Subnational variation for care at birth in Tanzania: Is this explained by place, people, money or drugs?

Background: Tanzania achieved the Millennium Development Goal for child survival, yet made insufficient progress for maternal and neonatal survival and stillbirths, due to low coverage and quality of services for care at birth, with rural women left behind. Our study aimed to evaluate Tanzania’s subnational (regional-level) variations for rural care at birth outcomes, i.e., rural women giving birth in a facility and by Caesarean section (C-section), and associations with health systems inputs (financing, health workforce, facilities, and commodities), outputs (readiness and quality of care) and context (education and GDP). Methods: We undertook correlation analyses of subnational-level associations between health system inputs, outputs, context, and rural care at birth outcomes; and constructed implementation readiness barometers using benchmarks for each health system input indicator. We used geographical information system (GIS) mapping to visualise subnational variations in care at birth for rural women, with a focus on service availability and readiness, and collected qualitative data to investigate financial flows from national to council level to understand variation in financing inputs. Results: We found wide subnational variation for rural care at birth outcomes, health systems inputs, and contextual indicators. There was a positive association between rural women giving birth in a facility and by C-section; maternal education; workforce and facility density; and quality of care. There was a negative association between these outcomes and proportion of all births to rural women, total fertility rate, and availability of essential commodities at facilities. Per capita recurrent expenditure was positively associated with facility births (correlation coefficient = 0.43; p = 0.05) but not with C-section. Qualitative results showed that the health financing system is complex and insufficient for providing care at birth services. Bottlenecks for care at birth included low density of health workers, poor availability of essential commodities, and low health financing in Lake and Western Zones. Conclusions: No region meets the benchmarks for the four health systems building blocks including health finance, health workforce, health facilities, and commodities. Strategies for addressing health system inequities, including overall increases in health expenditure, are needed in rural populations and areas of highest unmet need for family planning to improve coverage of care at birth for rural women in Tanzania. Our analysis used a health systems evaluation framework [22] modified to reflect care at birth services (Fig. 1), which outlines the essential components within each building block of the health system (inputs, outputs, outcomes and impact) tailored to care at birth. We adapted this health systems framework within the context of the national level Tanzania Countdown case study [7], incorporating the same outcome measures for this second analysis, and utilising best available data for the health systems building blocks. We describe contextual data for each region in Tanzania according to a priori known associations with care at birth [24, 25] We extracted 2012 census population data [26]. Subnational gross domestic product (GDP) was abstracted from the National Accounts 2000–2010 [27] and estimated per capita with census data [26]. The proportion of women with complete primary education or higher was abstracted from 2010 DHS data [10]. United Nations adjusted births and pregnancies data (2010) were obtained from a previous analysis [28] (Additional file 1) – we extracted total fertility rate (TFR), number of live births, number of rural births, proportion of all births by rural women, birth density (mean births per square kilometre) and rural birth density. We collected the best available data at subnational level in mainland Tanzania corresponding with each health system building block in the evaluation framework (Fig. 1), selecting proxy indicators according to data availability and reliability (Table 1). Health system tracer indicators and data sources related to coverage of care at birth for rural women in Tanzania Financial input indicators of recurrent (government) health expenditure from 2012/13 [29], 2007 average annual household Out of Pocket health expenditure (OOP) [30], and 2013/14 Official Development Assistance for MNH (ODA) [31] (Table 1) were obtained (Additional file 1). We calculated per capita expenditures [26], converting to 2013 USD using Bank of Tanzania conversion rates and World Bank deflators [32]. Community Health Fund (CHF) – a community based health insurance scheme – 2013 coverage data were obtained from National Health Insurance Fund (NHIF) reports [33] (Table 1). Health workforce density data for those cadres involved in MNH service provision were derived from the human resources data in the health country profile [34]. We assumed that Assistant Medical Officer, Assistant Nursing Officer, Medical Consultant, Medical Doctor, Medical Specialist, Nurse, Nurse Midwives, and Nursing Officer are capable of providing skilled birth care [35]. Health workforce densities were reported as a ratio per 10,000 capita and per 10,000 births. We estimated the total number of facilities providing basic and comprehensive emergency care at birth (health centres and hospitals) in each subnational region from data provided by Ministry of Health and Social Welfare (MoHSW) – now the Ministry of Health, Community Development, Gender, Elderly and Children [36]. Health facility densities were reported as a ratio per 10,000 capita and per 10,000 births. The proportion of all facilities with no stockouts of essential commodities (Additional file 1) were extracted from the 2014 quarter four national RMNCH scorecard using Health Management Information System (HMIS) data [37] as a measure of commodities supply. We used the proportion of women who attended ANC and subsequently recalled being informed of signs of pregnancy complications, as a proxy for quality of care [10]. For health service readiness we used percentage of all health facilities with improved water source (Additional file 1) from the Tanzania Service Provision Assessment Survey 2014-15 [38], in accordance with recent evidence associating water and sanitation with maternal mortality [39, 40]. Reliable health service utilisation data were not available at subnational level. Building on the national Countdown case study [7], we used 2010 DHS data [10] to calculate the proportion of all births (inclusive of C-section) occurring in a health facility (hospital, health centre or dispensary) – a proxy for our outcome of skilled birth attendance [41] – and the percentage of births by C-section – a proxy for our outcome of emergency obstetric care [42] (Additional file 1). Both outcomes are self-reported by women. We restricted analyses of outcome indicators to births by rural women based upon findings of the Countdown country case study: rural/urban disparity is the strongest inequity [7]. Additionally, 70 % of Tanzania’s population is rural [10], and the literature illustrates that urban women generally access facilities for births (82 %) [7, 10]. In 2012, four new subnational regions were demarcated (Additional file 1); thus we recalculated district-level data for several indicators (total population, total births, recurrent expenditure, total expenditure, CHF, ODA, health workforce, health facilities, and commodities supply) to ensure consistency with the subnational boundaries in place at the time of the 2010 DHS (Additional file 1) [10]. Bivariate correlation analyses were performed across all levels of the evaluation framework, using Stata 13.1. Less than 5 % chance was considered statistically significant. A correlation coefficient (CC) of greater than 0.80 was considered a very strong association, 0.60–0.79 a strong association, 0.40–0.59 a moderate association, and <0.40 a weak association adopted from recent literature [43] and considered within the context of this analysis. Choropleth and proportional maps were generated using Arc GIS 10.3 software [44] to illustrate subnational variations in: (i) rural birth density; (ii) births by rural women in a health facility; (iii) births by rural women by C-section; (iv) per capita recurrent expenditure; (v) health workforce density; (vi) health facility density; and (vii) health facilities availability of tracer drugs. Health facility and health workforce data were mapped using both population and births as denominators, taking into account recent recommendations from Gabrysch et al. [45]. Implementation readiness barometers developed by the Countdown Health Systems and Policies Technical Working Group [46] were drafted for each Tanzanian region based upon the WHO health system building blocks, to be overlaid with choropleth maps showing variation in proportion of births by rural women a) in a health facility and, b) by C-section. This approach was applied to identify “good” and “poor” performing regions and to assess subnational variation in care at birth, with a focus on service availability and readiness. Implementation readiness barometers were constructed using data from HMIS [37], Human Resources for Health Country Profile (2012/13) [34], 2012 Census [26], the Prime Minister’s Office for Regional Administration and Local Government (PMO-RALG, now the President’s Office for Regional Administration and Local Government) Financial Reports database [29], DHS 2010 [10], and facility data provided by MoHSW [36] for the following interlinked indicators based on four WHO health system building blocks (health financing, workforce, commodities, and facilities): (i) per capita recurrent expenditure [29]; (ii) skilled health workforce density per 10,000 population [26, 34]; (iii) availability of tracer drugs at health facilities [37]; and (iv) health facilities per 10,000 population [26, 36]. Applying methodology developed by Countdown [46], data for each health systems indicator were categorised according to proportional achievement of a benchmark, as follows: (i) green: ≥ 75 %; (ii) yellow: 50– <75 %; (iii) orange: 25– <50 %; (iv) red: <25 % (Fig. 2). International benchmarks were used for categorising health workforce and health facilities data [47, 48] (Fig. 2). No international benchmarks exist for per capita recurrent expenditure or commodities availability. Thus, we allocated four groups representing the diversity in funding levels and categorised as green the subnational regions with the highest expenditure levels. We used ≥75 % as a benchmark for available tracer drugs. Benchmarks categorising health systems data to construct implementation readiness scores for regions in Tanzania The aim of the interviews was to explore the budget and decision-making process to understand resource flows and identify potential bottlenecks at different levels of the health financing system, and across different types of health expenditure. Twenty-two purposively sampled semi-structured interviews were undertaken with stakeholders from MoHSW, PMO-RALG, representatives of development partners and regional and council health management teams. A semi-structured interview guide was developed during a pre-fieldwork site visit. Interviews were conducted in one region and two districts between April and July 2012, in English, face-to-face and each lasted approximately one hour. All the interviews were conducted and analysed by one author (MMA). Where respondents agreed, interviews were recorded and transcribed; otherwise, notes were taken during interviews and immediately typed up. Data were analysed using thematic analysis [49], involving several stages: data familiarisation, code generation, search and review themes and defining themes.