Spatial Distribution and Analysis of Risk Factors Associated with Anemia Among Women of Reproductive Age: Case of 2014 Rwanda Demographic and Health Survey Data

1. INTRODUCTION

Anemia is an important public health problem affecting all age groups of population. Globally, it affects 1.62 billion people and is more prevalent among pre-school age children and women [1, 2]. There are various causes of anemia among adults as well as children. Anemia among adults may also be the result of also some specific nutrients deficiencies, namely iron and vitamin B-12, infectious diseases such as HIV, tuberculosis, malaria, helminthic infection, schistosomiasis, and hookworm [3-6] among others but the contributions of these potential causes to the overall anemia burden are not well known and vary by geographical locations, dietary practices, genetic, and season. In the study carried out by Fischer et al. [7] among adult Mexican women, it was found that household food insecurity was associated with anemia. Maternal anemia is also associated with a risk for low birth weight and inadequate iron store for the newborn [8, 9]. In addition, anemia in pregnant women increases the risks of maternal and child mortality; it can also increase the risk of having preterm delivery; and it has negative consequences on the cognitive and physical development of children and work productivity [8, 9, 10-12]. Iron deficiency with or without anemia is prevalent among women of childbearing age because of menstrual losses or higher physiologic needs for iron. Women are generally known to be more vulnerable to iron deficiency causing anemia than men. According to the World Health Organization, iron deficiency anemia is considered as a public health problem when its prevalence exceeds 5.0% of the population [13]. In Rwanda, in general, the trend was decreasing as there was a tremendous decrease from 2005 to 2010 (26% -17% respectively). But from 2010 to 2015, the prevalence of anemia slightly increased from 17% to 19% [14].

The previous studies conducted in Rwanda were restricted to geographical locations such as province or districts of the country [15, 16]. In addition, none of them accounted for the possible non-linear effect of continuous covariates or mapped structure spatial effects. Therefore, the main objective of the current study is to identify the risk factors associated with anemia among women of reproductive age and to map their spatial effects. To the best of our knowledge, there is no study in the literature using structured logistic regression to identify the risk factors and mapping their spatial variation in Rwanda.

2. METHODS

2.4. Statistical Method

2.4.1. Model Formulation

Suppose y_ij is the anemia status of woman i from district j, with i = 1, 2, ..., 6056 and j = 1, 2,..., 30; where y_ij = 1 if woman i is anemic and y_ij = 0 otherwise. Let X_ij = (x_ij1, x_ij2,..., x_ijm) be a vector of all m continuous independent random variables and W_ij = (w_ij1, w_ij2,..., w_ijk) a vector of k categorical independent random variables. The anemia status of the women of reproductive age is a binary outcome in this study and therefore follows a Bernoulli distribution;

(1)

where P_ij is the probability that a woman from district j is anemic and 1 - p_ijis the probability that a woman i from district j is not anemic. The risk of anemia can be associated with the explanatory variables using appropriate link function in Generalized Linear Models (GLM) approach as follows;

(2)

Model (2) is known as binary logistic regression, where W_ij^' is the vector of explanatory variables and β is the vector of coefficient parameters. But classical GLM has a rigid assumption that all observations are independent; however, this assumption is sometimes not satisfied as some observations may have spatial dependence. Therefore, there is a need to include the spatial variability in the model (2) and it is given by;

(3)

where β_i is the vector of fixed effect corresponding to categorical variables, f_k is the appropriate smooth function of continuous variables such as mother’s age at the birth, current age of the mother and the size of the family and f_spat (S_j) are the parameter of random effects which captures unobserved spatial heterogeneity at district S_j.

2.4.2. Parameter Estimation

The estimation of the parameters is done using a full Bayesian analysis approach, where the appropriate prior distributions have to be assigned to all unknown parameters. In the current study, diffuse priors P (β) ∝ const are assigned to all fixed regression parameters. The current study estimated non-parametrically continuous covariates by assigning to them second–order Gaussian random walk priors [20] and given as with Gaussian error , the variance controls for the smoothness of f. For the prior distribution of structured spatial effects f_i, the current study used the nearest neighbour Gaussian Markov random field model specified as an Intrinsic Conditional Autoregressive (ICAR) prior distribution [21] given as follows;

(4)

where N_S is the number of regions and denotes that region r is neighbor of region s. Two different regions or areas are neighbours if they share a common boundary; neighbouring regions are assumed to have similar patterns, otherwise these two regions are not neighbours.

2.4.3. Posterior and Inferences

Posterior distributions are the distribution of the parameters after observing the data and are obtained by updating the prior distribution with observed data. A full Bayesian inference is based on the analysis of the posterior distribution of the model parameters. In the present study, the Bayesian posterior marginal distribution was estimated based on integrated nested Laplace Approximation [22], SPSS version 24 was used for data management and data exploratory and R-INLA 3.3.1 was used for inferential analysis.

2.5. Model Selection

Model goodness of fit was assessed using Deviance Information Criteria (DIC). The smaller the value of DIC the better fit is the model [23, 24]. DIC is a generalization of Akaike Information Criterion [25] and the DIC value is given by;

(5)

where is the posterior mean of the deviance and pD is the number of effective parameters in the model that penalizes the complexity of the model. DIC takes both the model fit (summarized by ) and model complexity (captured by pD) into consideration when comparing models. When comparing two competing models, the size of the difference in DIC values of these two models in order to declare that one model is better than other is not clear cut. But some studies have stated that a difference in DIC of 3 between two models cannot be distinguished whereas a difference between 3 and 7 can be weakly differentiated [23, 26].

3. RESULTS AND INTERPRETATIONS

In order to make sure that the estimates obtained from the present study were represented at the national level, the survey weights provided in Rwanda Demographic and Health Survey data set were used in the analysis. The initial descriptive statistics were done using cross-tabulation analysis and it was assessed based on Chi-square test to investigate the relationship between anemia status among women of reproductive age and categorical variables at 5% level of significance; we considered a median and interquartile range for continuous variables and the results are summarized in Table 1. A total of 6056 non-pregnant women of childbearing age was considered in this study, among them, 1205 were from urban and 4850 from rural area; the majority of participants were literate (72.8%) and more than half were married or living with partner (52%). The median age of the respondent was 28 and the interquartile range was 16. The median number of members per household was 5 whilst the interquartile range was 3; the median age of the respondent at first birth was 21 and the interquartile range was 5. The overall prevalence of anemia among non-pregnant women of reproductive age in Rwanda was 18.9%.

Tobacco use was significantly associated with anemia status of women of childbearing age (p-value=0.003). The prevalence of anemia was 19.0% among women who do not use tobacco and 28.9% among women who use tobacco. Women’s marital status is also a significant risk factor associated with anemia status of women of childbearing age (p-value < 0.0001). The prevalence of anemia was 19.0%, 18.2%, 27.6% and 23.4% among single, married, widow and divorced or separated women, respectively. The respondent’s body mass index was also significantly associated with anemia status (p-value < 0.0001). The prevalence of anemia was 18.8% among normal or obese women and 26.3% among underweight women. The wealth quintile of the household is significantly associated with anemia status among women in Rwanda (p-value <0.0001). The prevalence of anemia was 22.7%, 18.8% and 16.2% among women from poor, middle and rich households, respectively. The bed net use was also significantly associated with anemia status of the women (p-value=0.001). The prevalence of anemia was 18.1% among women who used a bed net before the survey and 21.7% among women who did not use a bed net before the survey. The prevalence of anemia was 22.5%, 19.1% and 18.0%, respectively among women with no level of education, primary education and women with secondary or tertiary education level (p-value=0.028). The prevalence of anemia was higher among women from households where they used water from unprotected well 30.7%, protected well 30.0% and less among women whose household used water piped into dwelling or yard 17.7% (p-value=0.001). The prevalence of anemia was 16.9%, 18.8%, 23.0% and 31.8% among women whose household used charcoal, wood, straw or grass or shrub and other sources of cooking fuel other than charcoal, wood and straw or grass or shrub respectively (p-value <0.0001). The prevalence of anemia was higher among women from rural compared to women from an urban area (20.0% and 16.5%, respectively).

The prevalence of anemia was higher in Southern province 22.9% and low in Kigali city 14.6% (p-value=0.000). The prevalence was 15.0% among women who did use contraceptive methods and 22.4% among women who did not use any type of contraceptive method (p-value <0.0001).

In the final analysis, only significant variables from cross-tabulation analysis were included in multivariable spatial modelling.

It is seen from Table 2 that model A4 has the smaller DIC compared to DICs of model A1, model A2 and model A3. Therefore, based on the principles of DIC, model 4 was found to be the better model fit and thus it is used in the final analysis.

Table 2.

Statistics	Model A1	Model A2	Model A3	Model A4
DIC	5789.44	5811.30	5745.63	5741.18
	5737.71	5750.55	5667.55	5645.65
pD	25.865	30.38	39.04	47.77

3.3. Spatial Effects

The present study found positive and negative structured spatial effects on anemia status among women of reproductive age. Fig. (4) presents structured spatial effects on anemia among women of childbearing age; where yellow shows significant negative effects on anemia, green shows significant positive effects and yellow-green non-significant effect.

4. DISCUSSIONS

Anemia among women of reproductive age is a common problem encountered mainly among pregnant women mostly in developing countries and is the main cause of morbidity and mortality. The overall prevalence of anemia in the current study was 18.9% and this was considerably lower compared to what was reported from DR Congo 38% [27], Tanzania 45% [28], Burundi 39% [29], Uganda 32% [30], and Ethiopia 24% [31] to name few.

The findings of this study showed that the risk of anemia is higher among women from poor households. This finding is similar to that of NISR [14] and Kamruzzaman et al. [17]. This may be explained by the fact that a woman from a poor household may not easily find enough and healthy food, appropriate and timely health service, and better maternal health care. But the study did not find a statistical significant difference between a woman from middle-income households and a woman from rich households.

The findings from this study also showed that the risk of anemia was higher among underweight woman. This is consistent with findings from previous studies [17, 19, 32-34]. Although higher body mass index may not always indicate appropriate micronutrient intake, an underweight (BMI<18.5g/m²) person is more likely to have other associated co-morbidity illness and hence be deficient in some necessary micronutrient that may be associated with anemia [33].

The results from this study revealed a significant inverse relationship between literacy and anemia status among women of reproductive age. A literate woman may access information related to maternal health more easily, and have a job or business to generate an income for the family which in turn contributes to the well-being of her and her family. The education level is known to have an important effect on the socio-economic and demographic context of anemia prevalence, mostly in developing countries, but the current study did not find a statistically significant association between respondent’s education levels and anemia status.

Another finding from this study was that the risk of anemia among women lessened with the use of mosquito bed nets. The use of bed nets in this present study was used as a proxy for malaria infection. In a study by Wirth et al. [35] done in Sierra Leone, they found that malaria and inflammation were each associated with anemia. The risk of anemia was observed to decrease among women who were using a contraception method at the period of survey. This finding is similar to the findings of Kamruzzaman et al. [17]; Massawe et al. [36]; Heck et al. [37] and Asres et al. [38]. The results from this study revealed that a hygienic toilet can serve as a protective factor against anemia among women. A similar finding was obtained by Kamruzzaman et al. [17]. The type of toilet facility is among the most basic level of sanitation. The women from households without toilet facilities may be susceptible to infection by hookworms and parasites [18, 39, 40] and this may support our findings that lack of toilet facilities in household or community increases the risk of anemia. The association of unhygienic toilet with anemia can also be explained in general by poor health or chronic blood loss through gastrointestinal parasite infection [17].

The present study did not find any statistically significant association between lactating, pregnant women, amenorrhea, woman’s education level, number of children ever born, place of residence and tobacco use and anemia among women of reproductive age in Rwanda in multivariate structured logistic regression.

The study also considered non-linear effects of continuous covariates on anemia status among women. It was observed from Fig. (2) that the risks of anemia increase with increasing the size of the family. Further, it was observed from the same figure that between five and ten members, the credible intervals were a bit narrowed and this means that there were many observations in this interval and few observations towards the end which made credible intervals to widen (few households have 22 members).

It was also observed from Fig. (3) that giving birth at an earlier age increases the risk of anemia, especially between 13 and 23 years old. It was found that the risk of having anemia increases with increasing age. In the study by Kamruzzaman et al. [17] among Bangladesh women, they found that older women (30-49 years old) were more likely to be anemic compared to young women (15-29 years old).

The finding of this study showed a positive spatial pattern of anemia distribution in districts Nyanza, Huye and Gisagara (southern province), and Kirehe and Ngoma in Eastern province. This finding is similar to that of NISR et al. [14]; in their report from RDHS (2014/2015); they found a higher prevalence of malaria in Southern and Eastern provinces and prevalence of anemia was higher in Southern province followed by Eastern province. In a recent study by Habyarimana et al. [41] on risk factors of childhood anemia, found that the prevalence of anemia was higher in Southern and Eastern Provinces. But malaria may not be the only cause; therefore, a thorough epidemiological study is needed to find out the cause of this higher prevalence of anemia in the above-named districts.

CONCLUSION

This study applies a Bayesian semi-parametric modeling methodology. This model is different from classical modeling approach concepts in its flexibility that can reveal the important features of continuous variables. The objective of the current study was to identify the risk factors associated with anemia and to map their possible spatial effects on anemia among women of childbearing age. The findings of this study showed that body mass index, contraceptive use, bed nets use, drug intake for malaria in pregnancy, wealth quintile of the household, type of cooking fuel, type of toilet facilities and type of source of drinking water, age of the woman, age at the first birth and the size of the family were the risk factors associated with anemia. The findings from this study also highlighted five districts from southern and eastern provinces with the highest number of anemic women. The negative spatial effects were found mostly among districts from Northern Province and Kigali city. The findings from this study may serve to help public health planners, policymakers and other related institutions in making effective decisions for fighting anemia efficiently. They may also serve as useful empirical grounding to the growing literature on geographical and sociological dimensions of health and disease [42, 43].

REFERENCES