In this study we examined trends in PMTCT service utilization and assessed the rate of MTCT in Addis Ababa, Ethiopia. The HIV counselling and testing service utilization improved substantially in 2009 following policy shift to routine opt-out approach. The HIV prevalence appeared to decrease steadily paralleling the increased number of women tested for HIV. Irrespective of policy changes, the uptake of ARV prophylaxis and loss to follow up remained unimproved. Out of the 10.6% (896) exposed babies tested for HIV, the cumulative probability of HIV infection decreased to 15.0% by 2007 among babies on sdNVP regimen tested at >=18 months of age, and to 8.2% by 2009 among infants on ZDV regimen tested at >=45 days of age. The proportion of HIV positive pregnant women referred for treatment, care and support increased eighteen-fold in 2009 compared to 2004. Meanwhile, the proportion of partners tested in the PMTCT setting declined significantly by 14% from 2004 to 2009.
The proportion of women who received HIV counselling and testing among new ANC attendees increased significantly from 50.7% in 2007 to 84.5% in 2009 following the shift to routine opt-out testing. Consistent with our findings, studies from other resource poor settings have revealed significant improvement in HIV testing at ANC from 45% to 99% following the shift to routine opt-out approach [18–20]. Nevertheless, over 10% of the new ANC attendees had not received pre-test counselling or testing in 2009. In this regard the opt-out approach, which seems to be robust to maximise HIV testing in many settings [18–20], did not show remarkable success in our study. This could be attributed to two major reasons. First, most of the studies that reported nearly a 100% HIV test acceptance when testing was offered routinely as an opt-out approach, were pilot initiatives [18–20], unlike our study that compiled PMTCT reports from a national programme. In line with our argument, a study in Kenya based on data from 43 PMTCT service outlets reported a 80.6% HIV test acceptance . Second, the expansion of the PMTCT programme to private facilities seem to have contributed to the persistence of a large proportion of women who had not received HIV counselling or testing even after the shift to routine testing. Site factors are reported to be more relevant than participant factors in determining HIV test acceptance in a study from Kenya . In particular, characteristics of the provider appear to be an important determinant for HIV test acceptance. For instance, characteristics of the midwives were found to be an independent determinant for HIV testing uptake in England , whereas high test refusal was associated with HIV testing being offered by general practitioners in Canada [21, 23]. In almost all the private facilities in Addis Ababa, ANC and HIV testing was offered by physicians who had little or no training on HIV counselling and testing in an opt-in approach. This indicates that there are gaps in the implementation of the routine opt-out testing strategy, particularly in private facilities. On top of that, the utilization of post-test counselling and collection of test result by almost all tested women throughout the years, irrespective of the testing approach employed, implies that the persistent gap in pre-test counselling and HIV testing service utilization could be an evidence of failure of the health system to deliver the programme rather than a failure of the women.
The HIV prevalence among the tested women had reduced by 54% in 2009 compared to the level in 2004, in parallel to the increased number of women being tested. Similar declining trend in HIV prevalence has been observed among adults and from a sentinel surveillance report in Ethiopia. Synergy between the natural progression of the HIV epidemic and behaviour changes among the general population in terms of increased condom use and reduced risky sexual behaviour seem to contribute to the declining prevalence [3, 4]. The HIV prevalence estimate in our study was lower than that of the ANC sentinel surveillance report, i.e 6.2% vs 9.3% in 2007/2008 . Consistent with our finding, a study that compiled data from Kenya, Ethiopia, and Zimbabwe for 2005 reported that the HIV prevalence in Ethiopia was 6.4% from PMTCT programme reports and 8.2% from sentinel surveillance reports [24–26]. The two reports, although generated from the same population, have differences in population size and timing of data collection [24–26]. The PMTCT report includes all women who participated in the PMTCT programme all year round. In the sentinel surveillance, leftover blood samples were collected from a smaller number of ANC attendees from selected facilities, often for a period of 3 months biennially.
The decreasing trend in the HIV prevalence estimate across the years could also be attributed to a more representative sample of pregnant women tested for HIV in our study. In other words, the inclusion of more service outlets and therefore reaching out to a large population in our study seems to be more representative, unlike the sentinel surveillance report that relied on reports from a few public facilities. Even in the sentinel surveillance reports, as the number of sentinel sites increased the estimated HIV prevalence declined . More importantly we included reports from private facilities where the HIV prevalence is reported to be lower compared to ANC attendees in public facilities, 2.4% (unpublished report from Addis Ababa City Administration Health Bureau) vs 9.3%  respectively in 2007/2008. Moreover, the HIV prevention potential of Highly Active Antiretroviral Therapy (HAART) should not be ignored .
Despite the marked increase in HIV testing following the shift to routine opt-out approach, neither the proportion nor the number of women receiving ARV prophylaxis have increased. In 2009, only 53.7% of the women and 40.7% of their babies received ARV prophylaxis. Consistent with our findings, in Addis Ababa, 49.3% of the HIV positive women and 35.3% of their babies received ARV prophylaxis in 2007/2008  and the programme achieved only 30% compared to 80% as the target . Actually, poor ARV prophylaxis uptake is not a new story, even those intervention studies that show almost 100% successes in HIV testing were in short of ARV prophylaxis uptake [18–20]. However, the lack of improvement in ARV prophylaxis utilization following the shift to routine testing is an issue of great concern as the shift to an opt-out approach was primarily intended to increase the proportion of women and infants receiving prophylactic ARV drugs . In the era of routine testing for HIV, the PMTCT programme appears to be an effective screening programme than a prevention programme due to the large number of dropouts after testing. Yet, the gain in HIV testing turnout limits the weakness of the routine opt-out approach in terms of subsequent dropouts and poor adherence to ARV prophylaxis . Currently the availability of potent ARV prophylaxis is changing the landscape, and even mother-to-child HIV transmission through breast feeding has become less of a concern . Yet, still more lives are at stake because of the lack of improvement in uptake of this critical component of the PMTCT programme.
According to Kasenga et al., skilled attendance at birth is an important determinant of ARV prophylaxis uptake that requires thorough consideration . In Addis Ababa, only 30% of the pregnant women had skilled attendance at birth in 2008  corresponding to the proportion of infants who received ARV prophylaxis . Since infants are given prophylaxis within 72 hours of birth at the health facility, infants delivered at home have less chance to receive ARV prophylaxis than infants delivered in health facilities. An intervention study from Zambia gives some hope that dropouts and non-adherence to ARV prophylaxis could be reduced to zero. This study, which employed multiple interventions, increased the ARV prophylaxis uptake from 29% at baseline to 100% within 3 years . This gives reassurance that the official 80% ARV prophylaxis uptake goal for Ethiopia could be achieved through concerted effort and renewed commitment.
According to the findings, an increasing proportion of HIV-positive pregnant women were referred for treatment, care and support services, from 3.2% in 2004 to 59.9% in 2009. The referral is intended for prompt initiation of ARV prophylaxis or treatment for eligible women using CD4 or lymphocyte count and WHO staging criteria . By doing so, the programme is addressing the most important issues that PMTCT programmes have been criticized for, i.e the lack of sensitivity to the needs of pregnant women. The PMTCT programme is increasingly successful in bridging the gap between prevention and treatment to address the moral and ethical questions raised over the years . The eighteen-fold increase in the proportion of positive pregnant women being referred for treatment, care and support from 2004 to 2009 demonstrates the good will and the potential to integrate new policies and strategies. Yet, more has to be done to ensure that all HIV-positive pregnant women have access to prophylaxis or treatment, care and support.
The ultimate objective of a PMTCT programme is to avert new HIV infections among children. However, only 10.6% (896) of the HIV positive pregnant women completed their follow up to infant HIV testing. The rates of MTCT were therefore assessed based on test result of the 896 babies tested for HIV. In the absence of any PMTCT intervention the cumulative probability of HIV infection among exposed babies aged >=18 months is in the range of 29% to 47% according to a cohort study conducted in an orphanage in Addis Ababa . In our study, the cumulative probability of HIV infection among exposed babies on sdNVP regimen tested at >=18 months was 14.3% in 2006 and 14.9% in 2007. According to the HIVNET 012 randomized trial, sdNVP regimen has a 41% efficacy. In this trial, consistent with our findings, the cumulative probability of infant HIV infection is 15.7% among breast fed infants tested at >=18 months . However, a methodologically similar study from Malawi that compiled monthly reports showed a 15.5% HIV infection rate among infants on sdNVP regimen tested at 6 weeks postpartum without accounting for the breast feeding transmission . Nevertheless, there is a possibility that our estimate could be biased due to the large loss to follow up.
In 2009, 8.2% of the exposed infants on ZDV regimen tested at 45 days were HIV positive. In the Petra clinical trial, multidrug ZDV regimen showed 63% efficacy in reducing MTCT. In this trial the rate of HIV transmission among infants on multidrug ZDV regimen tested at 45 days was 5.7% . Considering the fact that our data are generated from a national PMTCT programme and the obvious methodological difference with the Petra trial, the 8.2% infant infection rate reported in our study indicates the success of the national PMTCT programme among those who completed their follow up to infant HIV testing. A cohort study from similar resource poor settings that evaluated the effectiveness of a PMTCT programme among predominantly formula fed infants on ZDV regimen tested at >=45 days reported a 9.1% cumulative infant HIV infection, higher than the rate of HIV infection reported in our study . However, since the HIV testing was done at >=45 days, those HIV negative infants who continue to breast feed are still at risk of acquiring new infection. In general, the rate of MTCT averted by the national PMTCT programme appears promising among those who adhered to the programme. Nevertheless, the possibility of underestimation cannot be excluded since we lack information on loss to follow up. In line with this limitation, Ahoua et al. found that the cumulative probability of infant HIV infection among tested infants was 8.3%, whereas it was 15.5% when HIV related deaths were included in the analysis .
The last, but not least important PMTCT programme outcome indicator examined in our study was partner testing. The proportion of partners tested remained very low with a 14% significant decline in 2009 compared to 2004. A review on couple centred counselling shows that partner involvement in HIV testing not only helps to increase disclosure, condom use and uptake of ARV prophylaxis but also contributes to the lower rate of seroconversion compared to individual counselling . We noted a parallel trend in the number of partners tested with the number of women who tested positive across the years. This indicates that most of the partners who came for HIV testing were those whose wives tested positive. Partner testing in the context of PMTCT seems to facilitate women's coping, yet missing out the important prevention aspect by not advising HIV negative women to bring their partner for testing.
A study from South Africa shows that 3% of the pregnant women who were found to be HIV negative in their first HIV testing during pregnancy became HIV positive in repeat test in late pregnancy, giving a 10.7% incidence per year . This indicates that women are at risk to acquire new HIV infection from their HIV positive partner anytime during pregnancy and even during breast feeding. In eastern and southern African region, including Ethiopia, 36-85% of HIV positive individuals are believed to live with an HIV negative partner [4, 10]. Discordant couples are the newly identified high risk group in Ethiopia, as most infections are occurring within marriage. Because of mucosal and hormonal changes during pregnancy, the HIV incidence is four times higher among pregnant women compared to their non-pregnant counterparts . Meanwhile, women having recent HIV infection are more likely to transmit HIV infection to their babies . Therefore, it is crucial to focus on partner testing and involvement in the PMTCT programme to optimise programme effectiveness. The current strategies in Addis Ababa, that include giving priority for women coming with their partner for testing and sending an invitation home to partner should be encouraged.
One of the limitations of our study is that the rate of MTCT was examined based on infants tested for HIV, which could actually be underestimated due to the large loss to follow up. Considering the lack of a system to trace loss to follow up, our finding still highlights the potentially averted infections. Another limitation is that the PMTCT programme reached out to only 80% of pregnant women due to incomplete ANC attendance  and the findings seem not to represent the whole nation, where a high proportion of the population is rural. However, since it was generated from a national programme, lessons learnt herein could benefit the PMTCT programme across the country. We also believe that our findings are generalisable to the big cities where the HIV prevalence is higher.
Retrospective data collected primarily for reporting purposes always have weakness, especially in resource poor settings where the quality of reports are often questionable . By limiting our study objectives to those indicators that could be calculated from the reports we minimized the risk of having incomplete data. We obtained almost all the reports from February 2004 to August 2009 from the 10 sub-cities. To validate our data we checked Addis Ababa City Administration Health Bureau and NGO reports, and our data were found to be consistent. For missing reports we obtained data from log books and reports at service outlets.