Point-of-Care CD4 Testing to Inform Selection of Antiretroviral Medications in South African Antenatal Clinics: A Cost-Effectiveness Analysis

Background Many prevention of mother-to-child HIV transmission (PMTCT) programs currently prioritize antiretroviral therapy (ART) for women with advanced HIV. Point-of-care (POC) CD4 assays may expedite the selection of three-drug ART instead of zidovudine, but are costlier than traditional laboratory assays. Methods We used validated models of HIV infection to simulate pregnant, HIV-infected women (mean age 26 years, gestational age 26 weeks) in a general antenatal clinic in South Africa, and their infants. We examined two strategies for CD4 testing after HIV diagnosis: laboratory (test rate: 96%, result-return rate: 87%, cost: $14) and POC (test rate: 99%, result-return rate: 95%, cost: $26). We modeled South African PMTCT guidelines during the study period (WHO “Option A”): antenatal zidovudine (CD4 ≤350/μL) or ART (CD4>350/μL). Outcomes included MTCT risk at weaning (age 6 months), maternal and pediatric life expectancy (LE), maternal and pediatric lifetime healthcare costs (2013 USD), and cost-effectiveness ($/life-year saved). Results In the base case, laboratory led to projected MTCT risks of 5.7%, undiscounted pediatric LE of 53.2 years, and undiscounted PMTCT plus pediatric lifetime costs of $1,070/infant. POC led to lower modeled MTCT risk (5.3%), greater pediatric LE (53.4 years) and lower PMTCT plus pediatric lifetime costs ($1,040/infant). Maternal outcomes following laboratory were similar to POC (LE: 21.2 years; lifetime costs: $23,860/person). Compared to laboratory, POC improved clinical outcomes and reduced healthcare costs. Conclusions In antenatal clinics implementing Option A, the higher initial cost of a one-time POC CD4 assay will be offset by cost-savings from prevention of pediatric HIV infection.

(NIAID) (K01 AI078754 (ALC), R01 AI058736 ((KK, K. Doherty, TH, KAF, RPW), R01 AI093269 (KAF, RPW), East African IeDEA Consortium U01 AI069911, *and IMPAACT network U01 AI068632** (KK, SC, K. Doherty, TH, KAF, RW)), the Massachusetts General Hospital Executive Committee on Research (ALC), and the Elizabeth Glaser Pediatric AIDS Foundation (ALC, LM). Pointof-care CD4 measurement devices for use in the pilot evaluation were donated by Alere Healthcare. The funders and the suppliers of CD4 measurement devices had no role in study design, interpretation of results, or decision to publish. *Overall support for the IeDEA network is provided by the National Institute of Allergy and Infectious Diseases (NIAID), the Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD), and the National Cancer Institute. **Overall support for the International Maternal Pediatric Adolescent AIDS Clinical Trials Group (IMPAACT) was provided by the National Institute of Allergy and Infectious Diseases (NIAID) of the National Institutes of Health (NIH) under Award Numbers UM1AI068632 (IMPAACT LOC), UM1AI068616 (IMPAACT SDMC), with cofunding from the Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD) and the National Institute of Mental Health (NIMH). The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH. Support of the sites was provided by the National Institute of Allergy and Infectious Diseases (NIAID) and the NICHD International and Domestic Pediatric and Maternal HIV Clinical Trials Network funded by NICHD (contract number N01-DK-9-001/ HHSN267200800001C). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing Interests: The authors have read the journal's policy and have the following competing interests: POC CD4 devices were donated by Alere Healthcare for use in the services described herein. The authors also confirm that all of the funders, including Alere Healthcare, had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript, and that donation of the point-of-care CD4 machines by Alere Healthcare does not alter the authors' adherence to PLOS ONE policies on sharing data and materials. The authors confirm that Dr. Landon Myer is currently serving as a PLOS ONE Editorial Board member, and that this does not alter their adherence to PLOS ONE Editorial policies and criteria.

Analytic overview
In 2012, a point-of-care CD4 assay was introduced and evaluated in the Gugulethu Midwife Obstetric Unit (MOU), an antenatal clinic near Cape Town, South Africa (Appendix) [15]. We used data from this evaluation, with published clinical and cost data, to simulate a cohort of pregnant women identified as HIV-infected in antenatal care and their infants [13,14,[21][22][23][24]. We linked three validated computer models: 1) a decision analytic model simulating a cohort of women through a single pregnancy and delivery (the "MTCT model") [25][26][27]; 2) a Monte Carlo model of HIV disease among postpartum women (the Cost-effectiveness of Preventing AIDS Complications-International or "CEPAC-Adult model") [28,29]; and 3) a Monte Carlo model of perinatal and postpartum HIV infection among HIV-exposed infants (the "CEPAC-Pediatric model") [26,30]. Together, these three models simulate each mother-infant pair from the time of presentation to antenatal care (ANC) through the lifetimes of both mother and infant.
We projected short-and long-term clinical and economic impacts for two CD4 testing strategies: flow cytometry performed in a central laboratory ("laboratory"), and point-of-care testing performed in the antenatal clinic ("POC"). Clinical outcomes included MTCT risk at birth and weaning, pediatric life expectancy from birth, maternal life expectancy from presentation to care, and combined (maternal+pediatric) life expectancy. Economic outcomes, from the healthcare system perspective, included ANC costs, lifetime maternal HIV-related healthcare costs, lifetime pediatric healthcare costs, and 1-5-year maternal and pediatric health care costs (2013 USD). We calculated incremental cost-effectiveness ratios (ICERs) in $/life-year (LY): difference in combined healthcare costs (antenatal+maternal+pediatric costs) between the two strategies divided by difference in combined projected life expectancy (maternal+pediatric life expectancy). For ICERs, all outcomes were discounted at 3%/year [31]. We considered a strategy to be "very cost-effective," compared to the alternative strategy, if its ICER was <1x South African per-capita gross domestic product (GDP: $6,600 in 2013)/LY, "cost-effective" if the ICER was <3x GDP/LY, and "cost-saving" if it led to greater combined life expectancy and lower combined costs [32,33].

Modeled population and testing/treatment strategies
We projected outcomes for a cohort of HIV-infected, ART-naïve pregnant women and their infants in South Africa (Table 1), following a positive HIV test at the first ANC visit [13]. In laboratory, CD4 specimens were shipped to the national laboratory for flow cytometry, with results returned to patients at a second visit three weeks later. In POC, CD4 testing and result-return both occurred during the first ANC visit.
We simulated South African PMTCT guidelines at the time of the study, which reflected WHO Option A (lifelong maternal three-drug ART if CD4 350/μL or WHO Stage 3-4 disease; maternal zidovudine (AZT) in pregnancy, then daily infant nevirapine (NVP) throughout breastfeeding if CD4 >350/μL; Fig. 1) [3,5]. In the antenatal period, we therefore modeled provision of AZT to women who were awaiting CD4 results (laboratory), who never received CD4 results (either strategy), or who received results indicating CD4>350/μL (either strategy). We modeled provision of antenatal three-drug ART to women who received results indicating CD4 350/μL or who had evidence of WHO Stage 3-4 disease (either strategy). As a result, the model permits women with CD4 <350/μL to "incorrectly" receive AZT instead of ART, and thus have higher MTCT risks (Table 1), in order to incorporate the critical role of CD4 testing in selecting antenatal ARV regimens.
Women with initial or current CD4 350/μL continued lifelong ART, and those with initial and current CD4 >350/μL stopped maternal AZT and provided daily nevirapine syrup to their infants until weaning [3,5].
To isolate the impact of the CD4 testing strategies, in the base case, we varied only CD4 testing rates, CD4 result-return rates, and CD4 assay costs. We otherwise assumed guidelineconcordant care based on receipt of CD4 results: all women were accurately identified as HIVinfected, all mothers and infants adhered to prescribed PMTCT regimens, and all mothers and infants linked to postnatal HIV-related care, received ART if eligible after delivery, and were retained in care. To be conservative with regard to the benefit of POC, we assumed in the base case that laboratory CD4 had 100% sensitivity and specificity to detect true CD4 350/μL, and that POC had 93% sensitivity and 86% specificity to detect true CD4 350/μL (Table 1) [18]. We varied all of these assumptions in sensitivity analyses.

Model structure
We linked three computer models to simulate mother-infant pairs through pregnancy, breastfeeding, and the lifetimes of both mothers and infants (S1 Appendix; S1-S2 Figs.) [25][26][27][28][29]34]. At the time of presentation to ANC, mother-infant pairs enter the MTCT model, in which they face probabilities of key clinical events during pregnancy and delivery. MTCT model outcomes are assessed after delivery, and include maternal and infant vital status, infant HIV infection, and costs accrued during pregnancy and delivery. From delivery through death, clinical and economic outcomes are simulated for mothers in the CEPAC-Adult model and for infants in the CEPAC-Pediatric model. In these models, individuals are subject to monthly risks of clinical events, including opportunistic infections, response to ART, medication toxicities, and mortality, and the costs associated with these events (Appendix).

Model input parameters
Cohort characteristics, disease progression, and ART. We simulated the cohort of women seeking care at the Gugulethu MOU, with median age of 26 years and median gestational age at first visit of 26 weeks (Table 1, S1 Table) [13]. Monthly risks for opportunistic infections (OIs) and HIV-related death in the absence of ART were from Cape Town (adults) and from the International Epidemiologic Database for the Evaluation of AIDS (IeDEA; children) [30,35,36]. First-line ART was tenofovir/emtricitabine/efavirenz (TDF/FTC/EFV) for women and abacavir/lamivudine/lopinavir/ritonavir (ABC/3TC/LPV/r) for HIV-infected children [3,[37][38][39]. Further details of ART initiation, CD4 and RNA responses to ART, and switching to second-line ART regimens are provided in the Appendix [40][41][42][43][44].
MTCT risks. Modeled MTCT risks during pregnancy and breastfeeding, which substantially impact projected pediatric life expectancy, were the average values from published clinical studies in African breastfeeding populations, stratified by maternal CD4 count and ARV regimen received (Table 1, S1 Table, S1 Appendix) [26]. In sensitivity analyses, we also examined the impact of the highest and lowest published transmission risks for each regimen and CD4 stratum (S1 Appendix).
Effectiveness of CD4 testing strategies. We defined two key parameters for each CD4 testing strategy: the proportion of HIV-infected women undergoing CD4 testing, and the proportion of CD4-tested women receiving CD4 results and initiating three-drug ART if CD4 350/μL (result-return, Table 1; result-return rates below 100% reflect the proportion of women lost to follow-up before receiving CD4 results). For the laboratory strategy, data for testing (96%) and result-return (87%) were from the Cape Town MOU [45]. For the POC testing strategy, data for testing (99%) and result-return (95%) were from the pilot study of POC CD4 measurement at the MOU [45]. Based on MOU data, we modeled a 3-week interval between CD4 testing and CD4 result-return for the laboratory strategy [14].
Costs. POC CD4 assay costs (base case: $26) were derived according to Larson et al., substituting healthcare worker time observations from the MOU and local salary data in place of the Larson estimates (Table 1, S1 Table) [21]. Laboratory-based CD4 assay costs (base case: $14) were from published data [22]. During pregnancy, we included the costs of routine antenatal care and delivery (Appendix). After delivery, we included maternal and pediatric costs for routine HIV-related healthcare, acute care for opportunistic infections, ART, laboratory monitoring, and care in the final month of life (Appendix) [22,35,46,47]. All costs were in 2013 US dollars.

Model validation and sensitivity analyses
In previous work, we validated model-projected MTCT risk, pediatric survival, pediatric HIVfree survival, and maternal postpartum OI rates against published data, and we reported extensive sensitivity analyses on clinical, cost, and access-to-care parameters [25][26][27][28]. For this analysis, we examined additional variations in test sensitivity, specificity, testing rates, and resultreturn rates for the POC strategy, as well as antenatal and postnatal loss to follow-up (LTFU) rates, breastfeeding duration, healthcare and medication costs, MTCT risks, and the discount rate for both POC and laboratory strategies (S1 Table). We also examined both decreased POC CD4 costs, reflecting new POC assays in development, and increased POC CD4 costs, to incorporate possible costs not captured in the base-case estimate, for example: additional healthcare worker time to process CD4 specimens, undergo training, or conduct quality control activities; or reduction in staff capacity to perform other patient-related activities [21,[48][49][50]. Finally, we conducted multiway sensitivity analyses, varying POC assay cost, sensitivity, testing rates, and result return rates simultaneously.

Low laboratory access scenario
In many settings, access to laboratory-based CD4 testing is more limited than in Cape Town, an urban area close to central laboratory facilities. We therefore examined a second, "low laboratory access scenario," in which POC CD4 testing was introduced into a setting with a laboratory test rate (30%) and result-return rate (50%) based on UNAIDS data for low/middleincome countries [51].

Budget impact analysis
To inform short-term budgets, we projected not only lifetime outcomes, but also outcomes over a 5-year horizon. Outcomes included ANC costs, pediatric costs, total costs, and yearly pediatric survival rates for both laboratory and POC strategies. Because variations in access to laboratory-based CD4 testing had the greatest impact on lifetime cost projections, we repeated the budget impact analysis in the "low laboratory access" scenario.

Base-case results
Laboratory CD4 strategy. Laboratory resulted in a 4.2% MTCT risk at birth and 5.7% MTCT risk at 6 months ( Table 2). This strategy led to a pediatric life expectancy of 53.2 years (23.50 years discounted) and a maternal life expectancy of 21.2 years (14.8 years discounted), for a combined life expectancy of 74.3 years (38.3 years discounted). ANC costs were $310/mother, lifetime pediatric costs were $760/infant ($520 discounted), and lifetime maternal costs were $23,860/person ($15,440 discounted), for a combined cost of $24,930/mother-infant pair ($16,270 discounted).
Point-of-care CD4 strategy. POC resulted in a 3.8% MTCT risk at birth and a 5.3% MTCT risk at 6 months. This strategy led to a pediatric life expectancy of 53.4 years (23.6 years discounted) and a maternal life expectancy of 21.2 years (14.8 years discounted), for a combined life expectancy of 74.5 years (38.3 years discounted). ANC costs were $325/mother, lifetime pediatric costs were $710/infant ($480 discounted), and lifetime maternal costs were $23,860/person ($15,440 discounted), for a combined cost of $24,900/mother-infant pair ($16,250 discounted). Over a lifetime horizon for mother and infants, POC was cost saving compared to laboratory, with greater combined maternal and pediatric life expectancy and lower combined costs ( Table 2).

Sensitivity analyses
POC sensitivity and specificity. Holding test specificity at the base-case value of 86%, POC CD4 testing remained cost-saving (greater life expectancy and lower costs) compared to laboratory unless POC test sensitivity was 89%. POC resulted in a higher combined life expectancy compared to laboratory unless POC test sensitivity was 84% (Fig. 2: right panel). Holding test sensitivity at the base-case value of 93%, POC life expectancy increased compared to the base-case as the specificity of the POC assay decreased (i.e., more women with high CD4 were "incorrectly" assigned to ART than to AZT), and POC life expectancy never fell below laboratory life expectancy even at POC specificity of 100%. POC remained cost-saving (greater life expectancy and lower costs) compared to laboratory at all POC assay specificities.
POC testing and result return. POC remained cost-saving (greater life expectancy and lower costs) compared to laboratory unless the proportion of women tested and receiving POC assay results was 89%. POC led to a greater life expectancy compared to laboratory unless the proportion of women tested and receiving POC assay results was 85% (Fig. 2: right panel).
POC assay cost. Compared to laboratory testing, POC remained cost-saving unless the POC assay cost $51.
Other univariate sensitivity analyses. Results of sensitivity analyses on antenatal or postnatal loss to follow-up for mothers or infants, breastfeeding duration, healthcare and medication costs, a range of MTCT risks (including 6-week risk of 3.0% observed in a recent nationally representative sample [52]), and the discount rate are shown in S3 Table. Although projected MTCT risks, life expectancies, and costs differed as expected from the base case, the comparison between POC and laboratory was not sensitive to changes in these parameters, assuming they were varied similarly for both strategies.
Multivariate sensitivity analyses. When POC sensitivity, POC testing and POC result-return were high, POC was cost-saving at all POC assay costs from $13-52 (Fig. 3, upper right  corners). Conversely, at very low POC sensitivity, testing, and result-return rates, POC became more expensive and less effective than laboratory (Fig. 3, lower left corners). As POC assay cost increased, fewer combinations of sensitivity, testing rates, and result-return rates allowed POC to be cost-saving; however, POC was cost-effective or very cost-effective in many of these scenarios (Fig. 3, band from upper left to lower right corners).

Low laboratory access scenario
With lower rates of testing and result-return reflecting UNAIDS data [51], laboratory resulted in MTCT risks at birth and 6 months of 7.3% and 8.7%, as well as a pediatric life expectancy of 51.9 years (23.1 years discounted), a maternal life expectancy of 21.1 years (14.8 years per year, shown in parentheses. ANC costs were not discounted, because they accrued in the first year of the simulation. All costs are in 2013 USD. Projections are shown for a cohort of HIV-infected mothers after delivery, and for a cohort of their infants from birth (most of whom are HIV-uninfected). b. Laboratory testing strategies were dominated, meaning that they were more expensive (higher total ANC+maternal+pediatric costs) and less effective (lower total maternal+pediatric life expectancy) than the POC testing strategy.
c. POC results remain unchanged in the low laboratory access scenario.  Table 2). Cost savings from POC relative to laboratory were greater in this low laboratory access scenario than in the base case.

Budget impact analysis
In the base-case analysis, the higher upfront costs of POC were offset within 36 months after birth compared to laboratory, when both strategies reached costs of $490/mother-infant pair (Fig. 4, solid arrow). In the low laboratory access scenario, the upfront costs of POC were offset within 6 months after birth compared to laboratory (Fig. 4, open arrow). In both the base-case and low laboratory access scenarios, pediatric survival was slightly greater with POC than with laboratory at all time points (S4 Table).

Discussion
In antenatal care, use of laboratory-based CD4 assays can lead to delayed result-return, loss to follow-up, and missed opportunities for PMTCT. In South African PMTCT programs prioritizing ART for women with advanced HIV infection (Option A), our model-based analyses suggested that any improvement in testing rates or result-return rates will more than offset the cost of point-of-care CD4 assays, and that the benefits of POC testing will be even greater in settings where access to laboratory-based CD4 testing is limited. Projected cost savings were due to the greater number of pediatric HIV infections averted with POC; savings occurred within 36 months after delivery in regions with high access to laboratory CD4 testing and within 6 months in regions with lower access to laboratory CD4 testing [51], and persisted throughout the lifetimes of mother-infant pairs. Although POC CD4 assays are a newer, more expensive technology, they may provide the greatest clinical and economic benefits in settings with the most limited healthcare resources. Our analysis was based on South African PMTCT guidelines during the study period, which followed the 2010 WHO-recommended Option A strategy with an ART-eligibility CD4 threshold of 350/μL. As of 2013, South African guidelines now recommend that programs transition to WHO's Option B (three-drug ART for all HIV-infected women during pregnancy and breastfeeding regardless of CD4, with cessation of ART after weaning for women with high initial CD4). Based on a programmatic goal to harmonize treatment for pregnant and nonpregnant patients, many other countries are also planning to implement Option B; in addition, many programs are moving toward an ART-eligibility CD4 threshold for non-pregnant patients of 500/μL. As these transitions occur, the role of POC CD4 testing in antenatal care will evolve from the scenarios examined in our analysis. For example, with Option B, CD4 testing will be necessary to inform decisions about lifelong ART (women with lower pre-ART CD4) versus discontinuation of ART after weaning (women with higher pre-ART CD4), however, rapid POC CD4 result-return in ANC will not impact the choice of antenatal ART versus zidovudine [4]. If programs implement Option B+ (lifelong ART regardless of initial CD4) [53,54], POC CD4 assays may play a role in monitoring ART and switching to second-line ART [55][56][57][58].
The current analysis did not examine these different roles for POC CD4 assays in Options B/B+. At present, the majority of HIV-infected pregnant women in Africa are still treated under Option A [59]. Although a shift to Option B/B+ is planned in many countries, this will likely require months or years to complete [4,7,59]. Based on available MTCT data and PMTCT guidelines during the study period, we also modeled a CD4 threshold of 350/μL to determine use of AZT versus ART [2,5,60]. The transitions away from Option A and ART initiation CD4 thresholds of 350/μL will likely occur most gradually in remote and more resource-limited settings, which also have the poorest access to laboratory-based CD4 testing [51]. While the planned transition to Options B/B+ is ongoing, the majority of HIV-infected, pregnant women in Africa are still receiving interventions based on CD4 counts [59]. In these programs, the rapid result-return permitted by POC CD4 testing could substantially improve linkage to HIV care and ART initiation for pregnant women, leading not only to clinical benefits, but also to cost savings.
Notably, we found that with Option A, reducing modeled POC test specificity improved the projected clinical and economic benefits of POC testing. This occurred because, using the average of published MTCT risks for each regimen (Appendix), we modeled lower intrauterine/ intrapartum transmission risks with maternal three-drug ART than with AZT [61][62][63][64]. Reductions in POC specificity led more women with high CD4 to start three-drug ART "incorrectly," reducing MTCT risks and increasing pediatric life expectancy. Although this was not the focus of our analysis, this finding provides additional support for the universal initiation of ART during pregnancy [3,6]. Our modeled MTCT risks were not based on a direct comparison of the Option A and Option B regimens in women with high CD4, however; a randomized trial comparing these two regimens is in progress [65]. Additional impacts of Option B compared to Option A, such as effects on maternal disease progression, prevention of HIV transmission to sexual partners, and reduced risk of maternal or infant tuberculosis, were beyond the scope of this analysis [25,[66][67][68].
There are several limitations to this analysis. First, models necessarily simplify complex clinical and operational processes. Although our three linked models with differing structures did not permit probabilistic sensitivity analyses or generation of 95% confidence intervals, we followed international guidance and conducted extensive univariate and multivariate sensitivity analyses to examine which assumptions and data inputs led to changes in policy conclusions [69]. Second, we assumed a sensitivity and specificity of 100% for laboratory CD4 testing, as the gold standard. We chose this simplifying assumption to be conservative with regard to the benefits of POC CD4 testing, as well as to accurately map modeled CD4 strata to those in the PMTCT trials providing input data, which used laboratory CD4 measurements. Third, our analyses may exclude unforeseen impacts for women who receive HIV test results, learn CD4 results, and initiate ART at the first ANC visit. Although pilot data suggest high levels of acceptance for POC CD4 result-return and rapid ART initiation [14], if this process leads women to feel overwhelmed, and thus to maintain poor adherence and retention in care at later time points, the benefits of POC CD4 testing may be attenuated [70]. Finally, POC CD4 assays permit initiation of ART several weeks earlier than laboratory-based testing. Receipt of three weeks of ART in place of AZT will likely have minimal impact on maternal health, but may substantially reduce MTCT risk, especially late in gestation [11,12,71]. To include this effect, the model would require MTCT risks stratified by both duration of AZT and ART use and the gestational age at which each regimen is initiated. We were unable to identify such data, so we examined only improvements in testing and result-return rates related to POC. Because we did include the additional costs of three weeks of ART, however, this was a deliberately conservative assumption; including the MTCT reduction from three additional weeks of ART would likely show POC to be even more effective, with even greater cost-savings, compared to laboratory.
In conclusion, although additional funding will be needed to implement POC CD4 testing in the short-term, we find that POC CD4 testing will improve clinical outcomes and will save money within 1-3 years of delivery, compared to laboratory-based CD4 testing, in a range of settings. POC CD4 testing should be implemented in PMTCT programs that prioritize ART for women with advanced HIV infection.  6)). The MTCT model is a decision tree, coded in TreeAgePro software. Pregnant women enter the model at conception. Five possible PMTCT strategies are shown at the decision node, indicated by a square. This analysis examined only Option A. Circles indicate chance nodes, at which events occur based on probabilities derived from published literature. Triangles indicate terminal nodes, representing the clinical outcome of any single pathway through the model. Brackets reflect that the subsequent events emerging to the right of the bracket may follow any of the prior chance nodes included to the left of the bracket. At each chance node, the probabilities of all subsequent modeled events may depend on the PMTCT strategy being simulated and on the prior events leading to that node. For each modeled PMTCT strategy, the series of events shown in the figure may occur. For example, HIV-infected women may be ART-eligible (CD4350/μL or WHO Stage 3-4 disease) or non-ART-eligible; ART-eligibility may be identified by CD4 testing, identified by clinical evaluation, or not identified. All women may access ANC, undergo HIV testing in ANC, and receive HIV test results, or may fail to access these steps in the cascade. If identified as HIV-infected, women may be offered ARVs for PMTCT according to the PMTCT strategy being simulated, as well as ART if identified as ART-eligible (not shown). In the base case, women were assumed to receive all ANC services; service uptake was varied in sensitivity analyses. Probabilities for surviving pregnancy depend on receipt of ART; if maternal death occurs, infant death also occurs. Women who survive pregnancy may deliver at a healthcare facility or at home; if they deliver in a healthcare facility, they may access HIV testing (if previous status was unknown or negative), and if identified as HIV-infected at that time, may receive sdNVP in labor (sdNVP was excluded for this analysis). All women surviving pregnancy then experience probabilities of live birth and HIV infection in their infants, depending on PMTCT regimen received. Finally, women may link or fail to link to postnatal HIV-related care for themselves. At the end (far right) of any given path through the model, there are two sets of outcomes: infant outcomes and maternal outcomes. Infant outcomes include HIV infection status (infected or uninfected at birth, shown), risk of postnatal HIV infection if uninfected at birth, life expectancy, and per-person healthcare costs. Maternal outcomes include life expectancy and per-person HIV-related healthcare costs. These outcomes are derived from the CEPAC Adult and Pediatric models, through specific simulations of each possible scenario described at the end of the pathways shown in the MTCT model. As an example of infant outcomes from the CEPAC-Pediatric model, an HIV-uninfected infant with an ARTeligible mother who is in postnatal care (and thus on ART) would face monthly risks of HIV infection based on receipt of maternal ART during breastfeeding; if infected postnatally, the infant would face CD4-and age-stratified monthly risks of OIs, ART failure or toxicity, and AIDS-related and AIDS-unrelated death, leading to a LE and lifetime cost projection for the infant. As an example of maternal outcomes from the CEPAC-Adult model, this infant's mother would face CD4-dependent monthly risks of OIs, ART failure or toxicity, and AIDS-related and AIDS-unrelated death, leading to a LE and lifetime cost projection for herself. These CEPAC model outputs are then used as "payoffs" (outcomes) in the MTCT model, according to conventional methods for evaluation of a decision tree. The average value assigned to any modeled PMTCT strategy in the MTCT model is, in essence, a weighted average of the value of these outcomes at the end of each pathway (weighted by the probabilities of reaching each possible path endpoint). Abbreviations: ARVs: antiretroviral drugs; ART: three-drug antiretroviral therapy; ANC: antenatal care; AZT: zidovudine; sdNVP: single-dose nevirapine (excluded for this analysis).  Fig.) after delivery; for this analysis, all modeled women enter with chronic HIV infection. They then face monthly risks of clinical events including opportunistic infections, medication toxicities, and death; these risks are stratified by the parameters listed in the figure. Life months accrued between presentation to antenatal care and delivery are added to the CEPAC model projections.

Supporting Information
Infants enter the pediatric model (S2B Fig.) after delivery; for this analysis, all modeled infants enter either HIV-negative or with an intrauterine/intrapartum HIV infection. Infants HIVnegative upon entering the model can either develop a postpartum HIV infection via breastfeeding or remain HIV-negative. HIV-infected infants face monthly risks of clinical events including opportunistic infections, medication toxicities, and death; these risks are stratified by age and CD4%. (TIFF) S1 Table. Complete input parameters for a model of mother-to-child transmission in South Africa (includes parameters listed in manuscript Table 1).