This study has demonstrated the usefulness and limitations of using service data to inform HIV-free survival as a result of PMTCT programs. It measured not only the proportion of infants that were not HIV-infected, but also the ultimate outcome of PMTCT intervention: a living child free of HIV infection. A key contribution of this study has also been to describe the limitations of analyzing service data to measure outcomes of PMTCT programs.
Analysis of service data from the sdNVP based PMTCT interventions in rural Malawi provides some limited indication of reduced risk of MTCT [15
]. In this context of high breastfeeding rates, without any intervention, HIV-free survival at 24 months postnatal could be as low as 55% [17
]. In our data, we estimated the 24-month HIV-free survival probability to be 78% overall and 82% when the infant dose of NVP was also received. However, due to the sparseness of our data, the confidence intervals around our estimates are very wide and each spans 55%; thus, our data do not allow for any firm conclusions.
At the time the program was implemented, the sdNVP only regimen was offered in Malawi as the predominant option for PMTCT, especially in rural areas. More recently, Malawi-as with other Sub-Sahara African countries-is transitioning to more efficacious regimens. However, even in the context of more efficacious regimens, our analysis to inform HIV free survival remains relevant; a PMTCT program offering more complex ARV regimens still needs to assess the outcomes of the intervention especially when adherence to treatment might decrease with long term and complex regimens.
Moreover, this analysis revealed how a “minimalist” PMTCT approach-sdNVP only antiretroviral regimen-results in a relatively high proportion (around 22%) of HIV-exposed children dead or HIV-infected.
Detection of HIV infections in the study population was essentially based on HIV antibody tests and could not have occurred earlier. A proportion of early deaths could have been avoided if HIV diagnosis were established by 12 months of age and pediatric antiretroviral treatment initiated in a timely manner [3
]. If not diagnosed and treated, approximately one-third and one-half of HIV-infected infants will die before the age one year and two years, respectively [18
]. Improved coverage, more effective ARV regimens, and early infant diagnosis could have prevented a substantial proportion of these HIV infections and deaths among HIV-exposed children.
This analysis has limitations. First, most of the children, including those who tested HIV-negative, were still exposed to HIV through breastfeeding-99% of children had been breastfed-and this analysis did not adequately capture the definitive HIV status of these HIV-exposed infants.
Second, establishing diagnosis of HIV-infection requires HIV testing at two different points in time, children were tested only at one time [14
]. Reasonable conclusions, however, may still be drawn as the HIV test in this circumstance was not meant for diagnosis but for surveillance.
Third, our analysis could not ascertain that all children who were not alive had died of HIV infection; the registers used during home visits did not capture verbal autopsy information.
Next, our analysis likely overestimates the numbers of HIV infected infants. Infants who tested positive for HIV antibodies at
18 months of age, were considered HIV-infected in this analysis, even though conclusive HIV infection status was unknown.
Finally, this study demonstrates the limitations of analyzing routine service data. Despite efforts to maintain good data quality, routine service data are not comparable to data collected in formal research studies with rigorous protocols for data collection [19
]. For example, two-thirds of HIV-exposed children that received ARV prophylaxis could not be included in the analysis because data on their mother’s address were not registered. It is not possible to determine if the rate of HIV-free survival among this group is comparable to those who were followed up through home visits. During program implementation, the practice of documenting the address of mothers was not regularly conducted, which prevented the follow-up of a large proportion of HIV-exposed children and their mothers. This analysis highlights the need to have good quality service data-from both health facilities and home visits. As the quality of service data improves, so does its ability to measure HIV-free child survival rates.
Moreover, among the large proportion of mothers for whom home addresses were not available, there was not a significant difference in their age, parity, or distance between home and the clinic. Despite that there is no existing evidence of bias between the mothers who were LTFU versus those who were tracked through home visits, it cannot be completely ruled out.
Lastly, the study population does not represent a random sample, therefore the results may not be generalizable to a larger population. A greater sample size in a representative number of sites, availability of defined age at the time of HIV testing, as well as documentation of the number and results of previous HIV tests would have allowed more detailed analysis and possibly firmer conclusions.