Socioeconomic health inequality in malaria indicators in rural western Kenya: evidence from a household malaria survey on burden and care-seeking behaviour

Malaria is one of the most important diseases in many low- and middle-income countries, primarily affecting children and pregnant women in sub-Saharan Africa. In 1999, approximately 60% of global malaria deaths were concentrated among the poorest 20% of the global population [1]. In sub-Saharan Africa, 3.1% of all disability-adjusted life-years (DALYs) were lost to malaria in 2002 [2]. Although preventable and treatable, the number of deaths due to malaria remains high. In 2015, there were an estimated 429,000 malaria deaths (range 235,000–639,000) worldwide, and most (92%) of these deaths occurred in the Africa [3]. In Kenya, despite remarkable achievements in malaria prevention and control over the last 10 years, malaria remains a leading cause of morbidity and mortality [4]. In 2015, while the prevalence of microscopically-confirmed malaria was 8% amongst children less than 15 years (13% by malaria rapid diagnostic test [RDT]) nationally, it was 27% (43% by malaria RDT) in the lake-endemic region of western Kenya [5].

The relationship between malaria disease and poverty has often been described as a vicious cycle and whether malaria infection is a consequence of or a cause for low household socioeconomic status (SES) has been debated for decades [6]. In a systematic review of nine studies to establish the relationship between malaria and poverty, two studies found a significant relationship between poverty and malaria, four studies found no significant relationship and three studies demonstrated mixed results [7]. Malaria also imposes substantial costs to individuals, households and governments. Globally in 2015, total funding for malaria control and elimination efforts was estimated at US$2.9 billion; governments in malaria-endemic countries provided 32% of the total funding, of which 65% or US$612 million was expenditure by national malaria control programmes for programme implementation and 35% or US$332 million was expenditure on health service delivery [3]. In high malaria-transmission settings in Kenya where the average annual household expenditure was less than US$800 [8], households spent an average of US$10 (range US$9–12) monthly and approximately US$120 (range US$108–144) annually on malaria treatment in 2010 [9].

The microeconomic relationship between malaria burden and composite wealth indices is also mixed and contradictory [7]. A study among Tanzanian children, using principal component analysis (PCA) to rank households, established that malaria was associated with household SES when SES was the dependent variable, but an individual’s SES was not associated with malaria risk when malaria was the dependent variable [6]. Another study in Tanzania, which investigated causality between malaria and SES, established that the higher the household wealth quintile the lower the prevalence of malaria in individuals in the household [10]. The lack of consistent findings may partly reflect the inherent difficulty in measuring SES and differences in the populations studied, methods used to measure malaria infection and malaria intensity in the study areas [7, 11‐13].

Various methodologies for assessing SES employing broad quantitative and qualitative aspects of poverty have been recommended including PCA, polychoric PCA and multiple component analysis (MCA) [14‐19]. A study from rural western Kenya compared the three methods of ranking households into SES quintiles and established that although the methods produced similar results, MCA gave the highest percentage of the total variation for the household asset variables and thus the largest weights for the variables. The study concluded that MCA was a better model for generating asset weights than PCA or polychoric PCA [20]. The study further conducted comparison between ordinary PCA and MCA and established that 93% of households were placed in the same quintile by both methods, 87% of the households by ordinary PCA and Polychoric, and 91% by Polychoric and MCA and that ordinary PCA asset index was statistically significantly correlated with the index based on MCA (r = 0.997, p < 0.01) [20]. The MCA model also allows both quantitative and qualitative variables, which is not possible with traditional PCA methods [20‐22]. However, studies using MCA methods to investigate socioeconomic inequalities related to malaria indicators are lacking.

Health inequality data, including malaria parasitaemia prevalence, use of malaria prevention and treatment interventions and expenditures on malaria medication, are often collected but not analysed from an economic or equity perspective. Yet, such data and analysis are important for monitoring health inequalities and the impact of malaria control interventions at the microeconomic level. The aim of this study was to establish the relationship between household SES and inequalities in malaria-related health indicators including morbidity, use of insecticide-treated nets (ITNs), care seeking, and expenditure on malaria medications in a malaria-endemic area of rural western Kenya.

The study evaluated the relationship between burden of malaria infection and household SES within an area of rural western Kenya. This is the first published paper to assess the relationship between malaria indicators and SES using the MCA model to generate household wealth quintiles based on continuous and categorical variables. The findings show that individuals in the poorest households had a higher burden of malaria infection compared to those from less-poor households. Persons from the poorest households also spent significantly more money to purchase medications that are not recommended for malaria treatment, which are likely to have less clinical efficacy and lead to unnecessary risk of adverse effects and complications of taking inappropriate medications. No significant associations between care seeking and SES or medication use and SES were observed, and the study found high access to and use of ITNs, irrespective of household SES. Prevalence of malaria infection was significantly higher in Gem sub-county compared to Rarieda sub-county. This could be due to high vegetation coverage and the presence of River Yala which cuts across the sub-county. These findings contribute to the scarce published literature on malaria and socioeconomic inequalities. Although there is extensive literature on health inequalities and health outcomes more generally, no previous study has evaluated the relationships between malaria indicators and SES using MCA to analyse microeconomic data.

The study results are similar to findings by Somi and colleagues who reported a large variation in parasitaemia rates between socioeconomic groups, where individuals with the lowest SES were significantly more likely to have malaria parasites than less-poor individuals [8]. Findings from this analysis, however, contrast with those of de Castro and Fisher who found that SES had no association with malaria infection [6]. The de Castro study, however, was limited to children aged 6–59 months whereas a study by Somi et al. in which the analysis was not restricted to a specific age group [5, 8]. Both cross-sectional studies used household assets and proxies to measure SES using the PCA model [6, 10, 19]. This study addressed some of the limitations of these previous studies by using malaria confirmed by microscopy, as the main outcome of interest controlling for age group, and using generalized linear models instead of traditional concentration indices and Lorenz curves to estimate the risk of health indicators as a measure of inequity as recommended by World Bank and World Health Organization [13, 17, 19].

Previous studies on health outcomes, including malaria, and SES have traditionally used the PCA model to generate a household SES index. The PCA model relies heavily on dichotomous socioeconomic variables to achieve a composite household SES index [14, 15, 19]. Benefits of using the MCA model are inclusion of both continuous and categorical variables and larger weights for assets, which increases statistical power [21, 22]. Using PCA models to generate household SES indices was anticipated to facilitate a more robust evidence base for assessing the associations between health outcomes and poverty, especially at the household and community levels [14, 15, 19, 21]. However, recent literature has demonstrated the weaknesses of PCA models including the inability to accommodate continuous variables such as number of assets owned and generation of low asset weights, which makes it difficult to determine clear wealth quintile cut-offs particularly in settings where most households have the same assets and therefore the same or very similar SES scores [21, 22]. Based on evidence already published [20] that MCA is a better model than PCA, this study has applied MCA for assessment of socioeconomic status to assess malaria related health inequalities.

The study determined that nearly two-thirds of persons had fever in the past 2 weeks, and the majority (81%) of them took medication with an equal proportion of individuals among the poorest and less-poor households. Nearly half of those who sought care went to health facilities, but the other half sought care from pharmacies and informal drug shops. Research from Kenya shows that people who seek care from health facilities are more likely to get tested for malaria and receive the first-line recommended medications for treatment compared to those who go to pharmacies and informal drug shops [29, 31‐33]. There were no differences observed in the expenditure on all malaria medications per person between SES groups, and less than 5% of persons purchased non-recommended malaria medicines overall, which is a positive finding. However, the poorest households spent more to purchase potentially ineffective medicines compared to less-poor households; ineffective treatments potentially prolong parasitaemia or fail to clear parasitaemia, which can lead to recrudescence or severe malaria and increased expenditures on additional treatments or hospitalizations. The findings of this study suggests a need to encourage healthcare seeking in the formal health sector, especially among the poorest households.

No significant differences in ITN ownership or usage between the poorest and less-poor households was observed in this rural western Kenya community in 2012, which was less than a year after the first universal coverage ITN distribution in Siaya County. Ownership of at least one ITN per household overall reached near full coverage (94%) and was well above the national target of 80% [34]. Subsequent national household surveys have consistently demonstrated significant differences in ITN ownership, access and use between the lowest and highest wealth quintiles [4, 26]. Because this cross-sectional study was conducted in a relatively small geographic area (i.e., three sub-counties) of rural Siaya County, there is probably much less socioeconomic variation compared to the national population and more uniformity in programmatic distribution within a single county.

A key principle of the Kenya Health Policy 2014–2030 is to achieve equity in the distribution of health services and interventions by 2030 [35]. Findings from this study illustrate existing socioeconomic inequalities in the burden of malaria infection and expenditures on non-recommended malaria medication in this rural western Kenya setting. However, the lack of differences between SES groups in care seeking, overall medication use and expenditures, and ITN ownership and use demonstrate the progress toward achieving equitable access to health services and distribution of free malaria commodities, including first-line medicines for treatment and ITNs, in western Kenya. Analysis of malaria indicators in relation to household SES using MCA methodology can be used to monitor progress towards achieving health equity goals in line with the Kenya Health Policy 2014–2030 and global sustainable development goals [36].

The study has a number of limitations. Findings were based on one cross-sectional survey preventing any evaluation of cause-and-effect of SES on malaria indicators over time. A longitudinal or trend analysis of repeated surveys would have provided an opportunity to study changes in SES and monitor the gap in malaria indicators between the poorest and less-poor households over time as malaria control interventions, including free first-line malaria treatment at health facilities and ITNs, were implemented. The other limitation was inclusion of only households with children < 5 years of age based on protocol-specific objectives. While this study advances the knowledge related to the association between malaria, control interventions and microeconomics, under these limitations, it reduces generalizability. Additionally, expenditures were calculated per person rather than per household because not all persons in the household were interviewed or tested for malaria. Although all children < 5 years of age were surveyed, only a small proportion of persons ≥ 5 years of age were included in the survey sample. Finally, the use of assets as proxies for SES also has limitations including, most importantly, that the monetary value of assets was not collected, and hence the net worth of the household might be over- or under-estimated. Asset-based proxies, however, have been shown as a reasonable way to measure wealth status in the absence of household income or expenditure data, which is not commonly available in informal economies [30]. This study did not compare the current results with any other results which could have been analysed using other methods besides MCA because there is already evidence that MCA is better model than PCA. Such comparison would not be statistically different in assigning households into the quintiles [20].

In rural western Kenya, individuals in the poorest households had a higher burden of malaria prevalence compared to those from less-poor households. However, no significant differences were observed in care seeking, overall medication use and expenditure, or ITN ownership and use between households based on SES. This study demonstrates that the MCA model can be a useful tool for assessing malaria-related health inequalities at the microeconomic level and to monitor progress towards achieving equitable access to health services and distribution of malaria interventions in line with national and global health and development goals.