Globally, acute respiratory infections (ARIs) mortality [1], with the annual untreated cases of childhood ARI estimated to be over 50 million [2]. Furthermore, it is estimated that 80 % of all ARI-related deaths among children under 5 years of age occur in developing countries, making it a leading cause of infant mortality in these countries [3]. Symptoms of ARIs include respiratory rate greater than or equal to 70 breaths per minute, severe chest wall retractions, cough and inability to feed and drink [4].
Children are particularly vulnerable to ARIs, because of their inability to adequately protect themselves from the associated environmental risks, their relatively immature immune systems and physical development [4]. Children under five years are not usually able to seek health care themselves; they must rely on adults for this. This responsibility usually falls to the mother or other female caregivers in the family [5]. With differing sociocultural roles in the home, males are often regarded as breadwinners whereas females are seen as homemakers, which includes taking care of the children [5].
Securing child health depends on appropriate health care-seeking. Insight from existing studies indicates that such behaviour is associated with a variety of factors including maternal education, maternal age, household size, maternal ethnicity, and household socio-economic status [6,7,8,9]. Distance from health care facilities is known to play a key role. For instance, a study in rural Tanzania found that mothers who lived one or more kilometers from a health center were less likely to access health care [8]. Mothers were also more likely to seek health care for children aged 24 months or younger [8, 10]. Some studies also suggest that mothers seldom seek health care for their child in the early stages of a child’s illness [6, 10]. This practice would have negative implications for child health. Some studies indicate that over 30 % of child deaths can be attributed to late care-seeking [11]. Similar findings have been reported in Ghana, where ARIs accounted for 20 % of all annual deaths among children under five years [12]. Recent studies have also shown an association between socio-demographic and cultural factors such as distance from facilities, income, ethnicity and household size and access to health insurance to be associated with maternal health care seeking in Ghana [13, 14]. Another study also showed that child mortality and health-seeking behavior were a function of social factors such as maternal education, place of residence, and family income [15].
The healthcare system in Ghana can be considered to be pluralistic in that it is characterized by two parallel systems: the Traditional and the Orthodox (or medical) systems [16,17,18,19,20]. The traditional system is the oldest and most widely used care system due to its accessibility, acceptability, affordability, and availability [16, 17]. It includes the use of indigenous knowledge, spiritual therapies, herbal medicines, manual techniques, and in some cases, modern medical equipment to diagnose and treat ailments [20,21,22]. Traditional medical practice in Ghana is regulated by the Traditional Medical Practice Council which gains its mandate from the Traditional Medical Practice Act 575, 2000 [23]. Despite the efforts by the council, unlicensed practitioners and others who practice in secrecy hinder the proper regulation of traditional medical practice [24]. The use of traditional medicine is more widespread in rural rather than urban areas, partially due to the skewed availability of modern health care facilities [25, 26] but mainly due to cultural norms, the desire to be part of the healing process, perceived displeasure with the medicalization of western medicine, and perceptions on the quality of care [27].
Orthodox medicine, by contrast is characterized by the use of scientific methods and principles to arrive at a diagnoses and treatment of diseases [18, 21, 22]. In Ghana, adherents to orthodox medicine, seek care in health care facilities that are either private or publicly owned. These include hospitals, clinics, polyclinics, pharmacies, health centers, and Community-Based Health Planning and Services (CHPS) compounds. The inequitable spatial distribution of personnel and resources for orthodox medicine and corresponding access to these facilities has been reported [20]. A study in 2011 suggested that, overall, whereas the ratio of traditional medical practitioners to human population stood at 1:200, orthodox doctor–population ratio stood at 1: 20,000 [28].
Studies on maternal health care seeking for childhood ARI, especially in the Ghanaian context, are very limited. Existing studies have mainly focused on the determinants of ARI among children, with limited or no focus on care-seeking behavior among the mothers [29,30,31]. The findings of these studies show that maternal and household socioeconomic factors are significantly associated with the ARI symptoms among children under age five. There is also a limited knowledge on the temporal trend in care-seeking for ARI, given that policies such as the Child Health Policy 2007–2015, Community-based Health Planning and Services (CHPS) policy, and Free Maternal Health Policy, have been introduced in the healthcare sector in the last three decades [30,31,32,33]. To the best of our knowledge, studies on maternal health care seeking for childhood ARI have not considered a recent nationally representative survey of children in Ghana. We, therefore, contribute to this growing body of literature by reporting on the determinants of care-seeking for childhood ARI using data from five successive National Demographic and Health Surveys conducted in Ghana.
Knowledge of the determinants would be useful for planning interventions that could help improve health care-seeking and ultimately secure child health. This is most important, given that ARIs are one of the leading causes of morbidity and mortality among children under five years in Ghana and internationally [1, 3, 12]. Additionally, such information would assist in making policy decisions on the attainment of Sustainable Development Goal 3; Good health and well-being. The objectives of the study are two-fold. First, the study aims to assess the association between socioeconomic factors and the type of treatment sought for ARI. Secondly, the study explores the specific effect of place of residence on seeking medical care for childhood ARI symptoms using a Bayesian hierarchical spatial logistic regression.
Data and methods
The study used data from the following Ghana Demographic and Health Survey (GDHS) over the years: 1993, 1998, 2003, 2008, and 2014 [34,35,36,37,38]. The GDHS is a nationally representative survey of women aged 15 to 49 years and men aged 15 – 59 years. Its main objective is to capture information on fertility, maternal and child health as well as family planning and attitudes towards HIV/AIDS and other sexually transmitted infections [32]. Respondents for the surveys were selected through a two-stage sampling process. The design used 20 sampling strata from stratification of each of the ten administrative regions into urban and rural areas. The first stage of the sampling process involved selecting census enumeration areas (EAs) in each stratum. The probability of selection of each EA was proportional to the size of the EA – that is, the number of residential households in the EA. In the second stage, households were randomly selected within each EA, and all women (aged 15-49 years) who were members of the household or who had spent the night before the survey in the house were interviewed. The birth history and health information of children born to eligible women in the last five years before the survey were collected as part of the data. This information was kept in the child recode dataset – the data used for this study. Detailed descriptions of the GDHS surveys and the sampling methods are available in the final reports for the surveys [32,33,34,35,36]. The sample for this study was children under age five who were ill with acute respiratory infections (ARI). The samples of children under age five with ARI in the successive GDHS used in this study were for each survey 2 204 (GDHS 1993), 3 298 (GDHS 1998), 3 844 (GDHS 2003), 2 992 (GDHS 2008), and 5 884 (GDHS 2014).
Measures
The outcome variable for this study was the type of treatment sought for children with ARI symptoms. In the GDHS, ARI among children was identified by asking mothers and caregivers of children under age five whether their children had been ill with a cough accompanied by short and rapid breathing in the two weeks before the survey. The mothers and caregivers were then asked if they had sought treatment for the ARI symptoms and where the treatment had been sought. For our outcome variable, children with ARI symptoms whose mothers or caregivers did not seek treatment were categorized into a single group and coded as “0 = No treatment sought”. We operationally defined seeking medical care as a mother or caregiver seeking an expert opinion or treatment from a public or private hospital or clinic, outside the home, for a child who shows symptoms of ARI. Therefore, children whose mothers or caregivers sought treatment in public health care facilities (such as a government hospital, government health center/clinic, government health post or CHPS) or at private health care facilities (e.g. private hospital, private clinic, private doctors, mobile clinic and maternity home), but excluding a pharmacy or drug store, were classified into a single group and coded as “1 = Sought medical care”. Children whose mothers or caregivers sought other sources of treatment, such as ‘pharmacy or drug store’, ‘traditional healer’ and ‘drug peddler’, were classified and coded as “2 = Self-care or sought traditional treatment”.
The independent variables were child demographic characteristics, maternal socioeconomic status, household characteristics, and place of residence. The age and sex of the child were included as child demographic characteristics. Maternal socioeconomic and demographic characteristics were age, marital status, religion, education, employment, and health insurance coverage. Maternal education was grouped into four categories and coded as: “0 = No formal education”, “1 = Primary education”, “2 = Secondary or high school education”, and “3 = Higher education”. The responses for maternal employment status were derived from the GDHS question that asks respondents whether they had been employed during the 12 months prior to the survey. Respondents who indicated they had worked in the past year, were currently working and those with employment, but on leave were grouped into a single category and coded as “1 = Employed”; while those who indicated they had not worked in the past year were coded as “0 = Unemployed”. Another maternal characteristic captured was health insurance status – that is, whether the mother or caregiver was covered by health insurance – with a binary variable coded as “0 = Uninsured” and “1 = Insured”.
Household characteristics used as possible predictors of type of treatment sought for ARI symptoms were: household wealth index (a DHS construct based on assets ownership and housing characteristics of each household) and sex of household head. Households were classified as poorest, poorer, middle, richer, and richest under the household wealth index. Place of residence was also included as an independent variable; it was categorized and coded as “0 = urban” or “1 = rural”. Finally, a variable capturing the GDHS periods or years was also included as an independent variable, with the 1993 GDHS as the reference category.
Analysis
Univariate, bivariate, and multivariate analyses were performed concerning the objectives of this study. We first conducted a cross-tabulation analysis to examine the distribution of the study sample characteristics by type of treatment sought for children with ARI symptoms. We performed a Pearson Chi-square test of independence to identify any association between the predictor variables and the type of treatment sought for children with ARI symptoms; while the Cramer’s V test was used to determine the strength of the association. Next, we conducted a multinomial logistic regression analysis to examine the effect of maternal socioeconomic status and household characteristics on the type of treatment sought for children with ARI symptoms among children under 5 years of age. The outcome variable of this study was a nominal variable with three categories hence the choice of multinomial logistic regression modeling. To facilitate easy interpretation of the result, we report the relative risk ratio (RRR) of the multinomial logistic regression model – which is the ratio of the probability of choosing medical treatment and self-care or using a traditional healer over the probability of not seeking treatment (the baseline category). The cross-tabulation and multinomial logistic regression were conducted using STATA version 16 [37]. We controlled for the survey design effects using the ‘svyset’ command in STATA to adjust for the sampling clusters and weights. We also estimated the design effect to provide insight into the efficiency of the sample used in this study. The first design effect – Deff - is a ratio of the variance estimate from our sample and the variance estimate from a hypothetical sample of the same size drawn as simple random sampling (SRS). A Deff value greater than 1 implies the study sample is more efficient than SRS. The second design effect estimate – Deft – is the ratio of the standard errors in the study sample and in the SRS.
The final analysis was a Bayesian hierarchical spatial logistic regression to account for potential spatial dependence among the administrative regions used in the sampling process. The sampling method used in the GDHS means the data from the surveys are geographically distributed data. Due to the sampling approach, traditional regression models such as the multinomial logistic regression model used in the previous analysis do not account for potential spatial dependence. The concept of spatial dependency is based on Tobler’s First Law of Geography, which states, “everything is related to everything else, but near things are more related than distant things” [38]. This concept violates the independently distributed observations and error assumptions of regression models. To address this issue, we included a spatial random effect term in our Bayesian model to account for potential spatial dependency between the administrative regions. In the modeling, we assume a Besag-York-Mollie (BYM) specification. BYM specification proposed by accounts for both smoothened spatial structure of the data based on the concept of spatial dependency (spatial autocorrelation) and an unstructured spatial effect. The Unstructured spatial effect is based on the assumptions that the effect of the spatial units or administrative regions may be independent of neighboring units or regions – independent region-specific noise (unstructured spatial effect) [39]. The BYM modeling specification, thus addresses the issue of potential bias where there is no spatial dependency [39]. For the Bayesian hierarchical spatial logistic regression, we created a dummy variable out of the original outcome variable and coded it as “1 = sought medical treatment” and “0 = did not seek treatment or self-care or sough traditional healer”.
The Bayesian hierarchical spatial logistic regression was implemented in the open-access R software [40] using the R-INLA package [39,40,41,42,43]. We visualized the results for the region-specific odds ratio of using a medical treatment for ARI and the posterior probability that the odds ratio of using a medical treatment for ARI symptoms exceeds one (exceedance probability) in R software using the ‘tmap’ and ‘tmaptools’ packages [41].
