This research uses a mediation approach, disaggregated data on health aid and data from nationally representative household and health facility surveys and links malaria facility information to malaria service use at the child level. While existing literature has examined the effect of facility readiness to provide contraceptive services as well as facility readiness to provide care during delivery, these studies do not consider the role of health aid.12 13 Further, there is a considerable lack of knowledge on the effect of health aid on the readiness of facilities to provide malaria services and on malaria service utilisation in sub-Saharan Africa. This study sought to inform aid allocation strategies designed to strengthen malaria service delivery in Malawi using a theoretically driven health services approach, but found limited evidence that health aid relates to malaria health service utilisation through its influence on increasing malaria facility readiness in a defined service area. Despite this finding, results did reveal that health aid increased diagnostic capacity within health facilities. The finding that health aid does contribute to the availability of diagnostic tools, but not an overall ability to implement these tools indicates a clear gap for funders to consider when developing future allocation policies. This section will first discuss the results by each of the four research questions, then discuss study limitations and strengths and conclude with the contributions of this research.
Discussion of results
According to the Baron and Kenney mediation approach, Step 1 typically would necessitate a significant association between health aid and malaria service utilisation to justify continuing the mediation analysis. In recent years, the literature has supported dropping the Step 1 requirement that Pathway C or the ‘total effect’ is significant to assess mediation.14 Recent studies have demonstrated that mediation analyses provide explanatory value and should be pursued even when the total effect is not statistically significant.14–16 In addition, if the ‘total effect’ of health aid on utilisation of malaria services is not significant, there may be practical or theoretical reasons for estimating mediating effects.17 In the case of this study, identifying the association between health aid and malaria service readiness still has scientific merit in order to understand which components of the malaria healthcare system need to be strengthened or require improved measurement.17 Referring to the Baron and Kenney model, results from Step 2 indicate a significant relationship between aid and diagnostic service readiness, but not overall malaria readiness or training readiness. Further, these results appear to present conflicting evidence of the association between health aid and facility readiness such that aid appears inversely associated with facility readiness overall but positively associated with diagnostic capacity. This finding is interesting for several reasons. First, it indicates that health aid is translating to increased malaria diagnostic capacity. Early and accurate diagnosis of malaria is essential to lower mortality and stem the spread of disease.18 Diagnosis is also important, because not every fever contributes to malaria. Due to increasing concerns related to drug resistance, presumptive treatment is no longer recommended unless diagnostic tests are not accessible.18 Second, the finding that aid is translating into diagnostic capacity, but not overall facility readiness highlights a gap that could significantly impact the quality and accessibility of malaria diagnosis. Referring back to the Baron and Kenny model for steps 3 and 4, findings indicate that the coefficient for Pathway B was significant in two cases of overall malaria service readiness. This result is not surprising given the fact that 96% of health facilities in Malawi had full capacity to provide first-line treatment medicine. Individuals may have obtained effective treatment from facilities with a lower level of readiness. The results also suggest that both Pathways C and C’ are similar and not statistically different from zero, suggesting there is neither a direct nor indirect (ie, mediated via facility readiness) association between health aid and malaria service utilisation. Therefore, mediation was not formally tested using a Sobel test.19
Limitations and strengths
Several limitations should be acknowledged when interpreting these results. First, it is not possible to directly link the MIS to the facility. Therefore, KDE was used to assign a density of malaria service readiness to each child. Although this method allows for the incorporation of distance decay as well as the possibility to assign an aggregate measure of readiness to each child, it does not reflect the true travel distance to the clinic. Second, since the KDE disperses the effect of a facility across space by facility characteristics; the malaria service readiness index was derived using an equal weighting additive approach for interpretability. One could argue that the items in the index have different clinical values and thus should be weighted differently. Since a weighted additive index has not been validated within the context of KDE, there are still remaining assumptions that the dimensions do in fact carry equal weights.20 Third and related to the creation of the index, the malaria service readiness index ranked facilities as high, medium or low quality based on the distribution of the score. This was done in order to examine the differences among facilities while retaining a more intuitive understanding, Although it is a common practice in social science to dichotomise continuous variables, it may result in a loss of statistical power and result in an ability to detect significant differences.20–22 For this reason, I categorised the continuous variable into terciles. Discretisation in this way allows for the comparison of the highest and lowest groups and results in less efficiency lost.20 23 Fourth, this work is based on the assumption that the readiness of the facility to provide malaria services influences the decision to use a facility. Parents may choose to use facilities for reasons unconnected to service readiness.13 Fifth, the SPA and MIS data were collected within 1 year of each other. The malaria service utilisation data preceded the service provision data used to determine service readiness. Therefore, this research assumes that the health facility service provision capacities did not change much between the two surveys. Although this follows DHS guidelines for linking SPA and MIS data, data on malaria service availability collected during the SPA survey could have changed after the MIS survey.24 Sixth, the MIS data describe the 2-week period of prevalence of fever among children under five. While a history of fever is often strongly associated with malaria parasitaemia in malaria-endemic countries, children who had a fever may not have had malaria. Seventh, I restrict the AidData dataset to health aid projects that could feasibly have local impacts on malaria services and utilisation. However, I cannot subset the data to only include malaria-specific funding. In addition, AidData does not assign financial amounts to individual project locations. Many researchers choose to divide the total aid amount equally across all activities. In order to test the sensitivity of my results to other definitions of aid, I repeated the analysis defining health aid in terms of the dollar amount across each region. The results of this analysis were qualitatively similar to the more rigorous definition of aid based on counts of health aid projects that were used in this study. Also, this research uses date of planned completion to subset the timing of aid distribution. However, from this dataset, it is not possible to know exactly how much of this aid was actually transferred to the areas in question. A long lag between aid and the period the surveys were conducted may attenuate the association between aid and the outcomes of interest. In addition, enough time may not have elapsed for changes to occur in facility readiness that influence utilisation. In order to test the sensitivity of my results to temporality and endogeneity, I examined aid in two bins of early and late aid. Results of this analysis are located in online supplementary appendix G and indicate that it is appropriate for future research to consider binning aid into early and late periods. Eighth, this analysis assumes that health aid mainly influences the supply side of malaria care. It is possible that aid is used to create demand for malaria care and subsequently increase service use. Although, in theory improved quality of care can stimulate demand for care, in a resource constrained country like Malawi, cost associated with seeking care are formidable and unless these are addressed, improving facility readiness may not stimulate service use.25 Ninth, this analysis does not explicitly control for government and private investments in health that foreign donors do not fund, which could bias estimates by underestimating the aid amount. However, the Malawi government’s contribution to health expenditure has been consistently below the 15% recommended in the Abuja declaration, estimated at 7.2% in 2011.26 Consequently, any bias from excluding government funded interventions should be minimal. Tenth, the cross-sectional nature of both the SPA and the MIS prevent more rigorous estimates of the influence of service readiness on service utilisation. Specifically, I cannot examine whether changes in health aid in a geographic region cause changes in malaria facility readiness or utilisation.
Despite these limitations, this study has several noteworthy strengths. First, this paper linked facility census data and nationally representative household data to provide a more robust assessment of whether and how health aid, facility readiness to provide malaria services and the use of malaria services relate.13 The use of KDE to join these data provides a methodological improvement in the way the service environment was constructed and linked to household survey data. Similar to Wang, but with a different empirical approach, my analysis incorporates distance decay to measure the effect of the malaria service environment instead of only looking at the closest facility, reducing the likelihood that individuals were misclassified to a specific facility.13 Second, this is the first study to my knowledge to combine subnational aid flows with individual-level and facility-level measures of malaria service availability and use. Third, the mediation approach was innovative in its attempt to provide evidence on potential mechanisms by which health aid and malaria utilisation relate and identifies a need to strengthen or improve measurement of programme components.