameters produced from the primary analysis presented in Table 3 are given in Supplemental Table 2. In order to assess the stability of the results produced through multiple imputation, a sensitivity analysis was conducted using complete data only. All analyses were completed in STATA (27).

Factors associated with LTFU during first month of ANC care, pregnancy and the post-partum period

All three models presented above were adjusted for all of the variables listed in the table above.

LTFU - initial month of PMTCT: logistic regression.

LTFU - pregnancy: parametric survival analysis (supplemental Table 2).

LTFU - post-partum: parametric survival analysis (supplemental Table 2).