The Effect of Socioeconomic Drivers on Under-five Mortality Rates: A Survey Bayesian Meta-analysis Study

2.1. Study Area and Data Sources

The data used in this study are based on four recent secondary survey data on four countries located in the east sub-Saharan African region for 20 years from the Demographic health survey program on https://dhs program.com/ Namely Rwanda, Tanzania, Kenya and Uganda. These surveys are nationally representative of information on childhood mortality obtained in the birth files of women included in the survey. Such surveys are designed to update or estimate health and demographics. Fig. (1) represents the study areas of interest: Tanzania, Rwanda, Uganda, and Kenya. The data were obtained from a demographic health survey program.

2.2. Study Variables

The outcome variables considered in this study are the survival times of childhood mortality measured in months from birth until an event occurs or censored of a child below 60 months or five years old and considered based on a reference period. Children are considered based on their lives starting within the reference period selected. Children who died within the reference period are taken as uncensored cases; otherwise, they are censored. Also, six explanatory variables were considered in this study and may be classified as socioeconomic variables. These variables are the sex of the child, breastfeeding, education level, household income, family size, and type of place of residence.

2.3. Methods

The Cox proportional hazard survey regression model was adopted at the first of the analysis to obtain effective size, namely, hazard ratios with corresponding standard errors to be used at the second level of the analysis. The Survey Cox regression model addresses the problem of censoring while accounting for survey design [9-13]. Censoring is important to consider in survival analysis because failing to account for it properly can lead to biased or inaccurate estimates of survival probabilities and other related measures [9, 14, 15]. The presence of censoring introduces uncertainty into the analysis, as the exact time of the event is unknown for censored individuals [14-16]. The hazard ratio is the exponentiated linear part of the model, which describes the effect of one or more explanatory variables. The model addresses the censoring problem and the complex nature of the survey design. The ordinary Cox regression model is given by

(1)

here, λ₀(t) is the baseline hazard function at time t, x_i is a vector of model explanatory of individual i^th at time t, and β is the vector of unknown regression parameters, and they are assumed to be the same for all observations in the study. If we divide equation 1 by the baseline hazard function, we obtain.

(2)

The study adopts meta-analysis on a set of two groups, survival studies that may have censored data issues. Meta-analysis involves systematically collecting, evaluating, and synthesizing the results of individual studies to obtain an overall estimate of the effect size or association, increasing the statistical power and generalizability of the findings [17, 19], [20, 21] The log hazard ratios and the corresponding standard errors summarise each of the studies [19, 20]. We want to increase the statistical power of the overall results by integrating results across studies and having an overall interpretation of all studies. The fixed effect is considered and assumes homogeneity of study results. The Bayesian approach is based on several philosophies involving the test and intervals and is important for updating evidence [22-24]. The Bayesian analysis consists of three main parts: Prior, likelihood, and posterior distribution. A prior distribution represents the initial beliefs or knowledge about uncertain variables before observing any data [21, 22]. It is an essential component of Bayesian analysis and can be updated to a posterior distribution using Bayesian inference. Initially, uncertainty under Bayesian is expressed by a prior distribution of quantities of the study interest. Then, data and assumptions on their generations are summarised in the likelihood, combining prior distribution and the likelihood yield posterior distribution for the quantities of interest in a study [25-27]. The posterior distribution is often given as a point estimate and credible interval, and Bayesian is often carried out using WinBUGS, INLA, STAN, and rjags package in R.

(3)

where θ is the parameter of interest, y is the observed data, P(θ) denotes prior distribution, and P(y) denotes the likelihood. In the context of this study, the prior distribution represents the uncertainty of the effect being measured, such as risk ratios, odds ratios, and hazard ratios. They are expressions of subjective beliefs concerning the size of the effect and the size of the prior distribution shows the degree of uncertainty about the quantity of interest [26, 28, 29]. If there is less or non-information, a non-informative prior may be used and in a meta-analysis, the use of a non-informative prior often shows a position of prior ignorance [25].

2.4. Ethical Considerations

Addressing the impact of socioeconomic factors on under-five mortality requires careful consideration of ethical principles. The data used in this study was obtained from the Demographic Health Survey Progress website on request. Data should be anonymized and stored securely to prevent unauthorized access or misuse. The study does not identify the household location in each country.

3.1. Descriptive Analysis

The total number of children (1511550) included 12278 (8.1%) who experienced the event before their fifth birthday in Kenya. Of 120371 children included in the study from Rwanda, 17614(14.6%) died before their fifth birthday. A total of 109455 children were included in the study, with 12749 (11.7%) dying before their fifth birthday in Tanzania.

3.2. Bayesian Meta-analysis for Survey Data

Table 1 displays the Effect size, the mean, the median, the standard error, and the credible interval of the four countries of interest. The hazard ratio for education level (secondary or higher) means 0.602 with credible interval (CI) [0.583 – 0.6222]. The credible interval does not include one. This means that education level is significantly associated with child deaths; having secondary or higher education reduces child death. Lower levels of maternal education are associated with suboptimal child health practices, such as inadequate healthcare utilization, lower awareness of preventive measures, and poorer management of childhood illnesses, ultimately increasing the risk of child mortality. The effect size of a child’s place of residence (rural) is 1.138 with CI [1.108 – 1.168]. This does not include one; thus, the place of residence is significantly associated with child death. A child from a rural area has an increased risk of dying compared to a child from an urban area. Rural areas tend to have higher levels of poverty and lower socioeconomic development compared to urban areas. Poverty is associated with limited access to nutritious food, clean water, sanitation facilities, and healthcare services, all of which are crucial for child health and survival. Rural areas often face challenges in terms of healthcare infrastructure, availability of skilled healthcare professionals, and access to essential health services. Limited access to healthcare can lead to delayed or inadequate treatment of childhood illnesses and exacerbate the risk of child mortality (Table 2).

The effect size of the child sex is 1.140 with CI [1.121 – 1.157], which means the sex of the child (female) has an increased risk of dying before the fifth birthday. The effect size of breastfeeding status (yes) is 0.817 with CI [0.801 – 0.833], which does not include one; thus, economic well-being is associated with child mortality. Studies conducted by [29] and [30], suggest that being part of a larger family is associated with an increased risk of child mortality, even after controlling for other factors such as socioeconomic status and maternal education. Such studies highlight the importance of family planning and reproductive health interventions in reducing child mortality. The effect size of economic well-being (rich) is 0.833 with CI [0.728 – 0.637], which means that the average risk of a child dying from a rich household is lower than a child from a poor household. A family size greater than seven reduces child mortality. Poorer households often face barriers in accessing quality healthcare, including limited financial resources, lack of health insurance, and distance from healthcare facilities [31]. As a result, children from poor households are more likely to experience delays in receiving necessary healthcare, leading to increased child mortality rates. Poorer households often lack access to safe water sources and adequate sanitation facilities. This increases the risk of waterborne diseases and infections among children, leading to higher child mortality rates.