What does high and low have to do with it? Performance classification to identify health system factors associated with effective prevention of mother-to-child transmission of HIV delivery in Mozambique

Introduction: Efforts to implement and take to scale highly efficacious, low-cost interventions to prevent mother-to-child HIV transmission (pMTCT) have been a cornerstone of reproductive health services in sub-Saharan Africa for over a decade. Yet efforts to increase access and utilization of these services remain far from optimal. This study developed and applied an approach to systematically classify pMTCT performance to identify modifiable health system factors associated with pMTCT performance which may be replicated in other pMTCT systems. Methods: Facility-level performance measures were collected at 30 sites over a 12-month period and reviewed for consistency. Five combinations of three indicators (1. HIV testing; 2. CD4 testing; 3. antiretroviral prophylaxis and combined antiretroviral therapy initiation) were compared including a composite of all three, a combination of 1. and 3., and each individually. Approaches were visually assessed to describe facility performance, focusing on rank order consistency across high, medium and low categories. Modifiable and non-modifiable factors were ascertained at each site and ranking process was reviewed to estimate association with facility performance through unadjusted Chi-square tests and logistic regression. After describing factors associated with high versus low performing pMTCT clinics, the effect of inclusion of the 10 middle performers was assessed. Results: The indicator most consistently associated with the reference composite indicator (HIV testing, antiretroviral prophylaxis and combined antiretroviral therapy) was the single measure of antiretroviral prophylaxis and combined antiretroviral therapy. Lower performing pMTCT clinics ranked consistently low across measurement strategies; high and middle performing clinics demonstrated more variability. Association between clinic characteristics and high pMTCT performance varied markedly across ranking strategies. Using the reference composite indicator, larger catchment area, higher number of institutional deliveries, onsite CD4 point-of-care capacity, and higher numbers of nurses and doctors were associated with high clinic performance while clinic location, NGO support, women’s support group, community linkages patient-tracking systems and stock-outs were not associated with high performance. Conclusions: Classifying high and low performance provided consistent results across ranking measures, though granularity was improved by aggregating middle performers with either high or low performers. Human resources, catchment size and utilization were positively associated with effective pMTCT service delivery. © 2014 Gimbel S et al; licensee International AIDS Society.

Activities were conducted as part of a multi-methods, cross-sectional study designed to identify modifiable and non-modifiable health systems factors associated with pMTCT service performance. The results presented here contributed to the parent study which will be conducted in three sub-Saharan countries by developing the health facility performance ranking approach as a first step for identifying systems-level factors associated with high and low-performing clinics. Thirty public sector health facilities with pMTCT services in three districts along the Beira corridor (the main transport route connecting the port city of Beira with Zimbabwe) in Sofala province, central Mozambique, were included in the study. Sofala has an estimated population of 1.8 million of which 47% live in the study districts of Beira city, Dondo and Nhamatanda [32]. Sofala province has an estimated adult HIV prevalence of 15.5% [33], which has been consistently higher among women routinely tested for HIV in ANC (17.8%) [34]. Since its introduction in 2002, pMTCT expansion has increased to reach 100% of all public sector clinics with ANC services in the three study districts (and 86% of all facilities nationally) [35]. The gaps in the pMTCT cascade limit its effectiveness resulting in approximately 28% HIV infection rate in infants born to HIV-positive women in 2012 despite high availability of pMTCT and approximately 95% ANC attendance rate [36]. Study facilities included all public sector health facilities that met the inclusion criteria of: 1) location in the three districts; 2) provision of pMTCT at ANC in the last six months; and 3) consent to participate in the study. Two facilities were excluded because they were unwilling to participate or had limited access due to flooding. Clinic data for pMTCT performance ranking were sourced from provincial program reports covering January–December 2012. Data were based on monthly health facility data, which in parallel is entered into the national health information system at the district level. Monthly facility-level data were assessed for availability by the study team to identify missing reports, and irregular or missing data were crosschecked with facility-level registries to ensure accuracy. Missing data were recovered for all measures except cluster of differentiation 4 (CD4) testing data, which were inconsistently available at the facility-level. Data on health facility characteristics were collected using a survey developed for study purposes from November 2012 to January 2013 (Supplementary file). Descriptive, facility-level variables were identified through reviews of published literature on pMTCT and quality improvement, and the final list of facility characteristics was developed in consultation with provincial program managers and technical advisors [37–39]. The data collection form was developed in Portuguese and piloted in one facility before study assistants visited all 30 health facilities to collect information from facility managers and front line health workers. Data from each facility were double entered by study personnel to ensure their accuracy. Three performance measures were a priori selected based on their presence in routine reporting systems and importance for successful pMTCT, including: 1) the proportion of pregnant women in ANC tested for HIV at their first visit; 2) the proportion of pregnant women with a positive HIV test at first ANC visit who had a CD4 test in pregnancy; and 3) the proportion of women with a positive HIV test in the first ANC visit accessing effective PPO or cART in pregnancy (Table 1). HIV testing in the first ANC visit was selected (versus testing at any time during pregnancy) because testing in the first ANC is standard as per Ministry of Health (MOH) guidelines, and because an effective pMTCT package should initiate as early in pregnancy as possible [40]. Single-dose nevirapine was not included as an effective pMTCT PPO in the third measure because of its relative ineffectiveness in preventing HIV transmission [41]. Ranking measures ANC: antenatal care; cART: combination antiretroviral therapy. The 12 months of pMTCT cascade data were used to develop summary performance scores for each facility (n=30). First, each of the three indicators were estimated over a 12-month period (calendar year 2012). These individual indicators were then multiplied to generate the composite indicators. Facilities were ranked into three performance categories (high/middle/low) based on tertiles in the distribution of performance outcomes, rather than using a specific performance threshold level. Visual assessment was then used with one measurement strategy (HIV testing and ARV treatment and ARV PPO) defined as the reference strategy. This strategy was selected as the reference as it was composed of complete data sets and represented multiple steps in the pMTCT cascade. In order to assess the value of comparing high versus low-performing facilities in identifying modifiable health facility characteristics associated with pMTCT performance, we recoded facilities into two additional performance groups – one joining middle and high performers (to compare with low performers), and a second joining middle and low performers (to compare with high performers). Non-modifiable facility characteristics considered included facility type, classified into three levels – quaternary/tertiary hospitals, secondary hospitals and primary health centres, as well as catchment population sizes which were provided by provincial and district authorities. Additional non-modifiable factors collected included geographic location which was defined as urban/peri-urban/rural based on their location in Beira, Dondo or Nhamatanda municipalities (urban), in the outlying neighbourhoods of Beira (peri-urban), or outside of Beira, Dondo, or Nhamatanda municipalities (rural) and year of pMTCT initiation which was provided by facility leadership. Modifiable facility level characteristics included staffing which was defined as the number of health workers of cadres most relevant to pMTCT service delivery (physician, non-physician clinician (NPC), maternal and child health nurse, general nurse, midwife, custodian, social worker and activist). Distance to a laboratory with CD4 capacity was also considered modifiable as it could hypothetically be changed if CD4 capacity was introduced through new machines or new PIMA technology. This was estimated using driving distances between health facilities provided by provincial authorities. Other modifiable factors measures to describe pMTCT organization included integration with laboratory services (number of days per week with CD4 blood draws, availability of on-site laboratory capacity for CD4 and other laboratory monitoring), pharmacy services (medicines distributed via pharmacy or ANC/maternity) and outpatient care (adult and paediatric patient referral and tracking for continuity). Also, modifiable measures of community linkages included the presence of a mothers’ support group, whether community activists carried out patient tracking and whether health workers performed regular community outreach. General management practices were measured by the frequency of staff meetings. A list of essential medicines, supplies and materials related to pMTCT service provision was included in the factors list to assess the availability of key items over the preceding three months, as well as the length of stock outages, and was confirmed using stock registries. These were all deemed modifiable as innovations could be introduced to improve them at the facility, district or provincial level. Finally overall PMTCT service utilization was captured through four patient volume measures over the six-months before data collection, including the number of ANC consults, the number of postpartum consults, the number of institutional births and the number of modern family planning methods distributed. To refine the ranking procedures, we first explored whether the rank performance order for the 30 study facilities changed according to performance measurement (including each of the three outcome measures alone, a composite indicator multiplying indicators one and three for each facility (HIV testing and effective PPO or cART), and a composite indicator multiplying indicators one, two and three for each facility (adding CD4 testing to the previous indicator). Visual assessment focused on the consistency of rank order across high, medium and low categories depending on ranking strategy, using the composite indicator of HIV testing and effective PPO or cART as the benchmark, with the top 10 performing sites shaded, the middle 10 in white and the bottom 10 dotted. Assessment of the impact of ranking process on facility-level characteristics associated with facility performance was carried out for three of the five ranking approaches, excluding CD4 testing data which were found to be less available and reliable. Unadjusted Chi-square tests for independence were performed to estimate the association between performance and health facility characteristics. For continuous variables, logistic regression for performance outcomes was used to quantify the magnitude and statistical significance of any associations with performance. After describing factors associated with high versus low-performing pMTCT clinics, we assessed how the inclusion of the 10 middle-performing clinics affected the list of factors significantly associated with differential pMTCT performance. For this analysis, we developed ordinal logistic regression models including continuous clinic characteristics found to be significant in the bivariate analyses, and examined the magnitude of the associations between three-level (low, middle, high) compared to two-level (low, high, excluding the 10 middle-performing clinics). Next, we aggregated low and middle-performing facilities to compare their characteristics to high-performing facilities, and then aggregated middle and high-performing facilities to compare their characteristics with low-performing clinics. This analysis ranked facilities using indicators one and three (HIV testing at first ANC and receipt of effective PPO or cART), as these measures were most available and measured multiple, essential steps for successful pMTCT. Data analysis was performed using Stata v11.2 (College Station, TX). Study procedures were approved by the Ethics Committee of the Mozambique MOH and were determined to be non-research by the University of Washington Institutional Review Board.

N/A

• Dima Health