Study Design
We conducted a secondary data analysis of the 2016 Uganda demographic health survey (UDHS) data set.
Data Collection
This data were collected from 20th June 2016 to 16th December 2016 [11]. It was a nationally representative survey carried out by the Uganda Bureau of Statistics as part of the international MEASURE Demographic Health Surveys (DHS) with the support of ICF International and United States Agency for International Development (USAID). UDHS is a periodical survey conducted every five years. UDHS 2016 had four different questionnaires. Household questionnaire collected data on household environment, assets and basic demographic information of household members while women’s questionnaire collected data about women’s background characteristics, reproductive health, domestic violence and nutrition. The men’s questionnaire collected men’s health indicators’ data while the biomarker questionnaire collected data on anthropometry and blood tests [11]. Regarding anthropometry, weight was recorded in kilograms to the nearest one decimal point and was measured using an electronic scale ( SECA 878) [11]. Height was recorded in centimeters to one decimal point.
Study Setting
As of July 2018, Uganda had a population of 40,853,749 million people with 23.8% of the population residing in urban areas [22] and the country has a total area of 241,551 square kilometers [23]. Uganda’s health system has six levels ranging from the highest level of national referral hospitals to the lowest level at the community level [24]. Agriculture contributes about 24 percent of gross domestic product (GDP), providing half of export earnings and is the main source of income for the 84 percent of Ugandans living in rural areas [25].
Study Sampling and Participants
Samples were collected using stratified two stage cluster sampling design with census enumeration areas as the primary sampling units [11]. The first stage of sampling involved selecting 697 enumeration areas (EAs) including 162 urban and 535 rural enumeration areas selected from the list of the 2014 population and housing census sample frame [11]. One enumeration area in Acholi region was excluded due to land disputes hence ending up with 696 EAs. Enumeration areas with over 300 households were segmented and only one segment selected with probability proportional to the segment size as this helped minimize the burden of household listing [11]. The enumeration areas that were involved in the survey were chosen independently from each stratum with probability proportional to size. The second stage of sampling involved selection of households through equal probability systematic sampling. A list containing all households and maps in the selected enumeration area were made available and households that were in institutional living arrangements were excluded [11]. A power allocation with a small adjustment was done in the allocation of sample enumeration areas to ensure that the minimum number of clusters per survey domain are achieved. Finally, a representative sample of 20,880 households (30 per EA or EA segment) was randomly selected.
Women aged 15 to 49 years who were either the permanent residents or slept in the selected household the night before were eligible for inclusion in the Uganda’s demographic health survey 2016 [11]. Anthropometric measurements were done by trained technicians for about a third of the UDHS 2016 survey sampled women. The women sampled were neither pregnant nor had a birth two months before the survey [26]. Our secondary analysis only considered women aged 20 to 49 years and excluded women aged 15 to 19 years (adolescents) because the recommended anthropometric indicators (BMI and Height) for assessing undernutrition for those above 20 are different from those of adolescents (BMI for age and Height for age) and cannot be directly compared [27]. According to World Health Organization guidelines, measurement of adolescent nutritional status should account for age, as it is not until 19 years old that these curves approach convergence with adult models [27, 28]. According to the UDHS report, 18,506 women consented and filled in the questionnaires, of which 14,242 were aged 20 to 49 years. Of these, 4731 were sampled for anthropometry and 4640 had their anthropometry done while 91 were not present or refused anthropometry to be done [11]. To maintain the representativeness of the sample and possible differences in response rates across regions, sampling weights were used.
Outcome Variables
In this study, underweight and stunting were the outcome variables. Underweight was defined as body mass index (BMI) less than (<) 18.5 kg/m2 while stunting was defined as height less than (<) 145cm [3, 11, 20]. Underweight is further classified as severe (BMI less than 16.00 kg/m2), moderate (BMI 16.0016.99 kg/m2), and mild (BMI 17.00-18.49 kg/m2 ).
Explanatory Variables
This study included determinants of undernutrition basing on evidence from available literature and data [15, 20, 26, 29, 30]. These factors were divided into individual level (age, marital status, working status and education level), household level (wealth index, household size and sex of household head) and community level (region and residence) characteristics. Wealth index is a measure of relative household economic status and was calculated by DHS from information on household asset ownership using Principal Component Analysis [11, 31]. Different household assets were used to calculate separate wealth indices for rural and urban areas, combined into a national wealth index and these quintiles are; the poorest, the poorer, the middle, the richer and the richest quintiles [11, 31]. Place of Residence was aggregated as urban and rural. Region was categorized into four; Northern (Teso, Karamoja, Lango, Acholi, West Nile), Central (Kampala, Central 1 and Central 2), Eastern (Busoga, Bugishu and Bukedi) and Western (Tooro, Ankole, Bunyoro and Kigezi) [32]. Level of Education was categorized into: no education, primary education, secondary and higher education. Age was categorized into 20 -29, 30- 39 and 40-49. Household Size was categorized as less than six members and six and above members. Sex of Household Head was categorized as male or female. Working status was categorized as: not working and working. Marital Status was categorized into married and this included those in formal and informal unions and not married.
Statistical Analysis
The SPSS analytic software version 24.0 Complex Samples package was used for this analysis. Use of the complex samples package accounted for the complex survey sampling while use of sample weighted data accounted for the unequal probability sampling in different strata. Analyses were done by descriptive statistics and logistic regressions. Frequency tables and proportions/percentages were used to describe categorical variables while means and standard deviations for continuous variables. Initially, each exposure was assessed separately for its association with the outcome variables (stunting and underweight) using bivariate logistic regression and we present crude odds ratio (COR), 95% confidence interval (CI) and p-values. Independent variables found significant at p-value less than 0.2 [33-35] were included in the multivariable model. The non-significant variables whose associations with undernutrition had been shown in previous studies were also included in the multivariable logistic regression models.
In the multivariable analysis, we constructed two models based on the categorization of the independent variables into women individual and household and community level factors. We first performed a logistic regression model which included only individual characteristics (age, education level, working status and marital status). Then after, we constructed a final model which included individual characteristics adjusted for household and community characteristics (wealth index, residence, region, household size and sex of household head). Adjusted odds ratios (AOR), 95% Confidence Intervals (CI) and p-values were calculated with statistical significance level set at p-value < 0.05. Sensitivity analysis was done with only women who were underweight and had normal BMI after excluding those with BMI above 25.