Constructing a malaria-related health service readiness index and assessing its association with malaria mortality in under 5-year-old children in Burkina Faso

Background : The Service Availability and Readiness Assessment (SARA) surveys generate data on the readiness of health facility services. We constructed a readiness index related to malaria services and determined the association between health facility malaria readiness and malaria mortality in children under the age of 5 years in Burkina Faso. Methods : Data on inpatients visits and malaria-related deaths in under 5-year-old children were extracted from the national Health Management Information System (HMIS) in Burkina Faso. Bayesian geostatistical models with variable selection were fitted to malaria mortality data. The most important facility readiness indicators related to general and malaria-specific services were determined. Multiple correspondence analysis (MCA) was used to construct a composite facility readiness score based on multiple factorial axes. The analysis was carried out separately for 112 medical centres and 546 peripheral health centres. Results : Malaria mortality rate in medical centres was 4.8 times higher than that of peripheral health centres (3.46% vs. 0.72%, p<0.0001). Essential medicines was the domain with the lowest readiness (only 0.1% of medical centres and 0% of peripheral health centres had the whole set of tracer items of essential medicines). Basic equipment readiness was the highest. The composite readiness score explained 30% and 53% of the original set of items for medical centres and peripheral health centres, respectively. Mortality rate ratio (MRR) was by 59% (MRR = 0.41, 95% Bayesian credible interval (BCI): 0.19-0.91) lower in the high readiness group of peripheral health centres, compared to the low readiness group. Medical centres readiness was not related to malaria mortality. The geographical distribution of malaria mortality rate indicate that regions with health facilities with high readiness show lower mortality rates. Conclusion : health are associated with lower malaria mortality rates. Health system readiness strengthened regions Sahel, and du Mouhoun. Emphasis should be given to improving the management of essential medicines and to reducing delays of emergency transportation between the different levels of the health system.


6
Thus far, Burkina Faso has conducted two SARA surveys; in 2012 and in 2014. The data have been used to assess readiness of surgical [20], obstetric [13] and family planning services [14].
However, no studies have been carried out to date to investigate the relationship between health service readiness and health outcomes in Burkina Faso. Hence, to fill this gap, we focused our research on malaria-related services and determined the extend at which malaria services readiness is effective and able to prevent malaria deaths in children under the age of 5 years. Our findings will help to optimize resources allocation and improve SARA survey analyses for Burkina Faso and other LMICs.

Study area and national health system
Burkina Faso is a malaria endemic country. Indeed, malaria is the leading cause of consultation, hospitalisation and mortality in under 5-year-old children [21]. The health system of Burkina Faso is pyramidal and consists of three levels [22]. The peripheral level is formed by the health district and includes the "Centre de Santé et de Promotion Sociale (CSPS)", medical centres, isolate dispensaries, delivery centres and district hospitals. The latter serve as referral centres of the former health facilities. The second level is made of the regional hospitals, which are the reference structures for the district hospitals. The third level comprises the national and teaching hospitals and is the highest level of referral care providing specialized services. In 2016, there were approximately 1,760 CSPS, 47 district hospitals, eight regional hospitals and five national and teaching hospitals.

Data sources
The 2014 SARA survey We analysed health facility data from the Burkina Faso SARA survey carried out in 2014 that included 786 health facilities grouped in three strata: (i) 19 teaching hospitals, private polyclinics and regional hospitals (stratum 1); (ii) 90 district hospitals and medical centres (stratum 2); and (iii) 671 CSPS, isolate dispensaries and delivery centres (stratum 3). Strata 1 and 2 correspond to a rather homogeneous group as they are staffed with physicians (in most cases), and hence, we combined them to increase the sample size and created two hierarchical levels of health facilities: medical centres (highest level) consisting of strata 1 and 2 and peripheral health centres (lowest level), including those of stratum 3. Medical centres are mostly staffed by physicians, while peripheral health centres are managed by nurses.
The items in the SARA questionnaire are specific to the services provided by the health facilities and remain the same across health facility levels for a specific service. As facility levels differ in terms of the services and health programmes they offer, the items have different importance or weights depending on the facility level. For example, electric power source is mostly found in medical centres as they are situated mainly in urban areas, while solar power is the main source of energy in rural areas. Medicines for chronic diseases or surgery, anesthesia and X-ray equipment are mainly part of the medical centres rather than peripheral health centres.
We defined as tracer items readiness indicator (i) for the general services and (ii) for the malaria-specific services, the proportion of health facilities having the tracer item available. The Mortality data were extracted from the Health Management Information System (HMIS) for a full year (January-December 2014). Malaria mortality in children below the age of 5 years was defined as the number of malaria-related deaths among all in-patient visits to a health facility of that age group. The mortality outcome was linked to the SARA database according to the health facility.

Statistical analysis
Bayesian negative binomial models were fitted on the number of malaria-related deaths at the health facility. We assumed that the number of malaria-related deaths at the health facility follows a negative binomial count distribution, and hence, Bayesian negative binomial models were fitted on the malaria deaths data. The total number of children below the age of 5 years visiting the facility (i.e. the denominator of the mortality rate outcome) was considered as an offset term in the model, that is the logarithmic transformation of it was introduced as a covariate with fixed regression coefficient equal to 1. The tracer items were included as covariates in the model. Bayesian variable selection was applied to determine the most important tracers associated with the malaria mortality rate. A separate analysis was carried out for each facility level, i.e. medical centres and peripheral health centres.
MCA was applied to the most important tracers, adhering to an approach put forth by Ssempiira et al. (2018) [19]. In short, let be the set of selected tracers, , = 1, … . and 0, and 1, be two binary indicators corresponding to the presence and absence of the from the facility , respectively, that is, 0, takes value 1 when the tracer is absent ( = 0) and 0 otherwise.
Likewise, 1, takes value 1 when the tracer is present in health facility (i.e. = 1) and 0 otherwise.
The readiness score for health facility , based on the ℎ factorial axis is defined by = 9 Asselin (2009), we define a composite readiness score as = 1 ∑ ∑ ∑ ( −   =1  1  ∈{0,1}  =1   ) , , , where is the number of factorial axes used in the composite score and ( − ) is the Dirac delta function, which takes the value 1 when the weights related to , are selected from the factorial axis and 0 otherwise, that is, Furthermore, we assessed the association between malaria mortality rate and the readiness index described above, using a geostatistical Bayesian negative binomial model. Locational random effects were included in the model to take into account spatial correlation. We assumed a Gaussian process with an exponential correlation function of the distance between health facilities. The analysis was adjusted for the type of health facility location (urban/rural) and of administrative status (public/private). Further details of the statistical methods are provided in Additional file 1. Redlands, CA, USA).

Health facility characteristics and malaria mortality
The SARA survey carried out in Burkina Faso in 2014 included 766 health facilities. Among these health facilities, 658 (85.9%) reported complete malaria mortality data, and hence, they were used for subsequent analyses. Seventeen percent of the facilities (n=112) belonged to medical centres.
Around 80% of medical centres are located in urban areas, while in peripheral health centres, more than 80% of the facilities are in rural zones ( Table 1). Most of the facilities are managed by the government (77% of medical centres and 93% of peripheral health centres). The malaria mortality rate in medical centres is 4.8 times higher than that of peripheral health centres (3.46% vs 0.72%, p<0.0001).  items across all domains of the general service offered by medical centres (Table 2) For peripheral health centres, 29% (10/34) tracers were selected in the general service. These are similar to those in medical centres with the exception of the essential medicines, as most of them were not available in peripheral health centres. Regarding malaria-specific services offered by peripheral health centres, readiness to the first line of antimalarial drugs (96.3%) and to malaria diagnostics (85.5%) was similar as observed in medical centres. Posterior inclusion probability: gives the probability of the tracer to be included in the final model and it is calculated by the proportion of all possible models in the variable selection procedure that include the specific tracer. For example, the posterior inclusion probability of 21.4 estimated for the power tracer indicates that this tracer was included in 21.4% of all possible models generated from all general services-related tracers. 3 Item not included in the variable selection procedure due to low relative frequency i.e. <5%

Health facility readiness index
MCA was applied on the tracers items selected from the variable selection procedure to obtain a readiness score. Fourteen and six factorial axes were sufficient to build the composite indices for medical centres and peripheral health centres, respectively. Standard coordinates of the selected tracers are provided in Table 3 (medical centres) and Table 4 (peripheral health centres).  peripheral health centres 53% vs. 18%). Furthermore, the composite score based on the subset of tracers explained more variation than the composite score based on the whole set (medical centres: 30% vs. 26%; peripheral health centres: 53% vs. 30%).
15      The geographical distribution of malaria mortality rate showed a similar pattern with that Regarding malaria-specific services, the average availability of "staff and guidelines" and 117 the "medicine and commodity" domains is higher for health facilities in peripheral health 118 centres than medical centres. More than 80% of them have their staff trained and know the 119 guidelines for malaria management. In addition, more than 95% in these facilities also possess 120 first-line treatment for malaria. Malaria is the most important cause of morbidity and mortality in under 5-year-old children, which explains that substantial efforts are being made to train 122 peripheral health facility workers, render medicines and other medical supplies available for 123 malaria case management at all levels of the health system. In recent years, there has been a 124 shift from first-line medicines to ACTs, introduction of RDTs, and ITN campaigns [35,36].