Data sources
The 2016 EDHS, the fourth Demographic and Health Survey (DHS), data was collected by the Central Statistical Agency (CSA) of Ethiopia and other stakeholders; processed and organized by ICF International into different datasets. The authors accessed these public domain datasets from the MEASUREDHS website by permission. Variables related to the VCT of HIV were extracted and processed for secondary analysis. Standard protocols and three types of tools were used for collecting DHS data, namely the Household Questionnaire, the Woman’s Questionnaire, and Man’s Questionnaire. Further standardization and contextualization of the questionnaires were also done by governmental and non-governmental shareholders to maintain the validity of the tools.
Study population and sampling procedures
Ethiopia has been administratively divided into 9 regions and two towns administrates. Regions are divided into Zones, and Zones into administrative units known as Weredas. Each Wereda is further subdivided into the lowest administrative units, called Kebele. During the 2007 census, each kebele was subdivided into census enumeration areas (EAs), which were convenient for the implementation of the census and subsequent surveys (12). The 2016 EDHS followed a two-stage sampling design with stratification into urban and rural. At the first stage of the sampling, 645 EAs, 202 from urban and 443 from rural were selected based on the 2007 Ethiopian population and housing census sampling frame called EA. For Man’s questionnaire, males aged from 15-59 in the selected EAs of the selected households were eligible. The second stage of the sampling involved selection from the complete listing of households in each selected EAs by a probability proportionate to the size (PPS) of each cluster. Approximately, 28 households from each cluster (giving a total of 18,008 households), of which 17,067 households were occupied and 12,688 eligible males were identified and interviewed (12).
Measurements
The outcome variable along with all other socio-demographic variables (region, sex, age, religion, wealth status, marital status, occupational status, place of residence, sex of household head, and education level), discussion about family planning with a health worker, coverage status of health insurance, owning of a mobile telephone, use of the internet, frequency of using the internet within a month, frequency of reading newspaper or magazine in a month, frequency of listening to the radio in a month, frequency of watching television, relationship with a most recent sex partner, decisions on personal health care, decisions on large household purchases and decisions on how to spend respondent's earnings, and respondent’s involvement during check-ups for the most recent child were measured by respective direct questions asked from the respondents. The three other explanatory variables, namely risky sexual behavior, stigma status, and knowledge of HIV were indirectly measured by asking different indicator questions for each of the three variables and scoring and grouping were done to measure the generic variables.
Operational definitions
The respondents were considered to have risky sexual behavior if the response was yes to either of the following six questions (having multiple sex partner in a lifetime, having multiple sex partner in last 12 months excluding the spouse, condom using status the last time had sex with a most recent partner in last 12 months, condom using status every time had sex with a most recent partner in last 12 months, consistent condom use for paid sex in last 12 months, condom using status every time had sex with 2nd to most recent partner in last 12 months, and condom using status every time had sex with 3rd to most recent partner in last 12 months). On the other hand, levels of stigma and HIV knowledge were developed by composite questions to measure each appropriately. Stigma score from 7 questions was calculated and then groupings of the scores were performed. The variables from which scoring was done were whether the respondents ashamed if someone in the family had HIV, buy vegetables from a vendor with HIV positive, whether attending school together with children living with HIV is right or not, hesitation to take HIV test, afraid bad label about people with or believed to have HIV, afraid of a loss of respect from other people if positive for HIV and afraid to get HIV from contact with saliva from an infected person. The scores calculated from these questions range from 0 to 7 and were grouped into four classes. Consequently, score 0 shows no stigma, score from 1-3 shows low stigma, score 4 was considered to be a moderate stigma, and a score of 5-7 as high stigma. By a similar method, ‘knowledge about HIV’ was constructed by combining the responses to 9 sets of questions. Scores were computed from the question that asks whether the respondents ever heard about HIV/AIDS, whether the respondent knows about reducing the risk of getting HIV by always using condoms during sex, whether the respondent knows about reducing the risk of getting HIV by having 1 sex partner only, who has no other partners, whether the respondents answered right answer on transmissions of HIV via mosquito bites and by sharing food with a person who has AIDS, whether the respondents know that a healthy-looking person can have HIV, whether the respondents know that HIV can be transmitted during pregnancy, delivery, and breastfeeding. These 9 questions were added together and the minimum sum was 1 and with a maximum of 9. After scoring, the grouping of scores was performed to form the ‘HIV knowledge’ levels of the respondents. The average score was 7, and scores strictly less than average were categorized to have low ‘knowledge on HIV’ and the respondents with scores 7 and 8 were categorized to have high knowledge on HIV while respondents with score 9 were grouped as those having comprehensive knowledge of VCT HIV.
Other definitions, specific to this paper, were also given. Respondents in the age groups of 15-29 were considered as youths, those in the age group of 30 to 44 are adults and 45 to 59 were considered as late adults.
Agricultural workers: refers to those males who were market-oriented skilled agricultural workers, market-oriented skilled forestry, fishery and hunting, and agricultural, forestry, and fishery laborers.
Professional workers: include chief executives, senior officials, and legislators, administrative and commercial managers, production and specialized services managers, science and engineering professionals, health professionals, teaching professionals, business and administration professionals, information and communications technology professionals, legal, social and cultural professionals, science and engineering associate professionals, health associate professionals, business and administration associate professionals, legal, social, cultural and related associate professionals, information and communications technicians, teaching associate professionals and special education teaching associate professionals.
Trade or sales workers: encloses sales workers, building and related trades workers, excluding electricians, metal, machinery and related traders, handicraft and printing workers and, electrical and electronic traders.
Elementary occupation: covers cleaners and helpers, laborers in mining, construction, manufacturing and transport, food preparation assistants, street and related sales and service workers and, refuse workers and other elementary workers.
Others workers: consists of hospitality, retail, and other services managers, general and keyboard clerks, numerical and material recording clerks, other clerical support workers, personal service workers, personal care workers, protective services workers, handicraft and printing workers, food processing, woodworking, garment, and other craft and related trades workers.
Data analysis
Extraction of relevant variables, data exploration, cleaning, coding, and recoding, generating new important variables, descriptive and inferential statistics were obtained using Stata 14.2 statistical software. The multilevel logistic regression model was fitted to assess regional variation of HIV-VCT uptake and to identify its determinants. The DHS surveys often follow a hierarchical data structure as the surveys are based on multistage stratified cluster sampling (12). The appropriate approach to analyzing such survey data is, therefore, based on nested sources of variability which come from different levels of the hierarchy. Often, two responses from the same group tend to be more like one another than two observations from different groups and this association is called within-cluster correlation. Effects arising from the specification of correlation among responses from the same clusters are commonly known as Random Effects. Models used for the analysis of this type of data must account for associations among observations within clusters (levels) to make efficient and valid inferences. When the variance of the residual errors is correlated between individual observations as a result of these nested structures, ordinary logistic regression is inappropriate. The abovementioned background is a basis for using multilevel logistic regression in this analysis.
In regression analysis, not only the effect of a given explanatory variable is of interest but also the variation of this effect in the population might be invaluable. Oftentimes, this variation is captured via interactions. In classical regression, estimates of varying effects can be noisy, especially when there are few observations per group; multilevel modeling allows us to estimate these interactions to the extent supported by the data. Multilevel models also allow us to study effects that vary by group. In the analysis of multilevel regression, the clustering effect plays a great role in the estimation of the parameters and this clustering effect can be quantified by intraclass correlation (ICC). ICC is the proportion of total variation in the response variable that is accounted for by between-group variation. The ICC can also be thought of as the correlation among units within the same group, i.e the degree of homogeneity of the outcome within the group (18). In the current study all predictors are at level 1 and the clustering variable is a region where the subjects were residing.
All the outputs for descriptive analysis were done using weights provided in EDHS 2016 data as per the recommendations by the DHS program. The ways weights are used to vary based on the purpose of the analysis. To carry out multilevel analysis, the weights from DHS are adjusted as per the recommendation by Adam (19). Subsequently, we have checked the goodness of fit after weighting the dataset by both candidate weights. As expected, the multilevel logistic regression fitted by using the adjusted weights resulted in lower AIC = 9, 162.102, and BIC = 9, 233.47 as compared to the results from unadjusted weights with AIC=10, 492.85 and BIC=10, 564.22. In addition to the choice of weights, the principle of parsimony dictates us to go for the model with few numbers of variables in the model. Consequently, significant variables retained in our final model which has smaller AIC and BIC, and the results are presented in Table 3.
Ethical consideration
The EDHS 2016 survey protocol, including biomarker collection, was reviewed and approved by the Federal Democratic Republic of Ethiopia Ministry of Science and Technology and the Institutional Review Board of ICF International. Additionally, written consent was obtained from each respondent. For analysis, the investigators received permission from the public domain MEASUREDHS website and reanalyzed the dataset on male respondents.