Assessing delivery practices of mothers over time and over space in Uganda, 2003–2012

In 2010, an estimated 287,000 women died as a result of pregnancy or delivery-related complications [1]. Over 99 % of these maternal deaths occurred in developing countries [2]. Ranking 161 of the 187 UN member nations in the 2013 Human Development Index, Uganda is one of the least developed countries in the world [3, 4]. At 360 maternal deaths per 100,000 live births in 2013, the maternal mortality ratio (MMR) is one of the highest in the world, more than 22 times higher than more developed regions where the MMR is estimated at 16 deaths per 100,000 live births [5].

Between 1995 and 2009, 2.7 million third-trimester stillbirths occurred yearly [1]. The majority of these deaths took place in developing countries and most are preventable with appropriate antenatal, delivery and postnatal care [6]. In Uganda, the neonatal mortality rate was 26 per 1000 live births in 2010, while the estimated stillbirth rate was 25 per 1000 live births. About half of stillborns are classified as intrapartum deaths–deaths which occur during labour or delivery. A woman in a low-income country in sub-Saharan Africa is 24 times more likely to suffer an intrapartum stillbirth than a woman from a high income country [1].

Skilled birth attendance (SBA) is associated with reduced maternal and neonatal mortality and risk of stillbirth [1, 7‐9]. In low-income countries, newborns delivered by a skilled birth attendant in a health facility stand a greater chance of survival than newborns delivered elsewhere [7, 10]. Effective SBA requires an environment enabling skilled attendants to perform to the best of their abilities and gives them access to essential medications and equipment and timely intervention or referral options in the event of complications [11].

Promoting skilled birth attendance in health facilities is a global priority, especially to achieve the targets set for Sustainable Development Goal (SDG) 3— Ensure healthy lives and promote well-being for all at all ages—and to meet targets of other global initiatives, such as the Every Woman, Every Child initiative [12, 13]. The Ugandan government’s priority strategies to combat maternal and child mortality include improving access to emergency obstetrical care and developing an enabling environment for SBA [14, 15]. Facility-based delivery (FBD) is a proxy for lower-risk delivery as delivery with SBA can be difficult to quantify when patients assume that all clinicians are skilled.

Studies exploring factors associated with using health facilities for delivery have included: maternal characteristics, index pregnancy characteristics, access, socio-cultural beliefs and past experiences of the mother. Maternal characteristics such as young maternal age, high levels of education and increased autonomy are positively associated with FBD [16‐20]. A cross-sectional study in the district of Busia identified that parity less than four [AOR 2.9 (1.6–5.6)] and autonomy in deciding to attend ANC [AOR 1.9 (1.1–3.4)] are positively associated with FBD [18]. A large proportion of women less than 20 years old (65.8 %), with secondary education or above (81.4 %), in the highest wealth quintile (87.7 %), and mothers of first-order births (73.1 %) report delivering in an institutional setting [20].

Several traits relating directly to the index pregnancy also affect safe delivery practices. The timing of the onset of labour and duration of labour do impact delivery location. Labour onset late at night or short duration of labour can inhibit a mother from accessing a health facility for delivery services [21]. A cross-sectional study conducted in Bugesera district in Rwanda reports that attending more ANC visits (OR 1.567 [1.163–2.112]) is positively associated with FBD [22], while a study in southern Tanzania highlights that having been advised on FBD during ANC [AOR 1.82 (1.25–2.63)] is positively associated with skilled attendance at delivery [23].

Ease of physical and financial access to health facilities is positively associated with FBD [18, 19, 22, 24, 25]. Increased distance to facilities or facilities located in difficult terrain, as well as high costs, both formal and informal, mitigate against FBD [16, 17, 23, 26‐30]. A study conducted in four districts of Laos in 2009 reveals that eliminating user fees associated with delivery at the point of services increases by 9.8 percentage point (p < 0.1) the coverage of skilled birth attendance [26]. A cross-sectional study from the 2011 Nepalese DHS showed that increase distance to facilities or facilities located in difficult terrain explained up to 1.7 %-point of the 17.7 %-point regional gap in FBD between terai/hill and mountainous regions [28]. The 2008–2009 Indian district level household and facility survey revealed that the mean out-of-pocket expenditure on delivery care increased by 7 % for every 10 % increase in state domestic product per capita [29]. High levels of absenteeism and lack of supervision of health care workers are known barriers to using health facilities [31].

A woman’s knowledge of pregnancy and delivery-related risks also support using a health facility [1, 9, 11, 18, 20, 32]. A mother’s previous birth experiences, location of the penultimate birth, her beliefs regarding health providers’ skills and her perceptions of the quality of health facilities are additional factors influencing her decision to use FBD [16‐19, 22, 24, 27, 31, 33]. A woman’s supposition of how she will be received by health care workers and if her wishes will be respected by them also influence uptake of FBD [18, 20, 32, 34, 35]. Traditional beliefs, including the fatalistic perception that maternal death is a normal risk and to be expected, lead to underutilisation of health facilities [1, 15, 34, 35].

Uganda, a Millennium Countdown Country, is one of 75 countries where >95 % of maternal and child deaths between 1990 and 2010 took place; these mortality rates are high nationwide [1]. The World Health Organisation and UNICEF recommend that countries analyse data at both national and subnational levels to identify gaps and inequities in health services. Both UN agencies also urge improvements in surveillance and survey data to decrease missing data [1].

It is also important to uncover factors influencing FBD. The studies mentioned above generally include at most one geographic covariate, and most do not consider the spatial distribution of FBD. The aim of our study is to identify the simultaneous correlates of FBD, in order to provide a framework for prioritizing districts for support. Our study assesses variations over time and space in FBD and fits a statistical model to identify factors associated with FBD. We apply this model to 2003–2011 data to identify areas expected to have low indicator coverage in 2012 and validate this prediction with the 2012 data. This approach can therefore inform policy-makers and program managers on the status of FBD and trends and variations occurring over time and can identify locations needing further investigation.

The study was conducted by the USAID STAR E-LQAS project, which is implemented  Management Sciences for Health with Liverpool School of Public Health as a technical partner for LQAS. Trained district health managers collected data from individuals with household surveys conducted in 19–64 districts of Uganda at seven points in time during 2003–2012, using the Lot Quality Assurance Sampling (LQAS) methodology [35]. The surveys were financed by the World Bank and USAID [36] with questions adapted from accepted sources such as the Uganda Demographic Surveys. The District Health Management Team divided each district into 4–6 administrative subdistrict strata called supervision areas (SA) and selected 19 mothers of children 0–11 months (or 24 if 4 SAs) randomly from each SA. The SA sample size was selected so that when subdistrict data (the SA) are aggregated, the resulting district-level coverage proportion estimates for key indicators are calculated with a 95 % confidence interval not exceeding ±10 %. Villages were selected using probability proportional to size (PPS) sampling, wherein a comprehensive village population list supplied by each district was the sampling frame used to select villages from which the individual samples are taken. There was on average 88 villages in the sampling frame of each SA. PPS sampling ensures that sample villages are selected based upon their proportional representation of the entire population. Usually a sample of 19 villages was identified, sometimes less if some villages had a large population size relative to others in the same SA. Individual respondents were then randomly selected from the PPS-selected villages using a randomizing technique [35]. The main approached used was segmentation sampling. Segmentation was recommended as it was found to be a more rigorous second-stage sampling technique [37] and is now advocated in several survey guidelines [38‐40]. District Health Officers also requested a second approach be offered, namely, simple random sampling from an updated village listing of households. The latter was recommended only in cases in which a recently updated list existed and could be verified. With either approach once a reference house was selected the next closes house was selected for interview. This addition reduced the chance of a house having a zero probability of selection. The former approach was recommended in the trainings and used most frequently. Table 1 shows the number of districts in each Ugandan region that were surveyed in each year and the number of mothers interviewed in those regions. A total of 18,471 randomly selected mothers of children aged 0–11 months were interviewed, the inclusion criterion being that mothers had have been present in the village at least 3-months prior to the survey. Each maternal questionnaire included demographic characteristics and various health-related behaviours. Respondents with missing or erroneous responses were removed, leaving a total of 18,098 (98 %) records with complete information. These data were integrated into a superset, and in this study we analysed mothers’ responses to the question “Where did you give birth?”, their age at the time of the survey (in years) and their education level (none, primary, secondary, post-secondary). Uganda LQAS data reliability studies are available for review [41, 42].

We obtained district-level data from a variety of sources, including geospatial road and population data from 2009 [34] and 2010 Geographical Information System (GIS) locations of health centres. We calculated the number of health facilities per capita (per 100,000 inhabitants) based on the number of health facilities with in-patient beds (level III and above), since mothers are referred to these higher-level facilities for FBD. Household assets data from DHS 2011 [20] were used to stratify responses by economic quintiles. Altitude data was obtained from the US Geological Survey [43].

Our analysis consists of 3 phases: FBD mapping, model construction, and prediction of priority districts and population strata in them. Phases 1–2 used the 2003–2011 data, while phase 3 also included the 2012 data. All analysis was done using the statistical software R version 2.15 [44]; we used the R-package ‘maptools’ [45] to construct the maps.

We plotted spatially the percentage of mothers of children aged 0–11 months with FBD for different points in time (Fig. 1) and report the confidence intervals in Additional file 1: Table S1 (Appendix 1). During the time period over which the data were collected, some districts subdivided; to aid comparisons across time we plotted the indicator on the 2011 district map. The mean of the indicator across all districts in each year was 49 % in 2003–2004, 57 % in 2006, 58 % in 2009–2010, 65 % in 2011, and 66 % in 2012. For early survey years (2003–2006) most districts had <60 % of mothers of infants with FBD (77 and 58 % of the districts surveyed in 2003–2004 and 2006 respectively), except around the capital, Kampala (95 % in 2003–04 and 89 % in 2006). For later years (2009–2011), >60 % of mothers had FBD in most surveyed districts (61 and 68 % of the districts surveyed in 2009–2010 and 2011 respectively). Although much of northern Uganda was not sampled in the more recent time periods, the one district sampled in both 2003–2004 and 2011, Arua, increased from 26.3 to 73.4 % (Additional file 1: Table S1). The progress has not been replicated as dramatically everywhere, with half of the eastern Ugandan districts surveyed in 2011, for example, still reporting <50 % FBD.

All district-level covariates except for wealth index and mean altitude showed significant nonlinearity, and so were log-base-2 transformed before being incorporated in the model. Marital status was removed from the model since it did not have a statistically significant effect. Age was included as a continuous variable because the AIC value for this model was lower than models with 2, 3, and 5 age categories and the same as the model with 4 age categories.

Table 4 shows the logistic regression results. The second column reports the estimated effect of each covariate or interaction of covariates on FBD on the logistic scale. The third column reports the odd ratio of FBD related to each covariate. Each odds ratio was calculated between two ‘average’ mothers: one who has a secondary education, is 25-years old, was surveyed in 2007, and lives in a district where all district-level covariates take their average values (as given in Table 3); and the other who is identical except for a unit increase in the covariate in question. The results show that the odds of FBD were significantly lower for each one-year increase in maternal age [OR 0.98, 95 %CI (0.97, 0.99)], whereas they were significantly increased for each additional level of maternal education [primary: OR 1.59 (1.42, 1.78), secondary: OR 3.37 (2.88, 3.94), post-secondary: OR 10.4 (6.28, 18.1)]. The odds of FBD were significantly greater in districts with double the health facilities per capita [OR 1.12 (1.02, 1.23)], or road density [OR 1.13 (1, 1.26)], or in districts with a unit increase in the wealth index [OR 1.38 (1.24, 1.53)]. Living in the capital, Kampala, was strongly associated with FBD [OR 8.38 (2.24, 23)]. Districts with a double unit increase in the standard deviation of altitude (a proxy for the roughness and difficulty of the terrain) were strongly associated with a decrease in the odds of FBD [OR 0.89 (0.84, 0.94)]. Finally, there was a significant time trend: mothers surveyed in later years were more likely to have FBDs [OR 1.08 (1.04, 1.13)].

Our model had significantly lower AIC (AIC = 13,383) than the null spatial model (AIC = 13,690), and hence was a better model of the observed spatial variation. The area under the ROC curve reported a 71 % probability that our model ranked a true positive data point higher than a true negative one, which is significantly better than the 50 % probability predicted by a random model. Using cross-validation, the indicator predicted for each district based on unseen data was within 20 percentage points of the observed indicator for 95 % of the districts, and 77 % of the districts were within 15 percentage points of the observed indicator.

Using the 2003–2011 fitted logistic regression model, we predicted for all Ugandan districts the reasonable lower and upper limits for FBD during 2012. We used 18 (50) years old as most strongly (negatively) associated age value, and post-secondary (none) as most strongly (negatively) associated education value. We categorised districts into different priority levels using these predicted ranges. We validated these predicted priority groups by comparing them with indicator values calculated from a subsequent 2012 LQAS survey covering 61 districts (Fig. 2). Additional file 1: Table S2 in Appendix 2 gives the details of the predicted range and the observed indicator in 2012. The priority map identifies many north-eastern districts as being mild or high priority—classifications that agree with the low values for these districts seen in the observed data. In addition, all but four (6.6 %) of the observed indicator values fall within the predicted range.

The factors assessed in this study are not necessarily an exhaustive list of factors impacting uptake of FBD. We had limited data regarding household and personal wealth, for example, and despite free healthcare in Uganda, FBD has costs which reduce access to care [25, 62]. Additional data on maternal wealth could increase our understanding of the relationship between cost and use of FBD. District-level data was limited to one time point for each covariate, so the model relies on the assumption that the district-level covariates did not change significantly over time. This assumption is likely to be more reasonable for some covariates than others: mean and standard deviation of altitude will not have changed over the course of these surveys, but population density is more likely to have changed. Without supervision area locations it was not possible to calculate each supervision area’s distance to the nearest health facility, which is likely to be an important factor in FBD. We used health facilities per capita as the best proxy available for this variable.

We did not have data for parity in the assessment of maternal age and its impact on FBD. Further research should take into consideration their relationship and extricate their individual effects.

A hierarchical model may provide an alternative approach to estimate district-level information, as it could uncover a hidden ‘ease of access’ variable influencing FBD. Other studies attempting this approach [63] have tended to use data with greater spatial resolution than was available in this study. The model presented in this paper was designed to be fitted and interpreted by the survey collection teams as a regular part of monitoring and evaluation; hierarchical models would have introduced additional complexity to both fitting and interpretation. While these types of models may be investigated in the future, we showed by cross-validation that the current model provides a good fit without using hierarchical modelling.

The use of the model to predict the indicator in unsurveyed districts before an LQAS survey has been performed is of course limited by the lack of individual-level covariates in those districts. Since both age and education are strong predictors of FBD, it was not possible to give a point estimate for the indicator. However this method can be used to suggest which districts it would be most efficient to include in the next round of LQAS surveys.