What will it take to eliminate pediatric HIV? Reaching WHO target rates of mother-to-child HIV transmission in zimbabwe: A model-based analysis

Background: The World Health Organization (WHO) has called for the “virtual elimination” of pediatric HIV: a mother-to-child HIV transmission (MTCT) risk of less than 5%. We investigated uptake of prevention of MTCT (PMTCT) services, infant feeding recommendations, and specific drug regimens necessary to achieve this goal in Zimbabwe. Methods and Findings: We used a computer model to simulate a cohort of HIV-infected, pregnant/breastfeeding women (mean age, 24 y; mean CD4, 451/μl; breastfeeding duration, 12 mo). Three PMTCT regimens were evaluated: (1) single-dose nevirapine (sdNVP), (2) WHO 2010 guidelines’ “Option A” (zidovudine in pregnancy, infant nevirapine throughout breastfeeding for women without advanced disease, lifelong combination antiretroviral therapy for women with advanced disease), and (3) WHO “Option B” (pregnancy/breastfeeding-limited combination antiretroviral drug regimens without advanced disease; lifelong antiretroviral therapy with advanced disease). We examined four levels of PMTCT uptake (proportion of pregnant women accessing and adhering to PMTCT services): reported rates in 2008 and 2009 (36% and 56%, respectively) and target goals in 2008 and 2009 (80% and 95%, respectively). The primary model outcome was MTCT risk at weaning. The 2008 sdNVP-based National PMTCT Program led to a projected 12-mo MTCT risk of 20.3%. Improved uptake in 2009 reduced projected risk to 18.0%. If sdNVP were replaced by more effective regimens, with 2009 (56%) uptake, estimated MTCT risk would be 14.4% (Option A) or 13.4% (Option B). Even with 95% uptake of Option A or B, projected transmission risks (6.1%-7.7%) would exceed the WHO goal of less than 5%. Only if the lowest published transmission risks were used for each drug regimen, or breastfeeding duration were shortened, would MTCT risks at 95% uptake fall below 5%. Conclusions: Implementation of the WHO PMTCT guidelines must be accompanied by efforts to improve access to PMTCT services, retain women in care, and support medication adherence throughout pregnancy and breastfeeding, to approach the “virtual elimination” of pediatric HIV in Zimbabwe. We expanded a validated computer simulation model of MTCT [28],[29] to include each step in the “cascade” of PMTCT-related care, from first presentation at ANC through 2 y postpartum (Figure A in Text S1). The MTCT model simulates a single pregnancy, delivery, and postpartum period for each mother–infant pair, and projects infant outcomes at birth. Additional clinical events for mothers and infants through 2 y after delivery are simulated using the CEPAC (Cost-Effectiveness of Preventing AIDS Complications) model (Figure B in Text S1) [30]–[32]. The primary outcome of the linked CEPAC and MTCT models is risk of infant HIV transmission at the time of weaning; secondary outcomes include HIV infection risk at 4–6 wk of age, 2-y pediatric survival, and 2-y HIV-free pediatric survival. This analysis examined improvements along two dimensions of PMTCT care: more effective PMTCT regimens, and improved uptake of the PMTCT “cascade” (Figure 1). We examined three possible PMTCT regimens in Zimbabwe: (1) the 2002–2009 National PMTCT Program, based on sdNVP alone, (2) WHO 2010 guidelines’ Option A, and (3) WHO 2010 guidelines’ Option B. Infant outcomes were projected at four levels of PMTCT uptake (36%, 56%, 90%, and 95% of women/infants receiving medications by the time of infant delivery) [19],[23]. Key sensitivity analyses examined the impacts of duration of breastfeeding, maternal HIV disease stage, the range of published MTCT risks for each PMTCT regimen, and a “full” (100%) uptake scenario. The linked models were used to simulate two populations of pregnant and breastfeeding women in Zimbabwe, with mean age of 24 y (standard deviation [SD]: 5 y) [33]. Cohort 1 included women already HIV-infected at their first ANC visit (regardless of whether HIV status was known to patients or providers). Cohort 2 was composed of all women becoming pregnant each year in Zimbabwe, an estimated 392,460 women [34],[35], with HIV prevalence of 16% at first ANC visit and HIV incidence of 1%/year during late pregnancy and breastfeeding [36]. Cohort 1 was thus nested within Cohort 2. Cohort 1 was analyzed to project MTCT rates, and Cohort 2 to project the proportion and number of infants in an annual birth cohort anticipated to become HIV-infected by 18 mo of age. All women were assumed to breastfeed their infants for 12 mo in the base case, based on WHO infant feeding guidelines [3]. The antenatal, intrapartum, and postpartum/neonatal components of each modeled PMTCT regimen are detailed in Table A in Text S1. In all modeled regimens, women identified as eligible for ART (CD4≤350/µl or WHO stage 3/4 disease) were referred for ART. The availability of ART after referral depended on the modeled PMTCT uptake scenario; the availability of CD4 assays was varied in sensitivity analyses. Following convention, we refer to combination antiretroviral therapy regimens as “ART” when intended for treatment of maternal HIV disease (also effective for PMTCT) and as “triple-drug ARV regimens” when offered to non-ART-eligible women for PMTCT. To create the four uptake scenarios, the “cascade” of PMTCT services was categorized into three broad domains (Table 1; Figure 1): “care and testing,” “drug availability,” and “retention.” We then calculated a “product of participation” (POP), defined as the product of the proportions of women/infants receiving care in each of these three domains, both at delivery and at the end of the breastfeeding period [23]. Four primary scenarios of PMTCT uptake were simulated for each PMTCT regimen (Table 1). The first two scenarios reflected WHO and Zimbabwean Ministry of Health and Child Welfare estimates of uptake at each step in the PMTCT cascade for Zimbabwe in 2008 (total POP at delivery: 36%) and 2009 (total POP at delivery: 56%) [19],[23],[36],[37]. The “WHO target” scenario modeled the current WHO goal that 80% of pregnant women be HIV-tested and 80% of HIV-infected women receive PMTCT services [20],[37]. The “optimal” uptake scenario simulated 95% uptake of the complete PMTCT cascade through delivery, likely representing the best practically achievable outcomes and lowest MTCT risks for the evaluated PMTCT regimens. A scenario of full, 100% uptake was also examined in sensitivity analyses (Tables D–G in Text S1), to represent the maximum potential biologic efficacy of each regimen. The MTCT model is a previously published, decision-analytic simulation of a cohort of pregnant women from the time of conception through delivery (TreeAgePro) [28],[29]. The model uses a decision-tree (deterministic) structure, including probabilities of the following: presentation to ANC; offer and acceptance of HIV testing; receipt of HIV test results; clinical assessment for ART eligibility; CD4 testing and receipt of results; offer of, acceptance of, and adherence to ARVs for PMTCT; maternal mortality during pregnancy; HIV testing in labor for women with unknown or negative HIV status; live birth; infant HIV infection by the time of delivery; and linkage to postnatal care and ART for mothers and infants (Figure A in Text S1). The CEPAC model of adult HIV infection is a computer simulation model of HIV disease in adults. Technical details of the CEPAC model are described in Text S1 [30],[38]; outcomes of the CEPAC model for postpartum women have been validated against published data (Text S1) [32]. For this analysis, the adult CEPAC model was used only to project maternal mortality risks during the first 2 y after delivery, in order to inform infant mortality rates and duration of breastfeeding in the infant model. A first-order, Monte Carlo simulation model of infant HIV infection and survival was added to the adult CEPAC model (Figure B in Text S1) [29]. Infants enter this model at birth and are assigned one of three HIV categories (HIV-unexposed, HIV-exposed but uninfected, or HIV-infected), as well as one of three maternal disease categories (HIV-uninfected, HIV-infected and “ART eligible,” or HIV-infected and “non-ART-eligible”). Over the first 2 y of life, modeled infants face a monthly probability of four key clinical events: (1) incident maternal HIV infection during breastfeeding, if mother was previously uninfected, causing infants to transition from “unexposed” to “exposed, uninfected”; (2) maternal death, with risks derived from the adult CEPAC model as above, after which infants are no longer at risk for HIV infection but are at higher risk of death due to orphanhood [39]–[42]; (3) infant HIV infection through breastfeeding, if infant was previously uninfected; and (4) infant death from any cause. Risks of infant mortality are stratified by infant HIV exposure and infection status, by receipt of ART if infected, and by maternal vital status. The MTCT and CEPAC models were linked to allow a combined analysis in which each woman–infant pair is simulated together from the time of first presentation at ANC through pregnancy and delivery (the MTCT model), and then each woman and infant are simulated separately through the first 2 y postpartum (the CEPAC models). Details of the linkages between the CEPAC and MTCT models have been published previously [29] and are further described in Figures A and B in Text S1. During ANC in the MTCT model, women may be lost to follow-up between first ANC presentation and delivery [11],[24],[43]–[47], between delivery and 6 wk postpartum [11],[43],[44],[48]–[50], and after linkage to postnatal HIV care [43],[44],[51]–[56]. In the absence of data regarding rates of adherence to the Option A or Option B regimens during breastfeeding, the impacts of medication interruptions or discontinuations were assumed to be included in sensitivity analyses examining “highest risk” postnatal MTCT estimates. Baseline maternal characteristics reflected cohorts of pregnant women in Zimbabwe (Table 2). At first ANC visit, mean age was 24 y (SD: 5 y) [57], and the proportion with CD4≤350/µl was 36%, based on data from Zimbabwe [58]. In the base case, MTCT risks were derived as the average (mean or midpoint) of risks reported in PTMCT studies in Africa (Table 3; Text S1) [4]–[10],[12]–[15],[58]–[69]. To avoid underestimating early postpartum MTCT risks, we included only data from breastfed infants, necessarily excluding pivotal PMTCT studies among replacement-fed populations [8],[70]–. In the absence of reported postnatal transmission risks for infants older than 6 mo of age who continue to breastfeed with ongoing prophylaxis (Option A or B), we assumed a constant monthly risk based on data in younger children, as has been observed for postnatal transmission without prophylaxis [73]. Infant mortality rates (Table 4) were derived from Joint United Nations Programme on HIV/AIDS HIV-deleted mortality estimates (HIV-unexposed infants) [74], the ZVITAMBO study in Zimbabwe (HIV-exposed, uninfected infants) [75], and pooled analyses of several African cohorts (HIV-infected infants) [26],[76],[77]. Validation of model-derived MTCT and mortality risks against published data, and sensitivity analyses on many clinical and programmatic parameters, have been reported previously [29]. For this analysis, in addition to the impact of PMTCT uptake and regimen examined in the base-case analyses, we conducted univariate sensitivity analyses on three key factors influencing transmission risk. First, to examine the impact of the range of published MTCT risks for each regimen, we defined “highest risk” and “lowest risk” scenarios. The “lowest risk” scenarios use the lowest published MTCT risks for each drug regimen, reflecting the best reported field effectiveness or trial efficacy, and likely representing the most adherent study populations; the “highest risk” scenarios use the highest published MTCT risks for each drug regimen. Second, because the proportion of women with CD4≤350/µl or WHO stage 3/4 disease at first ANC visit has been reported to range widely (23%–68% [78],[79]), we varied this proportion in sensitivity analyses from 0% to 100%. Third, we investigated the contribution of prolonged breastfeeding by projecting MTCT risk at 4–6 wk of age (primarily reflecting transmission that occurs in the absence of breastfeeding), as well as after 18 mo of breastfeeding (the median duration reported in a cohort from Zimbabwe [58]), with ARV prophylaxis continued throughout breastfeeding. We identified the parameters that exerted the greatest impact on MTCT risks by comparing the range of MTCT risks projected when each parameter was varied through clinically plausible ranges; the largest range of projected MTCT risks reflected parameters with the greatest influence on model results. We then performed multivariate sensitivity analyses to determine combinations of these factors necessary to reach MTCT risks less than 5%. To further examine the multiplicative impact of individual components of the PMTCT cascade, we varied uptake at each step in the care pathway (access to ANC, HIV testing in ANC, receipt of HIV test results, medication availability, adherence to medications, and linkage to postnatal HIV-related care), as well as the availability of CD4 counts and ART for women with CD4≤350/µl.