Study setting
Nigeria, a federal republic in West Africa, is administratively divided into 36 states and a federal capital territory (FCT). These states are further divided into six geopolitical zones: North-West, North-Central, North-East, South-East, South-West, and South-South. In each of the geopolitical zones, the people are homogeneous and unique in their ways of life.
Data source
This study is a retrospective cross analysis of nationally representative data from the Nigeria Demographic and Health Surveys (NDHS) for 2003, 2008 and 2013. In the NDHS, women of reproductive age (15–49 years) are interviewed about their reproductive health, children health and other related topics. Selection of survey respondents involved a stratified two-stage cluster design in which all the census enumeration areas (EAs) in the country are stratified into rural and urban areas. A specified number of EAs was then selected from each stratum. Households in selected EAs were listed and 45 were systematically selected in each one [18]. Data were collected from eligible women at selected households with the aid of a pretested questionnaire by trained interviewers. Data collected from women about their under-five children were specially processed to constitute the kids recode file which was utilised for analysis in this study. The kids recode file contain household, maternal and child-related variables. Detailed description of field procedures are well stated in the full report of the surveys [18,19,20].
Derivation of variables
Three sets of variables were analysed: household environment, maternal and children characteristics. The variables were created from the household and children recode file respectively for each of the three surveys. Variables for the household environment include the type of cooking fuel used in the household; smoking by adults in the household; source of drinking water; type of toilet, quality of housing material and over-crowding in the sleeping room. Cooking fuel was categorised as clean (electricity, liquefied petroleum gas, natural gas or biogas) and unclean (coal, ignite, charcoal, wood, kerosene, animal dung, straw, shrubs, and grass). Source of drinking water was classified as improved (piped into dwelling/yard/plot, public tap/standpipe, tube-well or borehole, protected well and spring, rain water, and bottle water) or unimproved (unprotected well and spring, tanker truck/cart with drum, surface water, sachet water, and other sources). Similarly, toilet type was either improved (flush/pour flush to piped sewer system, septic tank or pit latrine, ventilated improved pit latrine, pit latrine with slab, composting toilet) or unimproved (flush/pour flush not to sewer/septic tank/pit latrine, pit latrine without slab/open pit, bucket, hanging toilet/latrine, no facility/bush/field) [18, 21, 22].
The quality of housing material was derived from a composite index based on the type of floor, wall and roof materials. Floor material was coded 1 if made of cement, carpet/rug, ceramic tiles, vinyl asphalt strips, parquet or polished wood, and 0 otherwise. Also, wall material was coded 1 if made of cement, stone with lime/cement, cement blocks or bricks and 0 if otherwise. For roof material, the following were coded as improved (1): cement, roofing shingles, calamine/cement fibre, ceramic tiles, metal/zinc. To derive the final categories for quality of housing material, the scores for the floor, wall, and roof were aggregated and re-coded as follows: good (3), average (2) and poor (0/1). Sleeping room was deemed over-crowded if the average number of person per sleeping room in the household was greater than or equal to three. The groupings of environmental factors were adapted from the year 2013 Nigeria National Demographic Health Survey (NDHS) and the 2010 WHO and UNICEF document on progress on sanitation and drinking water [18, 21]. For the 2003 survey, there was no data collected on the type of roofing material. As a result of which the variable for housing quality was not derived.
The maternal characteristics which were extracted from the children recode file were: maternal age, education, occupation, household wealth quintile, type of residence, and geopolitical region. Children characteristics included current age in months, sex, birth order and stunting which was an indicator of chronic malnutrition. The outcome variable was ARI symptom which was derived from the response to two questions on whether under-five had a cough in the last 2 weeks and if a cough was accompanied by short rapid breaths.
Analysis
Data for each survey (2003, 2008 and 2013) were analysed separately. The analysis involved the use of frequencies and percentages to describe all the study variables. Next, univariate models were fitted to describe the relationship between the outcome and explanatory variables. At the third step in the analysis, two models were fitted. In the first model, variables for the household environment were included in the model (Model I). Secondly, other maternal and children characteristics were added to the first model. As such the second which is a full model was used to identify the independent factors associated with childhood ARI symptoms. The complementary log regression model with robust standard errors was employed because of the outcome variables. This model is suitable for situations in which the dependent variable is dichotomous and its proportion in the analytical sample is below 10% [23]. Model coefficients were exponentiated to derive odds ratio with their 95% confidence interval. Stata SE Version 12.0 was used for all analyses. Sample weights were computed and applied to account for the complex sample design of the surveys.
The model is of the form:
$$ \Pr \left({y}_j=1|{x}_j\right)=1-\exp \left[-\exp \left({x}_j\beta \right)\right] $$
Where yj = presence of ARI symptom in child j.
xj = vector of covariates (explanatory variables).
β = coefficients for the covariates (xj).