In 2003, the University of North Carolina at Chapel Hill (UNC) initiated its HIV prevention, care, and treatment programs in the DRC (UNC-DRC) by strengthening Kinshasa ANC/L&D sites through the introduction of a minimum package of maternal and infant services, including HIV counseling and testing and PMTCT prophylaxis.²⁵ This package, instituted in June 2010 in 37 of the city’s highest volume facilities, has been described previously.^24–26

In April 2010, we began rolling out an expanded package to these sites, and 53 others (for a total of 90) were added by September 2012 to cover 32/35 provincial health zones. The prior standard of care was bolstered by training personnel on the 2010 WHO PMTCT guidelines²⁷ that includes Option A, empathetic counseling skills, co-located delivery care, and the importance of patient retention, paying delivery fees to encourage safe birth, and providing (1) CD4 testing, cotrimoxazole, and AZT for mothers, and (2) cotrimoxazole, extended nevirapine, and DNA PCR testing for infants. Transportation costs to the first visit at either of 2 affiliated (but separately located) HIV C&T centers were provided to promote long-term care uptake; however, once the new interventions were available at ANC/L&D sites, women could opt to remain there for PMTCT and HIV-exposed infant services. At 45 sites, volunteer HIV-infected “mother-mentors” worked as clinic assistants, helping women to navigate care while encouraging C&T uptake. Generally, women with CD4 counts ≤350 could access ART at C&T centers only. Details on the ANC/L&D sites are outlined in Table S1 (see Supplemental Digital Content, http://links.lww.com/QAI/A726).

An essential expanded package component was a new paper register that allowed for mothers and their infants to be linked and longitudinally tracked. Service delivery and patient dispositions were recorded in the registers by the ANC/L&D site providers and UNC-DRC personnel, who made supportive supervisory monitoring visits monthly as part of the expanded package. Data were recorded between a woman’s presentation for ANC or L&D, until her infant was confirmed as HIV-infected or ruled out as HIV-uninfected. Details on the populations of mothers and infants tracked in the registers are presented in Table S2 (see Supplemental Digital Content, http://links.lww.com/QAI/A726).

The study sites included the 90 ANC/L&D sites and 2 UNC-DRC C&T centers, Kalembe Lembe Pediatric Hospital and Bomoi Healthcare Center.^28,29 The data sources were the registers at the ANC/L&D sites, which were entered into an Epi Info database, the real-time Epi Info database at the C&T centers,³⁰ and an aggregate service delivery database from the ANC/L&D sites used to provide program context and quantify decentralization’s population-level impacts. The Epi Info databases contained unique patient identifiers to link mother–infant pairs and were linked to each other, as the ANC/L&D database included C&T codes and the C&T database included ANC/L&D codes. This facilitated tracking of individuals between facility types. All database records as of July 2013, when data collection ended, were included in analyses.

In the service delivery aim, we calculated the cumulative incidence of 2 outcomes between 2010 and 2013: (1) receipt of CD4 testing, and AZT or ART (pregnancy package), by women newly diagnosed with HIV before L&D, and (2) receipt of DNA PCR testing and cotrimoxazole (infant package) by HIV-exposed infants, at 2 facility types: (1) ANC/L&D site, providing decentralized services, and (2) C&T center, specialized to deliver these services. In the analysis of women, follow-up began at C&T enrollment if the pregnancy was registered at a site before it began offering CD4 testing and AZT; otherwise, follow-up began at ANC presentation. Follow-up ended at the earliest of (1) the outcome, (2) censoring (loss to follow-up, transfer, move, voluntary withdrawal, or end of study), or (3) a competing risk (delivery, miscarriage, or death). In the infant analysis, follow-up began on the date of first visit at ≥1 month of age (when cotrimoxazole is indicated³¹) at a facility offering DNA PCR testing and cotrimoxazole. Additionally, reaching 18 months of age (the end of the early infant diagnosis period, when HIV can be serologically confirmed) was an additional censoring event, and competing risks were limited to death and confirmation of HIV-negative status.

As individuals could opt to receive services at a C&T center, exposure could switch from “decentralized” to “specialized” (but not vice-versa) if this transfer occurred before either component of the intervention package was received; if 1 of the outcome services was received, followup was censored on the transfer date (date of C&T enrollment). To allow for this late entry, we used the counting process configuration as applied by Geskus.³² Individuals who received the intervention package on their enrollment date and those lost after a single visit³³ were assigned a follow-up time of 1 day. For others censored, if deemed lost to follow-up after a missed visit, follow-up ended on the last visit date. Only newly HIV-diagnosed women were eligible because the outcome included ARV receipt, and many previously diagnosed women were already on ART; additionally, time in care was likely a confounder and modifier of the associations of interest. Infants were eligible for the infant analysis even if their mothers were not in the analysis of women. DNA PCR testing was defined as specimen collected, not also receipt of result.

Competing risks are events such as death that preclude occurrence of the outcome and can create bias.³⁴ To account for competing risks in our data, the SAS macro %PSHREG was used to obtain competing risk cumulative incidence functions stratified by facility type (decentralized versus specialized) and to quantify the effects of facility type on delivery of the intervention packages.³⁵ Crude and adjusted hazard ratios (HRs) and 95% confidence intervals (CIs) were generated using the Fine and Gray subdistribution hazards model,³⁶ which produces an effect measure due to both the association of the exposure on the event of interest and the possibly differential impact of competing events on exposed and unexposed individuals.³⁷ We also employed the cause-specific Cox proportional hazards model, which treats competing events as censored, and results in biased effect estimates if its inherent assumption of independence between the outcome and competing events is incorrect.³⁷

Causal diagrams were used to identify confounders for adjusted models.³⁸ In the analysis of women, covariates were age and gestational age at registration along with CD4 count during pregnancy; infant analysis covariates were age at first visit and 3 maternal factors: gestational age, HIV status (undiagnosed or previously diagnosed) at registration, and CD4 count during pregnancy. We employed multiple imputation with fully conditional specification methods³⁹ to account for missing covariate data; 5 imputed datasets were generated. To relax the linearity assumption in modeling continuous variables, we used Stone and Koo’s additive splines constrained to be linear in the tails, with knots at the fifth, 35th, 65th, and 95th percentiles. χ² or Cochran-Armitage trend tests were used to compare proportions, with medians compared by the Mann–Whitney test.⁴⁰

In the aim to quantify decentralization’s population-level impacts, we calculated the proportions of HIV-infected women who received the pregnancy package pre-decentralization (at affiliated C&T centers only) versus post-decentralization (at either the C&T centers or ANC/L&D sites). Similarly, proportions of their HIV-exposed infants who received the infant package, pre- and post-decentralization, were calculated. All analyses were completed in SAS 9.3 (SAS Institute Inc., Cary, NC).

The study was approved by the Kinshasa School of Public Health Ethics Committee and the UNC Institutional Review Board. Consent was obtained at C&T enrollment; activities at the ANC/L&D sites met criteria for waiver of consent.

Between 2010 and 2013, in our cohort of linked mother–infant pairs, we identified 1482 newly HIV-diagnosed women eligible for the analysis of CD4 testing and ARV regimen delivery during pregnancy, as well as 1142 HIV-exposed infants eligible for the analysis of DNA PCR testing and cotrimoxazole delivery. Their characteristics and outcomes are described in Table 1. Women at a C&T center did not differ from those at only a ANC/L&D site in terms of age (P = 0.88) or gestational age (P = 0.05) at registration. The cumulative incidence of receiving the pregnancy package at 14 weeks was less at the ANC/L&D sites (66%; 95% CI: 63% to 69%) than that at the C&T centers (88%; 95% CI: 83% to 92%) (Fig. 1), and both the unadjusted (0.59; 95% CI: 0.51 to 0.69) and adjusted (0.62; 95% CI: 0.55 to 0.69) subdistribution HRs indicate inferior delivery of the package at the ANC/L&D sites compared with the C&T centers (Table 2). No marked changes in delivery of the package over time at either facility type were noted.

Compared with infants at a C&T center, those only at an ANC/L&D site were older at first visit (P < 0.01) and more likely had mothers who were previously HIV undiagnosed (P = 0.01), of greater gestational age (P < 0.01) at registration, healthier as reflected by pregnancy CD4 count (P < 0.01), and did not receive AZT or ART (P < 0.01). The cumulative incidences of receiving the infant package at 10 months (Fig. 1) at the ANC/L&D sites (89%; 95% CI: 82% to 93%) and the C&T centers (97%; 95% CI: 93% to 99%), as well as the unadjusted (0.85; 95% CI: 0.75 to 0.97) and adjusted (0.84; 95% CI: 0.76 to 0.92) subdistribution HRs (Table 2), suggest that delivery of the package at the ANC/L&D sites was poorer than that at the C&T centers. No marked changes in delivery of the package over time at either facility type were noted. Using age rather than months in care as the timescale, the median age at package receipt at the ANC/L&D sites was 7.3 weeks (interquartile range, 6.6–11.7), similar to that at the C&T sites (7.0 weeks; interquartile range, 6.4–8.9).

Only 14.2% of newly diagnosed cases remaining at the ANC/L&D sites received all 4 components of the pregnancy and infant packages, markedly lower than the 53.5% observed among cases at the C&T centers (Fig. 2). Altogether, just 21.1% of total newly diagnosed cases received all 4 components of the packages, which emphasizes that overall impact was driven more by the ANC/L&D sites where most individuals remained for care.

During the pre-decentralization period of 2009–2010, considering all HIV-infected pregnant women identified at ANC/L&D sites including those not presenting to C&T (N = 1518), package uptake at affiliated C&T centers was approximately 20% (pregnancy) and 16% (infant). In 2011–2013 when services were decentralized (N = 1405), uptake across facilities increased to about 64% (pregnancy) and 31% (infant). These increases are depicted in Fig. 3. During the period of decentralization, 1260 women and 634 infants received at least 1 intervention at an ANC/L&D site; no temporal changes in pregnancy and infant package uptake were noted. Greater proportions of women (P < 0.01) and infants (P < 0.01) remained at ANC/L&D sites in later years (Table 1), reflecting the increased dissemination of decentralized services over time.