In this study, we analysed aggregate data from 283,410 pregnant women attending 269 PMTCT sites. We identified several factors operating at the country-level that significantly influenced nevirapine coverage in women, including PMTCT knowledge, HIV prevalence, rural population proportion and proportion of GDP dedicated to healthcare expenditure. These factors were independent of site-level factors, such as the site activity or the type of health facility, and program-level factors, such as the status of the institution or the program burden measured as the number of sites included. To our knowledge, this study is the first to explore the contextual determinants of PMTCT program performance within the formalized framework of multilevel modeling. The contextual predictors identified explained almost 100% of the initial variance at the program/country level. Our data also showed a huge heterogeneity of performance between sites within the same programs.
Among 269 PMTCT sites, we found a nevirapine coverage ratio of only 44%. Our result is only somewhat more optimistic than a study by Temmerman et al. reporting a NVP administration rate of 20%, using a similar whole-ANC-population-based calculation process [19
]. The take-up testing rate by site was 61.5%, while other studies reported testing rates ranging from 21% to 95% [20
]. This variability is echoed in our data, since women tested percentages differed across programs from 7% to 97%. Test acceptability after pre-test counselling is variable but overall high ranging from 72% to 97%, as indicated in several previous reports [19
]. In regards to NVP provision to HIV positive pregnant women by site, our finding of 70.2% is consistent with other previous reports indicating rates from 56% to 94% [19
Specific PMTCT and general comprehensive knowledge about AIDS were significantly associated with higher nevirapine coverage, whereas general awareness about AIDS – as measured by the proportion of persons who had heard of AIDS – and accepting attitudes or stigma were not significantly linked to NCR. This is consistent with previous reports showing knowledge as an important factor of success in antiretroviral adherence [32
] or breastfeeding practices [33
]. Above all, this result is encouraging to programs aiming to increase knowledge about PMTCT and AIDS, which is a potentially modifiable contextual determinant. The absence of a significant result for stigma in our study requires cautious interpretation since methods for quantifying stigma are neither straightforward nor consensual. Beyond the difficult medical prognosis, knowledge of HIV positive status is of great social consequence, and stigma and violence are still actual threats for African women [34
]. Other factors must be taken into consideration for improving HIV test acceptance, such as educational background, familial environment and most importantly male-spousal involvement. Disclosure of HIV status remains problematic [37
] and a woman's perception of her husband’s approval of testing plays a crucial role in this context [38
The national prevalence of HIV remained significant after accounting for knowledge about AIDS and healthcare expenditure. In this context, the interpretation of this indicator might be of interest because it likely encapsulates several other contextual features (e.g. testing coverage, which lost its statistical significance in multivariate analysis) or other factors not assessed in our analysis. Thus, we may interpret national HIV prevalence as an awareness and readiness marker indicative of both population and health services.
Rural population as a percent of total population was negatively associated with NCR. This result is consistent with the loss to follow-up of pregnant women [21
], as maternal NVP coverage has been reported to be associated with the number of ANC visits [39
]. The quality of follow-up remains a challenge for rural and remote facilities [40
]. Interestingly, positive results can been achieved by providing antiretroviral prophylaxis for mother and infant early in pregnancy [12
] or by offering labor ward-based services for PMTCT [42
]. Beyond single-dose nevirapine regimens, rural primary health facilities likely face other difficulties as well, such as the capacity to determine the CD4 cell count required to implement more efficacious antiretroviral regimens [43
National health expenditure as a proportion of GDP was an independent and positive predictor of nevirapine coverage, probably suggesting an overall positive impact of increased investment in the organization and infrastructure of health services. This finding may confirm the importance of overall health system strength to ensure effective PMTCT services in the field, where performance is notably linked to sites’ infrastructure and testing capacities [13
] as illustrated in our study by the strong association found between this indicator and the testing rate by site.
At the site-level, the highest and lowest nevirapine coverage results could be observed within the same program, emphasizing challenges to the on-the-ground feasibility of PMTCT programs. Beyond infrastructure and testing capacities, lack of sufficient and well trained staff have been suggested to explain some of the discrepancies observed between sites. Integration of voluntary counselling and testing (VCT) services in pre-existing ANC settings represents an undeniable additional workload for in-place health workers. At the same time and for economic reasons, human resources have not always risen adequately [19
]. Efficient counselling activity requires trained staff fulfilling specific capabilities [25
]. Test acceptability and women’s satisfaction are moreover strongly related to the capacity of PMTCT services to provide good quality counselling and follow-up care to pregnant women [45
]. In our study, a moderate ANC activity (30–100 pregnant women attending ANC services per month) was associated with lower nevirapine coverage ratios. This finding could illustrate the negative effect of increased workload in medium size facilities, where more difficulties may have been encountered to adjust to the increased resource requirements for PMTCT.
We consider the generalizability of our findings in the context of global PMTCT programs. We included VDP programs with data available by site or one-site programs, which raises the question of the representativeness of the sites included, both in regard to the whole VDP program and to other existing PMTCT programs. We did not identify any significant difference between included and not-included VDP programs regarding their main characteristics, which suggests limited selection bias within the VDP program itself. As for the other PMTCT programs, our program panel cannot be considered as fully representative of all programs, implemented either within the same countries or elsewhere in sub-Saharan Africa and/or based on different antiretroviral drugs and regimen designs. However, our analysis was conducted on a range of sites and programs, including primary or secondary/tertiary health facilities with various ANC activities within programs from several different sub-Saharan African regions, thereby supporting the external validity of our results. Secondly, a possible selection bias cannot be excluded in regards to the pregnant women included in this study, since only women actually attending ANC services were considered. Finally, our findings stem from an early era in large scale implementation of PMTCT services, when antiretroviral interventions were based on much simpler regimens than the multi-drug regimens currently recommended [9
]. However, we believe that most of the factors identified in the present work as determinants of NVP coverage are likely to apply to current antiretroviral interventions as well. Barriers such as a limited access to services in remote rural areas or the influential role of population knowledge about PMTCT still have resonance today. Thus there is still great interest in documenting such obstacles, especially because they were encountered with an intervention as simple as single-dose NVP and may also extrapolate to the much more complex interventions of today. More generally, our results also highlight the potential interest in promoting the use of multilevel analyses when assessing factors affecting program scale-up, be it in the framework of PMTCT or other public health domains. Multilevel approaches are useful for researchers and can bring relevant information to policy makers by weighing the relative impact of site-, program- or contextual-level predictors of successful implementation of health programs.
Our study has several limitations that should be noted. In particular, we were not able to control whether a pregnant woman actually took the NVP dose she was dispensed in PMTCT services. Specifically, we lacked the ability to directly measure the drug compounds contained in cord blood samples, which is an accurate and useful means to counter the limited quality of reported medical data when assessing program coverage [39
]. NCR tends to overestimate overall PMTCT coverage, since it does not account for women without access to PMTCT services. We believe NCR can still yield at a glance a direct indicator of the process quality of a PMTCT program or site, encompassing VCT and final NVP administration. Further, we were not able to use DHS sub-national data for this study, given low sample size by area unit (mean number of sites by region <5). Yet, even considering large entities such as countries, a high between-country heterogeneity was observed with respect to contextual factors and NVP coverage, indicating that country-level indicators could still represent relevant variables to assess the ‘average’ contextual environment of the sites within countries. Finally, our results should be interpreted cautiously since we lacked complementary data to describe health facilities (e.g. staff training), as illustrated by the low percentage of variance explained at the site-level in multilevel analysis, or programs (e.g. availability of adherence support services). It is also likely that some relationships identified between NCR and country-level factors, such as the proportion of rural population or healthcare expenditure, may partially but not entirely be explained by uncontrolled site-level factors (i.e. rural/urban facility, quality of infrastructure) that were not available in our dataset.