Menu
Menu
Menu
Menu
Background: Substantial progress has been made in reducing childhood mortality worldwide from 1990-2015 (Millennium Development Goal, target 4). Achieving target goals on this however remains a challenge in Sub-Saharan Africa. Kenya’s infant mortality rates are higher than the global average and are more pronounced in urban areas as compared to rural areas. Only limited knowledge exists about the differences in individual level risk factors for infant death among rural, non-slum urban, and slum areas in Kenya. Therefore, this paper aims at 1) assess individual and socio-ecological risk factors for infant death in Kenya, and at 2) identify whether living in rural, non-slum urban, or slum areas moderated individual or socioecological risk factors for infant death in Kenya. Methodology: We used a cross-sectional study design based on the most recent Kenya Population and Housing Census of 2009 and extracted the records of all females who had their last child born in 12 months preceding the survey (N = 1,120,960). Multivariable regression analyses were used to identify risk factors that accounted for the risk of dying before the age of one at the individual level in Kenya. Place of residence (rural, non-slum urban, slum) was used as an interaction term to account for moderating effects in individual and socio-ecological risk factors. Results: Individual characteristics of mothers and children (older age, less previously born children that died, better education, girl infants) and household contexts (better structural quality of housing, improved water and sanitation, married household head) were associated with lower risk for infant death in Kenya. Living in non-slum urban areas was associated with significantly lower infant death as compared to living in rural or slum areas, when all predictors were held at their reference levels. Moreover, place of residence was significantly moderating individual level predictors: As compared to rural areas, living in urban areas was a protective factor for mothers who had previous born children who died, and who were better educated. However, living in urban areas also reduced the health promoting effects of better structural quality of housing (i.e. poor or good versus non-durable). Furthermore, durable housing quality in urban areas turned out to be a risk factor for infant death as compared to rural areas. Living in slum areas was also a protective factor for mothers with previous child death, however it also reduced the promoting effects of older ages in mothers. Conclusions: While urbanization and slum development continues in Kenya, public health interventions should invest in healthy environments that ideally would include improvements to access to safe water and sanitation, better structural quality of housing, and to access to education, health care, and family planning services, especially in urban slums and rural areas. In nonslum urban areas however, health education programs that target healthy diets and promote physical exercise may be an important adjunct to these structural interventions. Copyright:
We used a cross-sectional study design and based our analyses on the general population in most recent Kenya National Population and Housing Census 2009 [25]. Data was collected by the Kenya National Bureau of Statistics with reference to the night of August, 24th/25th 2009. We followed the guidelines and recommendations to assure Good Epidemiological Practice (GEP) as defined by the German Society for Epidemiology [26]. The study was therefore conducted in accordance with ethical principles and respected human dignity as well as human rights and all information was stored and used anonymously in our analysis. Our study strived to report a qualified risk-communication to the interested public. Household types covered by the census exceeded those considered adequate for this study, since some household types were only covered by a reduced census questionnaire missing important data. We therefore concentrated on housing type 1, “conventional” and excluded “refugees”, “non-conventional” (e.g. schools, hospitals), “institutions”, “travelers”, “vagrants”, and “emigrants”, resulting in a population of 37,919,647. We subsequently concentrated on usual members of the household only, arriving at a population of 35,629,354 living in 8,491,789 households. Based on 8,491,789 conventional households from the census, we calculated infant death in two steps: Based on this measure, 21,891 (2%) of the mothers’ last-born children (born between September 1th 2008 and August 24th 2009) died before August 24th 2009. It is important to note that our measure does not include mothers that died during pregnancy or delivery and it further does not reflect on those with short birth intervals, which likely underestimates infant mortality and is therefore not directly comparable to infant mortality rates. We base our study on the conceptual framework for cities and population health of Galea et al. [27] and Gruebner et al. [28] and focus preliminary on differences of living conditions, i.e. individual and socio-ecological risk factors for infant death in rural, urban and slum areas. For demographic variables, we used individual level information on mothers’ age (range 12–56 years), number of previously born children who died (range 0–14), mothers education (up to primary = 0, secondary+ = 1), and information about child’s sex (girl = 0, boy = 1) including the information whether the child was a twin or multiple (twin/multiple = 2) (Table 1). *Mean age of mothers 26.59 (range 12–56 years, standard deviation [SD]: 6.62). **Mean number of ever born children that died 0.19 (range: 0–14, SD: .62). Note that this measure excludes infant death occurring within 11 months preceding the census, i.e., the period of the outcome infant mortality. ***Mean age of household heads 37.24 (range: 15–95 years, SD: 13.09). For capturing the social environment in which a mother was living, we used household level information on household head’s sex (female = 0, male = 1), age (15–95 years) and their marital status (not married = 0, married = 1) (Table 1). For the physical environment, we constructed new variables to account for structural quality of housing, quality of water supply and mode of human waste disposal (sanitation). Quality of housing was constructed from information on material used for floor, wall and roof construction of a household. For the floor of a household, we considered wood, earth and other non-durable materials as minor quality and coded as 0. Cement and tiles were considered durable, coded as 1. Walls made of wood, corrugated iron sheets, grass/reeds, tin and other were considered non-durables and coded 0. Walls made out of stone, brick/block, mud/wood, and mud/cement were considered durable and coded as 1. Main roofing material made of asbestos sheets, grass, tin, mud/dung, and others was considered non-durable and coded as 0. Main roofing material made of concrete, tiles, “Makuti” (i.e., reed/grass type roof finish), or corrugated iron sheets was considered durable and coded as 1. The numbers for each variable were combined and summed up ranging from 0 to 3, with higher values indicating better structural quality of housing (0 = non-durable, 1 = poor, 2 = good, 3 = durable). Following national guidelines for the quality of water access [29], we considered water sources reported as ponds, dams, lakes, stream/river, unprotected spring water, unprotected well, “Jabia”, water vendor, and other as not improved, and coded 0. Protected spring, protected well, borehole, piped into dwelling, piped, and rainwater collection were considered improved water sources and coded 1. For the type of human waste disposal (sanitation), we considered uncovered latrines, bush and other as unsafe sanitation and coded as 0. Main sewer, septic tank, “Cess pool”, “VIP latrine”, and covered pit latrine were considered as safe sanitation and coded 1. Socioeconomic status (SES) can be conceptualized in various ways and the most appropriate approach to measure SES depends in part on its relevance to the subject under study [30]. In our study, we conceptualized maternal and household SES based on higher maternal education, better structural quality of housing, improved water, and sanitation considering these variables to have significant relevance to infant death. Other variables of which the majority could also be considered as indicators for SES, such as the type of material used for cooking, type of lighting fuel, a variable indicating whether a household possessed livestock, or the number of ever born children of a mother were excluded to avoid problems with collinearity [31], based on the correlation matrix using a threshold of |r| >.5 to identify high collinearity. For the place of residence, we constructed a new variable using information on urban status (rural, urban, peri-urban) and residential status (formal, slum), with three categories, i.e., 0 = rural: Households located in rural areas, 1 = non-slum urban: Non-slum urban or peri-urban areas, and 2 = slum: Slums in urban or peri-urban areas. For the ease of interpretation, we solely use the terms rural, urban, and slum in the following although urban excludes urban slums but additionally includes non-slum peri-urban areas. Likewise, the term slum includes urban and peri-urban slums. First, we fitted bivariate logistic regression models with the binary outcome infant death (1 indicating infant death) in order to identify those predictor variables that were significant at the p<0.1 level, which were used in the subsequent multivariable regression. Second, multivariable logistic regression without interaction terms was used to adjust for all variables considered significant in the first step and to identify the main effects. Third, multivariable logistic regression with interaction terms was used to investigate moderating effects between places of residence, i.e., rural, urban, and slum areas and predictor variables. We used a backward selection approach to find the most important predictors including interactions based on the lowest AIC values. Further variables were excluded based on epidemiologic reasoning and bivariate model performance. Bivariate and multivariable regression analyses were done with packages MASS [32] and the population attributable fractions were calculated in epiR [33] in the statistical programming language and environment R [34].
N/A
