The structure of the model is illustrated in . Annual numbers of births to women who are HIV-seronegative at their first antenatal visit, and annual numbers of births to women who are HIV-seropositive at their first antenatal visit, are both estimated from the ASSA2003 AIDS and Demographic model, a model of the South African HIV epidemic that is calibrated to HIV prevalence data collected at first antenatal visits.14, 15
This model is also used to estimate annual HIV incidence rates in pregnant women, and the annual incidence rates and numbers of births are shown in an online appendix (Table 1 of Supplemental Digital Content 1
). The first antenatal visit is assumed to occur at 23 weeks gestation16-18
and delivery at 39 weeks,18
on average, so that the average time in which a woman seronegative at her first visit can acquire HIV before delivery is 20 weeks if a 4-week window period is assumed.19
The probability that a pregnant woman seronegative at her first antenatal visit acquires HIV before delivery is therefore calculated as the annual HIV incidence rate in pregnant women multiplied by a factor of 0.38 (20/52). The assumed probability that a woman who acquires HIV in late pregnancy transmits HIV perinatally is based on a review of studies conducted in settings where breastfeeding is rare (summarized in ) and on perinatal transmission rates from HIV-positive South African mothers who reported previously testing negative.20, 21
Multi-state model of mother-to-child transmission
Review of studies assessing HIV transmission from mothers who seroconverted during pregnancy or while breastfeeding, in the absence of antiretroviral prophylaxis
Assumptions regarding infant feeding practices in women who are HIV-negative or HIV-positive but undiagnosed are based on the results of the 1998 Demographic and Health Survey.16
87% of these women are assumed to start breastfeeding, and the duration of breastfeeding is modelled using a Weibull distribution with a median of 18 months and a shape parameter of 2. All of these women are assumed to practise mixed feeding, as exclusive breastfeeding (EBF) was rare prior to the introduction of PMTCT programmes.16, 34
Women who were HIV-positive at delivery and who practise mixed feeding are assumed to have a fixed monthly probability of transmitting HIV, h1
. Breastfeeding HIV-negative mothers are assumed to acquire HIV at the same rate as pregnant women, and for an average period of 3 months after acquiring HIV, are assumed to have a higher monthly probability of transmitting HIV through mixed feeding, h0
. The parameter h0
is estimated from studies of the cumulative HIV transmission risk from breastfeeding mothers who have seroconverted (summarized in ), by noting that this cumulative risk can be expressed as
is the average duration of breastfeeding after seroconversion. Setting μ
= 9 months (half of the average duration in the two largest studies32, 33
) and setting h0
= 0.16 gives a cumulative transmission risk of 0.28, consistent with the pooled estimate of 0.27 in .
In each year, a proportion of pregnant women are assumed to receive HIV testing, increasing from 3% in 2000 to 92% in 2010 and subsequent years (see of Supplemental Digital Content 1
). A fraction of those women testing positive start antiretroviral treatment (ART) if their CD4 count is below 200 (or below 350 following the change in South African guidelines in 201035
). Of the remaining women who are diagnosed HIV-positive, a fraction is assumed to receive single-dose nevirapine, and following changes in guidelines in 2008,36
a fraction of women receiving single-dose nevirapine are assumed also to receive short-course zidovudine. Of women who are diagnosed HIV-positive antenatally, 50% are assumed to avoid breastfeeding completely,37, 38
35% practise EBF and 15% practise mixed feeding.39
The monthly probability of postnatal transmission is reduced if the child receives extended nevirapine prophylaxis, if the mother receives ART or if the mother practises EBF. HIV-diagnosed women who practise mixed feeding are assumed to do so for a median of 7 months. HIV-diagnosed women who practise EBF are assumed to do so for a median of 2 months (up to a maximum of 6 months), after which 30% are assumed to discontinue breastfeeding completely and the remainder practise mixed feeding (i.e. continue breastfeeding while introducing complementary feeds), for a median of 7 months.39-41
Following the change in guidelines in 2010,35
a proportion of HIV-diagnosed women who choose to breastfeed are assumed to administer extended nevirapine prophylaxis to their children, with this proportion rising to 80% by 2013. Following the more recent announcement of a phasing out of free provision of formula milk in public clinics, the proportion of HIV-diagnosed women who avoid breastfeeding is assumed to decline from 50% in 2010 to 20% in 2013. Assumptions about vertical transmission rates and the efficacy of PMTCT are summarized in , and a more detailed description of the model is provided in Supplemental Digital Content 1
Mother-to-child transmission assumptions
Children who acquire HIV are assumed to progress to a state of ART eligibility, after which they may start ART. Rates of progression to ART eligibility and rates of AIDS mortality in ART-eligible children are assumed to depend on age and mode of transmission (perinatal or postnatal), as described elsewhere.42
Numbers of new infections in children and numbers of children in different exposed and infected states are calculated at monthly time steps, starting in 1985. To ensure that the model assumptions regarding MTCT and paediatric HIV survival are plausible, the model is fitted to age-specific paediatric HIV prevalence data from national household surveys conducted in 2005 and 2008,60, 61
using a Bayesian uncertainty analysis approach. Beta prior distributions are specified to represent ranges of uncertainty around key parameters, and the means and standard deviations of these distributions are included in . Posterior distributions, representing the ranges of model results consistent with both the observed paediatric HIV prevalence data and the ranges of uncertainty around the input parameters, were simulated numerically using Incremental Mixture Importance Sampling.62
Three possible interventions are considered to reduce vertical transmission from mothers who acquire HIV after their first antenatal visit:
- Maternal HIV incidence is assumed to reduce by 50% after 2010. A 50% reduction in HIV incidence by 2011 is a target of South Africa’s National Strategic Plan,63 and could potentially be achieved in pregnant and breastfeeding women through intensified condom promotion, partner outreach, microbicides and pre-exposure prophylaxis.
- The offer of antenatal screening is assumed to be repeated at 34 weeks gestation in women who tested negative or did not receive testing at their first antenatal visit, from 2010 onwards. This has been recommended in recent PMTCT guidelines,35, 36 but implementation has been limited. It is assumed that 80% of women who previously tested negative would accept the offer of retesting,17 and 50% of women who did not receive testing at their first antenatal visit get tested at 34 weeks.
- Mothers and infants are assumed to be screened for HIV at 6-week immunization visits, from 2010 onwards. Although not part of current South African guidelines, this has been proposed by Rollins et al.20 92% of mothers are assumed to attend 6-week immunization visits,16 and 66% of women testing positive are assumed to receive their test results.21 Of those women who are diagnosed HIV-positive, 50% are assumed to discontinue breast feeding immediately (this proportion reducing to 20% by 2013), and those who continue to breastfeed are assumed to do so for a shorter period (median 7 months), with 80% administering nevirapine prophylaxis to their infants.
For ease of reference, we use the term ‘recently-infected mothers’ to refer to women who seroconvert after their first antenatal visit, either during late pregnancy or while breastfeeding.