Reconstructing the PMTCT cascade using cross-sectional household survey data: The PEARL Study

Abstract

Background

Given the ambitious targets to reduce pediatric AIDS worldwide, ongoing assessment of programs to prevent mother-to-child HIV transmission (PMTCT) is critical. The concept of a “PMTCT cascade” has been used widely to identify bottlenecks in program implementation; however, most efforts to reconstruct the cascade have relied on facility-based approaches that may limit external validity.

Methods

We analyzed data from the PEARL household survey, which measured PMTCT effectiveness in 26 communities across Zambia, South Africa, Cote d’Ivoire, and Cameroon. We recruited women who reported a delivery in the past two years. Among mothers confirmed to be HIV-infected at the time of survey, we reconstructed the PMTCT cascade with self-reported participant information. We also analyzed data about the child’s vital status; for those still alive, HIV testing was performed via DNA PCR.

Results

Of the 976 eligible women, only 355 (36%) completed every step of the PMTCT cascade. Among the 621 mother-child pairs who did not, 22 (4%) reported never seeking antenatal care, 103 (17%) were not tested for HIV during pregnancy, 395 (64%) reported testing but never received their HIV-positive result, 48 (8%) did not receive maternal antiretroviral prophylaxis, and 53 (9%) did not receive infant antiretroviral prophylaxis. The lowest prevalence of infant HIV infection or death was observed in those completing the cascade (10%, 95%CI: 7%–12%).

Conclusions

Future efforts to measure population PMTCT impact should incorporate dimensions explored in the PEARL Study – including HIV testing of HIV-exposed children in household surveys – to better understand program effectiveness.

Background

Despite well-resourced efforts to dramatically reduce pediatric AIDS globally, an estimated 240,000 children were infected with HIV in 2013.¹ As national programs for the prevention of mother-to-child HIV transmission (PMTCT) expand their coverage, optimize services, and enhance standards of care, ongoing assessment of population-based impact is essential. The “PMTCT cascade” – the series of steps an HIV-infected pregnant woman must navigate to minimize HIV transmission risk to her newborn – has served as an important framework for assessing program performance and identifying bottlenecks in the delivery of care.² Data sources that have been used to construct these cascades run the gamut from routinely collected program indicators³ to intensive external surveillance activities.⁴ However, most are facility-based approaches that are unable to account for individuals who do not access institutional antenatal and/or delivery services. As a consequence, they may overestimate PMTCT effectiveness.⁵

Surveys using household-based sampling, such as Demographic and Health Surveys (DHS), hold the promise of improved external validity.² For example, in their recent report of the 2012 Kenya AIDS Indicator Survey (KAIS), Sirengo and colleagues assessed key steps along the PMTCT cascade using such cross-sectional individual-level information, including those who visited antenatal care during pregnancy, tested for HIV during pregnancy, initiated any maternal or infant prophylaxis, and reported an HIV-negative infant test 6 weeks after birth.⁶ We sought to further refine this household survey approach using data from the PMTCT Effectiveness in Africa: Research and Linkages to Care (PEARL) study conducted in Zambia, South Africa, Cote d’Ivoire, and Cameroon. Our overall aim was to describe a monitoring framework to illustrate gaps in PMTCT service provision and their relative contribution to negative health outcomes among HIV-exposed infants.

Methods

PEARL was designed to assess PMTCT program effectiveness at the facility and community levels. The study’s methodology and approach have been published elsewhere.^{2, 5, 7–10} In this analysis, we focused on the community-based household survey, which utilized a two-stage sampling technique. The first stage involved random selection of health facilities providing PMTCT in the country as a whole (Zambia) or in specific regions/provinces (South Africa, Cameroon, Cote d’Ivoire). In the second stage, the geographic area served by each facility was carefully defined, and households randomly sampled from each catchment area. Trained study teams enumerated participating households and obtained birth histories.

Households reporting a birth within the past 24 months were deemed eligible for further participation and recruited to participate. This included administration of a 165-question survey, modeled after the DHS,¹¹ as well as collection of maternal and infant blood specimens. Although specific practices varied by country,⁵ maternal specimens were first tested to identify the population of HIV-infected women using the Determine HIV-1 antibody test (Abbott Laboratories, Abbott Park, IL, USA). Among infants and children who were HIV-exposed, HIV DNA PCR testing was conducted to determine HIV transmission. During the period of study implementation (April 2007 to March 2009), available antiretroviral regimens at sites included peripartum single-dose nevirapine (NVP) alone, antenatal zidovudine with peripartum single-dose nevirapine (ZDV+NVP), or combination antiretroviral therapy (ART) for women who were eligible based on national HIV treatment guidelines. The exception was the Western Cape Province of South Africa, where a few sites provided combination antiretroviral regimens for prophylaxis alone. Antiretroviral prophylaxis during breastfeeding – either maternal or infant – had not yet been incorporated into country guidelines.¹²

We reconstructed the PMTCT cascade based on maternal or caregiver responses to specific questions about antenatal care access, HIV testing during pregnancy, receipt of a positive HIV test, initiation of maternal prophylaxis, and initiation of infant prophylaxis (the extent of interventions recommended at the time). We then identified individuals who dropped out at each step of the cascade and calculated the composite outcome of infant HIV infection or death (including 95% confidence interval) for each group. We also calculated the prevalence of infant HIV infection or death for those who completed the cascade. Statistical analyses were performed using Stata version 12 (Statacorp, College Station, Texas).

Results

From April 2007 to May 2009, 10,236 children under 2 years of age were identified from the sampled communities, of whom 2,251 (22%) who were excluded for missing child laboratory data. Reasons for exclusion included refused participation (n=1,465), refused specimen collection only (n=588), and missing results (n=198). Of the remaining 7,985 mother-child pairs, 1,014 (13%) mothers tested HIV-positive; 976 (96%) had complete data for both mothers and infants and were included in this analysis. Median age was 10.5 months (interquartile range [IQR]: 4.5, 17.4; Figure 1). 444 (45%) resided in Zambia, 365 (37%) in South Africa, 114 (12%) in Cameroon, and 53 (5%) in Cote d’Ivoire. Among the 410 women reporting use of maternal antiretroviral prophylaxis, the most common regimen was NVP only (37%), followed by ZDV+NVP (36%), ART (18%), and ZDV only (4%). The remaining 5% did not remember their regimen. Similarly, infant prophylaxis comprised 302 (76%) who reported using ZDV+NVP, 50 (13%) who used only NVP and 44 (11%) who used only ZDV. Other population characteristics are shown in Table 1.

Figure 1.

Age distribution of 976 HIV-exposed children enumerated in the PEARL community survey (2007–2009)

Table 1.

Characteristics of 976 HIV-infected mothers and their HIV-exposed infants enumerated in the PEARL community survey (2007–2009)

	Study population (n = 976)
Maternal characteristics
Age at survey in years, median (IQR)	27 (23–32)
Parity, median (IQR)	2 (1–3)
Marital status
Married or cohabitating	618 (64.5)
Other	344 (35.5)
Educational level
No schooling or primary	427 (44.1%)
Secondary or higher	541 (55.9%)
Enrolled for antenatal care during last pregnancy	949 (97.3%)
Gestational age when antenatal care started, weeks, median (IQR)	20 (16–24)
Delivery in a health-care facility	898 (93.2%)
Maternal antiretroviral prophylaxis initiated	410 (42.0%)
Antenatal ZDV and peripartum NVP	149 (36.3%)
Peripartum NVP only	150 (36.6%)
Combination antiretroviral regimens	74 (18.1%)
Antenatal ZDV only	18 (4.4%)
Cannot remember regimen	19 (4.6%)
Child characteristics
Age at survey in months, median (IQR)	10.5 (4.5–17.4)
Birth weight in grams, median (IQR)	3000 (2700–3400)
Birth weight ≤ 2500 g	104 (12.1)
Infant antiretroviral prophylaxis initiated	396 (40.6%)
NVP	50 (12.6%)
ZDV	44 (11.1%)
ZDV and NVP	302 (76.3%)

ZDV = zidovudine, NVP = nevirapine

The PMTCT cascade is illustrated in Figure 2. When we considered each step in relation to the study population of 976 mother-child pairs, 98% (95% confidence intervals [CI]: 97–99%) attended at least one antenatal visit, 87% (95%CI: 85–89%) tested for HIV, 47% (95%CI: 44–50%) received a positive HIV result, 42% (95%CI: 39–45%) initiated maternal antiretroviral prophylaxis, and 36% (95%CI: 33–39%) initiated infant antiretroviral prophylaxis. Overall, 355 mother-child pairs completed every step of the PMTCT cascade from antenatal care through completion of infant prophylaxis. Among the 621 mother-child pairs who did not, 22 (4%) reported never seeking antenatal care, 103 (17%) were not tested for HIV during pregnancy, 395 (64%) reported testing but never received their HIV-positive result, 48 (8%) did not receive maternal antiretroviral prophylaxis, and 53 (9%) did not receive infant antiretroviral prophylaxis.

Figure 2.

Graphic depiction of the PMTCT cascade from the PEARL community survey (2007–2009). The upper panel is the traditional representation of the cascade, showing those remaining at each step. Coverage in relation to the total number of HIV-infected women enumerated (n=976) is shown as a percentage above each column. The percentages listed between columns represent access from one step of the cascade to the next. In the lower panel, we show the numbers of mother-child pairs who dropped out at each step, along with the prevalence of HIV infection or death among HIV-exposed children within each separate group (measured according to right-sided y axis). A relatively large number of women reported not receiving an HIV-positive result, despite undergoing HIV testing (*). Several factors may have contributed to this, including fear of accidental HIV disclosure, incident HIV infections since delivery, and false-negative results during the antenatal period.

Overall, 103 (11%) children tested HIV-positive on HIV DNA PCR; another 40 (4%) had died by time of survey administration. (NB: These results differ from the 24-month HIV-free survival estimates from our primary analysis, which incorporated modeling techniques to account for the range of children’s ages at time of study-initiated HIV testing.⁵) The distribution of these outcomes by child age was as follows: 0–3 months (2 deaths, 13 HIV-infected among 164 children), 3–6 months (5 deaths, 17 HIV-infected among 137 children), 6–12 months (13 deaths, 23 HIV-infected among 244 children), and 12–24 months (18 deaths, 50 HIV-infected among 424 children). An additional seven children met eligibility criteria, but had missing age information. Two were reported dead and none tested positive for HIV. The highest prevalence of HIV infection or death, our composite outcome of interest, was observed among those who did not access antenatal care (30%, 95%CI: 10%, 51%). The lowest was noted among those who completed the cascade (10%, 95%CI: 7%, 12%). However, there was considerable overlap in the 95% confidence intervals for all groups (Figure 2). Because of a limited sample size, we were unable to construct country-specific PMTCT cascades.

Discussion

This analysis follows that of Sirengo and colleagues, who used data from the KAIS 2012 to reconstruct the PMTCT cascade at the population level.⁶ Specific features of the PEARL community methodology allowed us to build upon their approach. Rather than relying on self-reported clinical outcomes, we collected and tested biological specimens from HIV-exposed infants and children, to ascertain HIV status at the time of the survey. We also included information about infant and early childhood deaths. Incorporation of such outcomes represents a novel addition to the PMTCT cascade literature and is in line with World Health Organization calls for rigorous program assessment.¹³

The proportion of mother-child pairs completing the PMTCT cascade – from antenatal care to initiation of infant antiretroviral prophylaxis at birth – was low overall, at 36%. Nearly two-thirds dropped out at various steps along the cascade and this attrition appeared to correlate with higher rates of HIV infection and death, although the precision of infant and early childhood outcomes varied by subgroup. Our estimate of cascade completion was also lower than the 51% facility-based coverage reported in the PEARL cord blood surveillance.⁷ Although both study components were conducted in the same areas, source populations (facility vs. community) and the methods for defining coverage (drug levels vs. self-report) differed greatly. These inherent differences make direct comparisons of individual cascade steps difficult. However, when facility- and community-based measurements of overall “program coverage” (i.e., report or demonstration of antiretroviral drug use during pregnancy) were considered in previous analyses, there appeared to be high correlation between the two estimates.⁵ Not unexpectedly, facility estimates of coverage were consistently higher than those from the community survey.

Similar to KAIS 2012,⁶ the largest single loss in the cascade was observed among HIV-infected women who accepted HIV testing but reported not receiving a positive result. This finding is difficult to reconcile given the routine use of rapid HIV testing algorithms at surveyed sites. Although false-negative results during pregnancy and/or recent HIV infections may have contributed, we suspect that many participants were unwilling to disclose their positive HIV status to survey teams, even though they knew that blood was being collected for later HIV testing per study protocol. Many have reported on the stigma associated with HIV infection in the context of PMTCT.¹⁴ It is plausible that fear of accidental disclosure to household or community members led to systematic misclassification at this step of the PMTCT cascade. The design of future surveys should take into account this possibility. Strategies to better elicit sensitive information at the household level or to verify data from medical records should be explored further.

A key consideration for population-based household surveys is sample size, since it relates directly to the precision of outcome measures. In this analysis, we included only children at or under 24 months of age. Increasing the sample to include those under 5 years, as is common in the DHS and similar surveys, would expand the pool of eligible children. However, this could lead to reporting bias, since details of recent pregnancies are likely to be more accurately recalled. Incident maternal HIV infection may have also occurred after antenatal testing or after birth. Surveys can be enriched to include larger numbers of HIV-infected mothers who recently delivered. The KAIS 2012 cascade was nationally representative but based on only 79 mother-child pairs. Increasing the sample size through targeted recruitments would undoubtedly raise survey costs and may prove prohibitive in settings where adult HIV prevalence is low.¹³

Approximately 20% of eligible children were not included in our analysis, with the vast majority within this group refusing participation in the overall study. Because we were unable to collect demographic or health information about mother-child pairs who declined – or why they elected not to participate – it is unknown how this attrition may have influenced our outcomes of interest. Future household surveys should to devise and incorporate strategies to maximize participation, including novel community mobilization strategies, government engagement at multiple levels, and better alignment with – or integration into – more established household survey activities. Better systems for collecting and tracking blood specimens, as was introduced in KAIS 2012, can also minimize lost data among those who are enrolled.

The PEARL study was conducted in an earlier era of PMTCT, prior to the establishment of more efficacious regimens through the breastfeeding period. One-third of HIV-infected women reported use of peripartum nevirapine alone, a practice that is now obsolete in most HIV clinical guidelines and is no longer recommended by the World Health Organization.¹⁵ With the introduction of longer and more complex regimens for PMTCT, the traditional cascade must be expanded in order to remain useful as an evaluation tool. Given the significant attrition observed after antiretroviral regimen initiation,^16–18 for example, additional steps that account for adherence should be considered. Such data could be obtained through careful maternal questioning, linkage to facility-based records, and/or inclusion of maternal HIV viral load testing as part of the survey. As more countries adopt the “Option B+” approach for PMTCT (i.e., lifelong antiretroviral therapy for all HIV-infected pregnant women¹⁹), the PMTCT cascade will be become increasingly aligned to that of the HIV treatment cascade.²⁰ Greater emphasis must also be placed on infant HIV testing at key time point (e.g., at 4–6 weeks of age, at 18 months or cessation of breastfeeding), and final HIV status of HIV-exposed children following breastfeeding.

In this secondary analysis, we sought to refine the survey-based PMTCT cascade by incorporating infant and early childhood outcomes alongside points of attrition. Future efforts to measure population PMTCT impact should incorporate these dimensions of the PEARL study, including HIV testing of HIV-exposed children, to better understand overall program effectiveness and to identify barriers and weaknesses along the cascade. As countries move to more efficacious regimens for PMTCT, further refinement of this cascade – and its alignment with that of HIV treatment – will be needed.