The effect of a Mentor Mothers program on prevention of vertical transmission of HIV outcomes in Zambézia Province, Mozambique: a retrospective interrupted time series analysis

Introduction: Mentor Mothers (MM) provide peer support to pregnant and postpartum women living with HIV (PPWH) and their infants with perinatal HIV exposure (IPE) throughout the cascade of prevention of vertical transmission (PVT) services. MM were implemented in Zambézia Province, Mozambique starting in August 2017. This evaluation aimed to determine the effect of MM on PVT outcomes. Methods: A retrospective interrupted time series analysis was done using routinely collected aggregate data from 85 public health facilities providing HIV services in nine districts of Zambézia. All PPWH (and their IPE) who initiated antiretroviral therapy (ART) from August 2016 through April 2019 were included. Outcomes included the proportion per month per district of: PPWH retained in care 12 months after ART initiation, PPWH with viral suppression and IPE with HIV DNA PCR test positivity by 9 months of age. The effect of MM on outcomes was assessed using logistic regression. Results: The odds of 12-month retention increased 1.5% per month in the pre-MM period, compared to a monthly increase of 7.6% with-MM (35–61% pre-MM, 56–72% with-MM; p < 0.001). The odds of being virally suppressed decreased by 0.9% per month in the pre-MM period, compared to a monthly increase of 3.9% with-MM (49–85% pre-MM, 59–80% with-MM; p < 0.001). The odds of DNA PCR positivity by 9 months of age decreased 8.9% per month in the pre-MM period, compared to a monthly decrease of 0.4% with-MM (0–14% pre-MM, 4–10% with-MM; p < 0.001). The odds of DNA PCR uptake (the proportion of IPE who received DNA PCR testing) by 9 months of age were significantly higher in the with-MM period compared to the pre-MM period (48–100% pre-MM, 87–100% with-MM; p < 0.001). Conclusions: MM services were associated with improved retention in PVT services and higher viral suppression rates among PPWH. While there was ongoing but diminishing improvement in DNA PCR positivity rates among IPE following MM implementation, this might be explained by increased uptake of HIV testing among high-risk IPE who were previously not getting tested. Additional efforts are needed to further optimize PVT outcomes, and MM should be one part of a comprehensive strategy to address this critical need. This was a retrospective interrupted time series analysis using routinely collected patient data from VUMC/FGH‐supported HF in Zambézia Province, Mozambique. During the study period (August 2016–April 2019), the province was comprised of 18 districts in which there were 230 public HF, and VUMC/FGH supported 112 of these. Each district‐level health system consists of one large central HF/referral centre and smaller peripheral HF. The VUMC/FGH MM program was implemented at HF providing maternal–child health services (including ANC), and the number of MM affiliated with each HF was proportionate to the volume of PPWH served at each HF (Table S1). Unpublished FGH programmatic data indicate that during the study period, 25–53% of pregnant women living with HIV were not receiving ART at time of presentation to ANC, and ≥97% of these women were started on ART under the prevailing ART initiation policies (initially Option B+, then Test and Start) [1, 2, 3, 4]. All PPWH and their IPE were eligible for inclusion if they: (1) enrolled in PVT services at one of 85 VUMC/FGH‐supported HF in nine districts (Table S1); (2) newly initiated ART in ANC during the current pregnancy; and (3) enrolled in care from August 2016 (1‐year pre‐MM implementation in August 2017) to April 2019 (end of evaluation period; however, the MM program continues at all sites). We excluded 27 HF that: (1) did not support maternal–child health services; (2) were supported by mothers2mothers® (M2M)—a similar but independent mentoring program for PPWH that had been implemented at a small proportion of VUMC/FGH‐supported HF (i.e. excluded to ensure comparability across intervention sites); or (3) were not supported by VUMC/FGH during the pre‐MM period (i.e. those HF in the district of Quelimane). An additional 15 HF were excluded from some analyses because of systematic missingness of outcomes data (i.e. non‐random missingness before or after a certain time point), as specified in Table S1. Our primary outcomes were retention of PPWH in PVT services, viral suppression among PPWH and HIV DNA PCR test positivity rates among IPE. Retention among PPWH was defined relative to time from ART initiation in ANC; we determined the proportion of PPWH per month who were still in care at 1, 3, 6 and 12 months after ART initiation (see the Supplementary Methods for individual‐level retention definitions). Each month, the number of PPWH who initiated ART 1, 3, 6 and 12 months prior and the number of PPWH who were still in care were recorded in the Open Medical Record System (OpenMRS)™ for each HF. The retention proportions for each district were calculated using the aggregated district‐level numbers. Viral suppression among PPWH was defined as a viral load (HIV RNA PCR) <1000 copies/ml. For this analysis, we determined the proportion of PPWH per month per district with viral suppression among all available viral load results for PPWH during the period of observation. Each month, the number of PPWH who had viral load testing and the number of PPWH who were virally suppressed were recorded in OpenMRS for each HF. Viral suppression proportions for each district were calculated using the aggregated district‐level numbers. HIV DNA PCR test positivity among IPE was determined for the periods 0–2 months and 0–9 months postpartum. This was defined as the proportion of positive DNA PCR results among all DNA PCR tests performed during the specified period. Each month, the number of DNA PCR tests for IPE within 0–2 months and 0–9 months and the corresponding number of positive DNA PCR results were recorded in District Health Information Software (DHIS) for each HF. The DNA PCR positivity rates for each district were calculated using the aggregated district‐level numbers. Other outcomes of interest included the uptake of HIV DNA PCR testing, uptake of ANC, institutional delivery and registration of IPE in CCR. Uptake of HIV DNA PCR testing among IPE was determined for the periods 0–2 months and 0–9 months postpartum. Each month, the number of DNA PCR tests for IPE within 0–2 months and 0–9 months were recorded in DHIS, but the exact number of IPE eligible for testing was unknown, so the number of pregnant women living with HIV who registered at ANC 6 months earlier was used as a proxy denominator. Descriptions of the methods and results pertaining to the uptake of ANC, institutional delivery and registration of IPE in CCR can be found in the Supplementary Materials (Supplementary Methods, Results and Tables S2–S4). OpenMRS and DHIS data from 85 HF in nine districts were included. Data were captured from August 2016 through April 2019. For each of the outcomes, we used aggregate data from each HF included in the evaluation. Routinely collected, de‐identified data were extracted from both databases for this retrospective interrupted time series analysis. To account for the phased implementation of MM services at various HFs (Table S1), outcomes were assessed at the HF level, accounting for respective implementation dates and looking at 12 months before and 12 months after MM implementation. Then, the pre/post outcomes were aggregated at the district level for comparison. Specifically, monthly HF‐level data were aggregated to district level by: (1) setting the MM start timepoint for each HF; (2) defining a new variable for each HF representing the MM implementation months (mm_month = the calendar year/month – MM start year/month); and (3) aggregating HF data within each district based on mm_month. Since all outcomes varied temporally (i.e. from −12 to +12 mm_month) and spatially (i.e. across nine districts), descriptive statistics were calculated within each district for pre‐MM and with‐MM periods separately, as well as for the entire study period. Each descriptive statistic was then compared across all nine districts. A proxy denominator was used to determine the uptake of HIV DNA PCR testing, which led to 17 (7.6%) of 225 proportions being >1. To address these invalid proportions, three approaches were employed for a sensitivity analysis: (1) cap all proportions >1 at a value of 1; (2) randomly replace each of them with a number between 0.9 and 1; and (3) exclude all invalid proportions from analysis. All three approaches yielded similar results, so only the results from the first approach (i.e. capping at 1) are reported. For each outcome, we assessed the effect of MM implementation via interrupted time series analysis using monthly district‐level aggregate data. Specifically, an indicator variable named mm was defined by assigning “no” for the pre‐MM period and “yes” otherwise, and a multivariable logistic regression model focusing on mm, mm_month and district was built to explore the effect of MM implementation adjusted by district. Sensitivity analyses were performed using logistic regression with a quasibinomial link function to account for potential overdispersion and generalized linear mixed‐effect models to account for clustering (see details in Supplementary Methods), and the results were essentially the same as those from the original fixed effect models (Tables S5 and S6). Statistical analyses were conducted using R statistical software 3.6.3 [5]. This data use and evaluation plan was approved by the VUMC Institutional Review Board (#201887), the Institutional Research Ethics Committee for Health of Zambézia Province (#16‐CIBS‐Z‐18) and was reviewed in accordance with the CDC human research protection procedures and was determined to be research, but CDC investigators did not interact with human subjects or have access to identifiable data or specimens for research purposes. Individual informed consent was not required for this evaluation since only routinely collected, de‐identified, aggregate data were used, and a waiver of informed consent was approved under the umbrella program evaluation protocol covering this analysis.