Cost-effectiveness of World Health Organization 2010 Guidelines for Prevention of Mother-to-Child HIV Transmission in Zimbabwe

We projected outcomes for mothers and infants following World Health Organization–recommended regimens to prevent mother-to-child human immunodeficiency virus (HIV) transmission. Compared with Option A, Option B improves life expectancy and saves money; compared with Option B, lifelong maternal therapy is of comparable value to common HIV-related interventions.

transmission (PMTCT) can reduce perinatal HIV transmission to <2% in the absence of breastfeeding and to <5% by 6 months of age among breastfeeding infants [1][2][3]. As a result, the World Health Organization (WHO) has called for the "virtual elimination" of pediatric HIV [1][2][3]. Access to antiretroviral medications (ARVs) for PMTCT remains limited, however; only 59% of HIV-infected pregnant women received ARVs for PMTCT in 2010 [4]. As a result, nearly 400 000 new infant HIV infections occur annually, and HIV-infected women experience high postpartum morbidity and mortality [4][5][6].
In 2010, WHO issued revised guidelines for PMTCT [1]. The guidelines included a renewed emphasis on identification of pregnant, HIV-infected women with CD4 count ≤350 cells/µL or WHO stage 3-4 disease, who require lifelong 3-drug antiretroviral therapy (ART) for treatment of their own HIV infections and for PMTCT. For women with less-advanced disease, WHO recommends a country-or program-level choice between Option A (maternal zidovudine in pregnancy; infant nevirapine [NVP] throughout breastfeeding), and Option B (maternal 3-drug ARV regimens throughout pregnancy and breastfeeding, with interruption after weaning). Select programs are considering Option B+, in which maternal 3-drug regimens are initiated in pregnancy (regardless of maternal CD4) and continued throughout life, including throughout breastfeeding and subsequent pregnancies [3,7].
HIV prevalence in antenatal care (ANC) is estimated at 16% in Zimbabwe, leading to approximately 61 000 births per year to HIV-infected women [8,9]. Through 2009, the Zimbabwe National PMTCT Program provided single-dose NVP (sdNVP) to all HIV-infected women, with ART for women identified clinically as ART eligible [8]. Like most countries in sub-Saharan Africa, Zimbabwe initially implemented the revised WHO guidelines with Option A (with antenatal coverage of 46% in 2010) and will soon be examining the feasibility of Options B and B+ [4]. We used validated computer models of HIV disease and PMTCT [10][11][12] to project the clinical outcomes and cost effectiveness of implementing WHO-recommended PMTCT regimens in Zimbabwe.

Analytic Overview
We used 3 validated, linked computer models for this analysis ( Figure 1): (1) a model of a single pregnancy and delivery (the mother-to-child HIV transmission [MTCT] model [10]); (2) the Cost-effectiveness of Preventing AIDS Complications (CEPAC) model of HIV infection and mortality among breastfed infants (the CEPAC infant model [13,14]); and (3) the CEPAC-International model of HIV disease progression among postpartum women (the CEPAC adult model [11,12,15]). Clinical outcomes of the linked models included infant HIV infection risk at weaning, maternal life expectancy (LE) from delivery, and infant LE from birth. Economic outcomes, from the healthcare system perspective, included ANC costs (through delivery), maternal HIV-related healthcare costs, and infant healthcare costs.
Incremental cost-effectiveness ratios (ICERs), in US dollars per year of life saved (YLS), were calculated from combined projected lifetime healthcare costs (antenatal + maternal + infant) and combined projected life expectancy (maternal + infant) [16], discounted at 3% per year. We used 2 criteria to interpret cost-effectiveness. First, following WHO guidance, an intervention was considered cost-effective if its ICER compared with the next least-expensive alternative was <3 times the 2008 Zimbabwe per capita gross domestic product, or 3 × $400 = $1200 per YLS [17,18]. Second, we compared results with the recently reported range of ICERs for ARTrelated interventions in developing countries ($550-$5200 per YLS) [19]. This work was approved by the Partners Healthcare Institutional Review Board, Boston, Massachusetts.

Modeled Population, PMTCT Regimens, and Uptake of PMTCT Services
The linked models were used to simulate a cohort of pregnant, HIV-infected women in Zimbabwe and their infants. We examined 5 PMTCT regimens: (1) no antenatal ARVs (comparator), (2) sdNVP, (3) WHO Option A, (4) WHO Option B, and (5) Option B+ (Supplementary Table 1). Women were modeled to present to care at 24-28 weeks' gestation and to breastfeed for 18 months, based on Zimbabwean data [20,21], with ARV prophylaxis (Options A, B, and B+ ) continued throughout breastfeeding.
To demonstrate the impact of guideline-concordant care, all women in the base-case analyses were assumed to be identified as HIV-infected at their first ANC visit. With no ARVs, women received no antiretroviral medications during pregnancy. With sdNVP, women initiated ART in pregnancy if clinical assessment indicated WHO stage 3-4 disease; CD4 testing was not included, reflecting its limited availability in the sdNVP-based National PMTCT Program in 2009. With Options A and B, women received ART during pregnancy if eligible by either CD4 or clinical criteria, and with Option B+, all women received lifelong ART. With all modeled regimens, women who linked to postnatal HIV care were assumed to undergo clinical and CD4 assessment at 6 weeks postpartum and to initiate ART if eligible, regardless of antenatal regimen received. In the base case, we assumed 100% adherence to PMTCT regimens (initiated at 30 weeks' gestational age), 100% linkage to postnatal care for mothers and infants, and 100% retention in care and ART availability for women and infants meeting WHO ART initiation criteria [22,23]. In sensitivity analyses, to reflect real-world programs, we examined reduced access to antenatal and postnatal care.  Figure 2B). Linkages between the models allow a combined analysis in which each woman-infant pair is simulated together from the time of first presentation at antenatal care through pregnancy and delivery, and then each woman and infant are simulated separately throughout their lifetimes. Abbreviations: ANC, antenatal care; ART, 2-drug antiretroviral therapy; ARVs, antiretroviral medications; HIV, human immunodeficiency virus; OI, opportunistic infection; PMTCT, prevention of mother-to-child HIV transmission; sdNVP, single-dose nevirapine.

Model Structure
The 3 simulation models are described in detail in the Supplementary Appendix and in previous publications [10,14,15]. The models were linked so that each mother-infant pair was simulated together from the time of first presentation at ANC through delivery (the MTCT model), and then each woman and infant were simulated separately over their lifetimes after delivery (the CEPAC adult and infant models), as in Figure 1 and Supplementary Figures 1 and 2.

Model Input Parameters
Maternal Characteristics, Disease Progression, and ART Based on Zimbabwean data, mean age at first ANC visit was 24 years [21]; mean CD4 count was 451 cells/µL (36% of women with CD4 count ≤350 cells/µL) [24]. Because detailed data to inform monthly risks of opportunistic infections (OIs) and HIV-related death in the absence of ART were not available from Zimbabwe, we derived these data from a cohort in South Africa (Supplementary Table 2) [25]. Details of ART initiation and switching, as well as CD4 and HIV RNA changes on ART, are provided in the Supplementary Appendix.

MTCT Risks, Infant Mortality Rates, and Infant Life Expectancy Estimates
Risks of MTCT during pregnancy and breastfeeding were calculated from PMTCT studies among breastfeeding populations in Africa, leading to estimates similar to those derived by the Joint United Nations Programme on HIV/AIDS (Supplementary Appendix) [10,26]. Data and assumptions to inform infant mortality rates and LE values are shown in Table 1 and detailed in the Supplementary Appendix.

Cost Inputs
Monthly medication costs were from the Clinton Healthcare Access Initiative [27]. Costs of clinical care were determined by estimating resource utilization (number of inpatient days and outpatient visits) for specified health conditions, then multiplying by the estimated costs of these healthcare encounters in Zimbabwe (Table 1 and Supplementary Appendix) [28]. For children aged >18 months, monthly utilization estimates (stratified by HIV and ART status) were multiplied by LE to estimate lifetime healthcare costs.

Model Validation and Sensitivity Analyses
Model-derived risks of MTCT, infant mortality, and postpartum maternal OIs were validated against published data, reported previously with extensive sensitivity analyses [10,14]. For this study, we conducted univariate and multivariate sensitivity analyses on key PMTCT, pediatric, maternal, and cost parameters.

Access to Care Parameters
We examined the impact of reported rates of PMTCT uptake, defined as the proportion of HIV-infected women receiving PMTCT services and ARVs by delivery (56%, estimated for Zimbabwe in 2009; 80%, the 2009 WHO target goal; 90%, the 2011 WHO target goal; and 95%, reported in neighboring Botswana in 2011) [5,8,29]. We varied the availability of CD4 assays from 25% to 100% in Options A, B, and B+; when CD4 count was unavailable in Option A, women were assumed to initiate ART only for WHO stage 3-4 disease. We also examined the impact of reduced pediatric ART availability (36%, estimated for Zimbabwe in 2009) [5] and of reported rates of maternal loss to follow-up (LTFU) from postnatal HIV care (Table 1) [30][31][32].

Clinical Health Parameters
We defined a lowest-MTCT risk scenario, using the lowest published risks (best reported effectiveness/efficacy) for each modeled regimen (Table 1); a highest-MTCT risk scenario, combining the highest published risks for each regimen; and a scenario assuming equal MTCT risks with Options A and B. We also used 4 assumptions about LE for HIV-exposed and HIV-infected infants: (1) a high pediatric LE scenario, using the upper bound estimates shown in Table 1, (2) a low pediatric LE scenario, using the lower bound estimates, (3) a largest difference scenario (lowest estimates for HIV-infected children; highest estimates for HIV-uninfected children), and (4) a smallest difference scenario (highest estimates for HIVinfected children; lowest estimates for uninfected children).
Finally, we investigated potential maternal health impacts of Option B and B+ in 2 ways. First, we varied the efficacy of firstline ART when resumed after ART interruption, reflecting potential interruption-associated drug resistance. Next, we examined the impact of "treatment fatigue" for women who begin ART with CD4 count >350 cells/µL solely for PMTCT, modeled as (1) an increased risk of virologic failure >6 months after ART initiation or (2) a reduction in second-line ART efficacy.

Cost Parameters
Because estimated costs of healthcare in Zimbabwe are markedly lower than in surrounding countries [28], we repeated the analysis using costs from South Africa (Supplementary Table 2) [33]. In the base case, we conservatively assigned lifelong costs of NVP-based ART to HIV-infected infants; in sensitivity analyses, as an upper bound on pediatric ART costs, we assigned the costs of lifelong lopinavir/ritonavir-based ART to sdNVP-exposed, HIV-infected children. Finally, the nondrug costs of providing 3-drug ARV regimens instead of zidovudine alone (e.g., personnel, laboratory costs) have not been reported; we also examined the impact of such implementation costs in the antenatal period.

Pediatric Costs
PMTCT regimens that prevented more infant infections resulted in lower pediatric healthcare costs over time. After the early cost of infant NVP during breastfeeding, the pediatric healthcare costs of Option A became less than those of no antenatal ARVs by 4 years after delivery (Figure 2A). This finding persisted over longer horizons; undiscounted lifetime costs per infant ranged from $730 (no antenatal ARVs) to $370 (Options B and B+) ( Table 2).

Maternal Costs
Although small differences in short-term maternal costs resulted from modeled drug resistance following sdNVP, 5-year costs were similar for no antenatal ARVs, sdNVP, and Option A ( Figure 2B). Options B and B+, requiring 3-drug regimens during pregnancy and breastfeeding, conferred the greatest initial maternal healthcare costs. Option B conferred lower maternal costs than Option B+ after weaning because of deferred ART costs when women without advanced disease interrupted ART, and maternal costs with Option B were less than with Option A by 5 years after delivery (Figures 2B and  2C). Undiscounted lifetime maternal HIV-related costs per woman ranged from $8450 (Option B) to $9820 (Option B+ ( Table 2).

Cost-effectiveness Analysis
Option B was projected to result in a discounted combined lifetime cost (ANC + mother + infant) of $5630 per motherinfant pair and a discounted combined LE (mother + infant)  Figure 2D; Supplementary Table 7).

Access-to-Care Parameters
The finding that no antenatal ARVs, sdNVP, and Option A were more costly but less effective than Option B was robust with reduced uptake of PMTCT services or access to CD4 testing, as well as with current availability of pediatric ART, and the ICER of Option B+ compared with Option B in these scenarios remained $1370 per YLS (

Clinical Health Parameters
Base-case policy conclusions were unchanged in all modeled pediatric LE and MTCT risk scenarios, including when MTCT risks were equal with Options A and B, as well as throughout a variety of "treatment fatigue" scenarios for women initiating 3-drug regimens with CD4 count >350 cells/µL (Supplementary Table 5). Results were sensitive, however, to the risk of virologic failure after 6 months on ART. When this risk was increased 1.5-fold from the base case (to >2.4% per month), Option B no longer dominated Option A; when it was increased 2-fold (to 3.2% per month), Option A dominated Option B (Table 3; Supplementary Table 5).

Cost Parameters
Policy conclusions were unchanged when lifelong lopinavir/ritonavir costs were assigned to sdNVP-exposed, HIV-infected infants (Supplementary Table 5). In sensitivity analyses using South Africa healthcare costs, the ICER of Option B+   Table 3). The difference in antenatal implementation costs between 3-drug regimens and zidovudine alone needed to be ≥$150 per person to change the comparison between Options A and B (Table 3); at $150 per person, Option B was no longer cost saving but remained very cost-effective ($2 per YLS), compared with Option A. Even with implementation costs as high as $400 per person, the ICER of Option B compared with Option A remained <$400 per YLS (Supplementary Table 6).

DISCUSSION
There are 4 key findings from this work. First, a strategy of providing no antenatal ARVs for PMTCT is more expensive and less effective over a lifetime horizon than strategies based on sdNVP, Option A, or Option B. This result, which occurs because the upfront costs of these PMTCT regimens are greatly outweighed by the downstream costs of caring for HIV-infected infants, lends strong economic support to the well-recognized clinical impact of expanding access to PMTCT programs, regardless of the specific drug regimen provided [5]. Second, in settings where 3-drug ARV regimens are not available for PMTCT [5,34], replacing sdNVP with Option A benefits infants and mothers and saves money over a lifetime horizon. Third, healthcare programs would decrease costs and improve outcomes further by implementing Option B instead of Option A. Although short-term drug costs are greater with Option B, the incorporation of healthcare costs for both mothers and infants leads Option B to cost less than Option A within 4 years after delivery, primarily because of averted pediatric HIV costs ( Figure 2D). Notably, however, if women with high CD4 counts develop poor adherence after Option B (increasing the monthly risk of late virologic failure by ≥25%) (Supplementary Table 5) or if mothers are lost to follow-up after delivery, Option B leads to shorter projected maternal LE than Option A.
Finally, these results strongly support lifelong ART for all pregnant, HIV-infected women (Option B+) [3,7]. The interruption of effective ART in Option B may have deleterious effects on maternal health. Randomized trial data comparing maternal health outcomes of Options B and B+ are anticipated soon [35]. In the interim, we assume a rapid rate of CD4 decline after ART interruption based on other trials [36][37][38], with an associated increased risk of OIs. As a result, Option B+ is projected to increase undiscounted maternal LE by 1.12 years compared with Option B (consistent with modeled impacts of other HIV-related interventions [12,39]), with an ICER of $1370 per YLS. Although this ICER exceeds the 2008 gross domestic product-based threshold for cost-effectiveness in Zimbabwe ($1200 per YLS) [17,18], it falls in the lower Figure 2 continued. all 3 of these strategies are assumed to begin ART when CD4 count falls to ≤350 cells/µL or stage 3-4 disease develops. Small cost differences result from assumptions regarding nonnucleoside reverse transcriptase inhibitor resistance following sdNVP, but the slopes of these 3lines are similar. In Option B+, all women continue their 3-drug regimens. In Option B, women who did not have advanced disease before pregnancy interrupt their ARVs but remain in care and re-initiate ART once CD4 count falls to ≤350 cells/µL or stage 3-4 disease develops. As a result, maternal costs after weaning are greater with Option B+ than with the other regimens, and costs for Option B (due to delayed ART use) are much lower after weaning (becoming less than the costs after Option A by 5 years after delivery) (arrow). C, HIV-infected women with CD4 count >350 cells/µL, post-delivery. ART costs for women not eligible for ART during pregnancy (CD4 count >350 cells/µL, no stage 3-4 disease), from the Cost-effectiveness of Preventing AIDS Complications (CEPAC) adult model. Three postnatal scenarios are shown: (1) initiate 3-drug ARVs in pregnancy and continue ARVs after weaning (as in Option B+); (2) initiate 3-drug ARVs in pregnancy and interrupt ARVs after weaning (Option B); and (3) do not initiate ARVs in pregnancy but remain in care and initiate ART when needed (CD4 count ≤350 cells/µL or stage 3-4 disease, as in the no antenatal ARVs, sdNVP, and Option A strategies). Interrupting ART at weaning saves money compared with continuing ART; however, this ART interruption may be associated with negative health impacts for HIV-infected mothers if retention in care is less than 100% (Table 2). Antenatal costs are not included in (A, B, or C). D, Total cohort costs over the first 5 years after delivery. These include antenatal care costs (through delivery), maternal HIV-related healthcare costs, and pediatric healthcare costs. Option B becomes cost-saving compared with Option A within 4 years after delivery (arrow). Abbreviations: ART, antiretroviral therapy; ARV, antiretroviral medication; sdNVP, single-dose nevirapine.
range of ICERs reported for ART-related interventions in developing countries ($550-$5200 per YLS) [19] and thus represents a return on investment comparable with many current HIV programs in Zimbabwe and other resource-limited settings.
Option B+ may represent an even better healthcare investment compared with Option B under specific conditions. First, ART interruption (Option B) may cause greater detriment to maternal health under real-world programmatic conditions than in our guideline-concordant simulations. When women are lost to follow-up after weaning, disease progression is unobserved and cannot lead to prompt ART reinitiation. Such disease progression is more rapid when ART was interrupted months before LTFU (Option B) than at the time of LTFU (Option B+) because of lower CD4 counts at LTFU in Option B. As a result, Option B leads to a projected discounted LE (11.64 years) even lower than no antenatal ARVs (11.71 years) [10], and Option B+ becomes more cost-effective compared with Option B ($850 per YLS). Second, analyses using cost data from South Africa (ICER, $1410 per YLS; 2008 gross domestic product, $5700) [18] suggest that Option B+ may be very cost-effective compared with Option B in higherincome settings where healthcare costs are greater. Third, this analysis excludes several additional benefits of Option B+ that may render it even more effective and cost-effective, including prevention of maternal tuberculosis (also reducing infection risk in infants) [40], HIV transmission to male partners [40], hepatitis B flares due to ARV interruption [7], and MTCT during subsequent pregnancies when women are already on ART at conception [7]. There are several limitations to this analysis. First, all models necessarily simplify complex processes; for example, assumptions about infant LE involved uncertainties about healthcare in the distant future. However, LE assumptions, cost assumptions, and projected clinical and economic results were similar to those previously reported [41,42], and we tested the impact of biologic and operational assumptions in extensive sensitivity analyses [10,14]. Except where noted, the impact on policy conclusions was minimal, primarily because assumptions were consistent across PMTCT strategies. Second, we excluded the potential impact of drug-related viral resistance in infants who become infected despite exposure to modeled ARV regimens, because of limited data about acquisition of such resistance [43,44] and its impact on later ART effectiveness. If resistant HIV is a greater concern for infants who become infected while exposed to maternal ARVs through breastmilk than to extended NVP monoprophylaxis, the benefits of Options B and B+ vs Option A will be attenuated. Finally, our analysis assumed a healthcare system perspective. If a societal perspective were assumed, interventions that avert HIV infections in infants and prevent morbidity and mortality in women would be even more cost-effective, avoiding transportation costs and lost wages for medical care and permitting the productivity gains of healthy women and of children who will become healthy adults. As in other studies, we find that PMTCT programs based on sdNVP are cost saving, compared with no PMTCT interventions [45]. This is the first analysis to compare sdNVP and Options A, B, and B+ and to consider both short-and longterm maternal and infant outcomes after PMTCT [16,41,42,45]. We find that, with guideline-concordant care, Option A is cost saving compared with sdNVP; Option B becomes more effective and less expensive than Option A within 4 years of delivery; and Option B+ offers additional clinical benefits and economic value comparable with other widely used HIV interventions. We anticipate that the clinical results of these analyses will be generalizable to many African settings where prolonged breastfeeding is the norm and that the base-case economic results may also be applicable in low-income African countries with healthcare costs similar to Zimbabwe. Although specific policies will depend on available resources as well as important considerations of fairness, feasibility, and priority populations [15,46], PMTCT programs should move rapidly toward these more effective and economically efficient strategies.

Supplementary Data
Supplementary materials are available at Clinical Infectious Diseases online (http://cid.oxfordjournals.org). Supplementary materials consist of data provided by the author that are published to benefit the reader. The posted materials are not copyedited. The contents of all supplementary data are the sole responsibility of the authors. Questions or messages regarding errors should be addressed to the author.