This facility survey was conducted in sites at which we already knew the PMTCT coverage level from the previously-published cord blood study 
. The exercise thus presented a unique opportunity to both describe facility factors associated with PMTCT coverage and to begin to validate the survey tool as a predictor of program performance. To our knowledge, this is the first study to examine in a systematic way the relationship between antenatal clinic and service characteristics and an objective outcome such as PMTCT coverage.
Many variables were associated with higher coverage, both individually and in two of the aggregate scores. Some, including availability of CD4 testing and infant PCR testing were intuitively satisfying and agree with our preconceived notions about how programs evolve to be more sophisticated and more successful. We did learn a few new things, however, including that variables pertaining to general antenatal care were even more predictive than PMTCT variables, and that the single factor distinguishing all of the worst-performing sites from the others was the lack of registers with PMTCT information. The predictive value of the non-PMTCT variables, and the lack of predictiveness of several PMTCT variables, reminds us that complex PMTCT services cannot be expected to function well where overall services are poor, even if all the elements of PMTCT are provided.
Overall, we concluded that this tool has limited but not zero utility as a method of evaluating PMTCT sites. Certainly programs do need to know if the essential elements of PMTCT and antenatal care are in place, and common sense dictates that PMTCT requires the presence of drugs, trained staff, and properly stored test kits. However, presence of these elements does not necessarily ensure successful implementation, and answers obtained in this simple checklist-based survey and patient exit interview do not necessarily reflect the experiences of patients visiting the facilities. The overall low coverage of NVP in our study sites (55%, range 33% to 68%) indicates that something is amiss in these PMTCT sites. The aggregate PMTCT score did have some relationship with coverage overall, but not in all countries, and the PMTCT scores were relatively homogeneous compared to the very large observed differences in coverage. Provision of longitudinal medical care and the population's receptiveness to this medical care is a complex process dependent upon attitudes, beliefs, trust, relationships, and complex elements of clinic flow and individual interactions. Most of this cannot be adequately described using simple inventories such as this one, even with direct observation of some elements of care.
Several variables that we expected to be associated with higher PMTCT coverage were not, and on close analysis of the questions and data we appreciated weaknesses in the survey tool that are important to help guide future program evaluation efforts. For example, we asked clinic managers if their site conducted “opt-out” HIV testing. Though several studies have shown that opt-out testing increases uptake 
. There was no association between opt-out testing and higher PMTCT coverage in our study. Asking clinic managers was probably not the right approach. In an operational sense, opt-out testing is defined not by the policy in place, but by the details of how testing is explained to clients and how they, their laboratory forms, and their blood move through the clinic during their visit. A checklist survey approach does not lend itself to an adequate description of how this occurs, and it is doubtful that we truly distinguished between opt-out and opt-in approaches. There are other examples as well. We found that availability of antiretroviral therapy (ART) at the clinic did not predict PMTCT coverage. Simply having ART at the same site does not tell us how well the ART and antenatal clinics work together to ensure that all eligible women receive ART during pregnancy. The details of how
the services are integrated will determine successful provision of ART in pregnancy. Meaningful exploration of these issues requires a different type of evaluation.
Our study has several recognized limitations. Since we decided to include only sites that provided both antenatal and delivery services, we included a relatively small number of facilities across four countries. The small sample size may have impacted the statistical power to detect associations between our composite scores as well as individual variables and PMTCT coverage. Second, with only six consecutive observations per facility selected, it is possible that the result for any one site might not be representative of that site's performance (sampling error). This could lead to exposure misclassification and undermine the ability to observe a true association.
Our results broadly support the general principle of strengthening the health care system as an important strategy for improving PMTCT coverage. This analysis exercise provides an important reminder for programs that service provision is complex, and that provision of drugs, test kits, policies, and training does not ensure program success. It is also an important caution to program evaluators that even this carefully conceived and detailed site survey could not predict which sites did well in a very useful way. Once basic elements of PMTCT are in place, detailed clinic-level quality improvement and problem solving initiatives that focus on operational factors are probably just as essential. Further study surrounding quality improvement methods and their impact on coverage and outcomes is warranted.
Members of the Pearl Study Team
Cameroon: Pius Tih (Cameroon Baptist Health Convention Health Board [CBCHB]); Thomas Welty (CBHC); Allison Spensley (Elizabeth Glaser Pediatric AIDS Foundation [EGPAF]), Christophe Grundman (EGPAF); Catherine Wilfert (EGPAF and Duke University).
Cote d'Ivoire: Didier Koumavi Ekouevi (Programme PAC-CI, Abidjan, Cote d'Ivoire and Université Victor Segalen, Bordeaux, France); Francois Dabis (Université Victor Segalen, Bordeaux, France); Serge Kanhon (Ministry of Health, Abidjan, Cote d'Ivoire).
South Africa: David Coetzee (University of Cape Town, Cape Town, South Africa [UCT]); Kathryn Stinson (UCT); Peter Smith (UCT); Andrew Boulle (UCT); Felicity Gopolang (UCT).
Zambia: Elizabeth M. Stringer; Jeffrey S A Stringer (Centre for Infectious Disease Research in Zambia [CIDRZ]); Benjamin H Chi (CIDRZ), Namwinga Chintu (CIDRZ); Mark Giganti (CIDRZ); Jessica Joseph (CIDRZ), Maximilian Bweupe (Zambian Ministry of Health); Nande Putta (CIDRZ); Alain DeGroot (CIDRZ), Humphrey Mulenga (CIDRZ); Wendy Z Mazimba (CIDRZ); Andrew O. Westfall (CIDRZ), Marc Bulterys (CDC-Zambia); Lawrence H Marum (CDC-Zambia); and Charles Cowan (Analytic Focus, LLC).
US Centers for Disease Control and Prevention: Tracy Creek (National Center for HIV, STD, and TB Prevention, Global AIDS Program, Atlanta Georgia).
World Health Organization: Nathan Shaffer (HIV Department, Geneva; formerly with CDC).