Study data and sample group
The study used data from GDHS. More information about the surveys can be found at https://dhsprogram.com. The specific data used in this study were from the GDHS 2003, 2008 and 2014 individual datasets. These datasets contain data on women and issues related to children. Analyses were restricted to mothers (15-49 years) who had a child less than two years of age (0-23 months) and that child should have been living with the mother at the time of the survey. The GDHS datasets are publicly available at http://dhsprogram.com/data/dataset/Ghana from which data for this paper were obtained, and Measure DHS approved a concept note before the datasets were downloaded. The specific data used in this study was the various survey years’ individual datasets. These datasets contain details on, but not limited to, mothers characteristics, breastfeeding practices and childhood morbidity (diarrhoea). An appended weighted sample of 4,675 mother-child pairs was used for the analyses (Table 1). Also, more about the sample procedures used in the GDHS are available at http://www.dhsprogram.com/publications/publication-FR307-DHS-Final-Reports.cfm.
Table 1: Sample group
Age
|
Survey year
|
2003
|
2008
|
2014
|
0-5
|
307 (26.15)
|
307 (26.14)
|
561 (47.71)
|
6-11
|
370 (29.68)
|
297 (23.81)
|
579 (46.51)
|
12-23
|
662 (29.38)
|
530 (23.50)
|
1062 (47.12)
|
Total
|
1339 (28.64)
|
1134 (24.25)
|
2202 (47.11)
|
N= 4675
Exposure variable
The exposure variable, breastfeeding practice(s), which is hierarchically and mutually exclusive, was coded as no breastfeeding = 0; exclusive breastfeeding = 1; predominant breastfeeding = 2; and partial breastfeeding = 3. These categories are defined as follows, within 24 hours preceding each survey: (i) no breastfeeding – children were not breastfeeding; (ii) exclusive breastfeeding – children were only fed with breast milk; (iii) predominant breastfeeding – children were breastfed with breast milk including plain water, non-milk liquids and other milk; and (iv) partial breastfeeding – children were fed with breast milk plus complementary foods.
Outcome variables
The outcome variable was diarrhoea. Data on ‘all diarrhoea’ and ‘diarrhoea with blood’ were collected. Mothers were asked, “Whether any of their children under five years of age had diarrhoea during the two weeks preceding the survey”. The responses to this question were: No=0; Yes, last two weeks=1; Don’t know=8. All the “Don’t Know” responses on the dependent variables were excluded from the analysis. This was done because the study was only interested in responses that were captured on either a child had diarrhoea or no diarrhoea. The outcome variable was dichotomized, with No = 0 and Yes = 1 format.
Explanatory variables
The review of related literature and extensive data mining informed the choice of afore-mentioned explanatory variables. These variables included: maternal education, working status, place of residence, preceding birth interval, source of drinking water, type of toilet facility, and type of floor material. The explanatory variables were arbitrarily assigned numeric codes starting with either ‘0’ or ‘1’.
Maternal education was categorized into no education, primary, and secondary or higher. For the preceding birth interval, births of mothers were recoded into those who had less than 24 months and those of 24 months or more. Environmental variables that may help explain diarrhoea episodes in children include a source of drinking water, type of toilet facility, and type of floor material. The various responses given by mothers, in the aforementioned variables, were recoded as either ‘improved’ or ‘unimproved’. Responses for source of drinking water were categorized as follows: improved sources (piped into dwelling, piped to yard/plot, public tap or standpipe, tube-well or borehole, protected well, protected spring, rainwater); and unimproved sources (unprotected spring, unprotected well, river/dam/lake/ponds/stream/canal/irrigation, tanker-truck, cart with small tank, bottled water sachet).
For type of toilet facility, improved facilities included flush to piped sewer system, flush to septic tank, flush to pit latrine, flush to somewhere else, ventilated improved pit latrine (VIP), pit latrine with slab, composting toilet, and a special case (i.e., flush/pour flush of excreta to a place unknown to the respondent); and unimproved facilities consisted of flush/pour flush to elsewhere (such as street, yard/plot, open sewer or a ditch), pit latrine without slab, bucket, hanging toilet or hanging latrine, shared sanitation, no facilities, bush or field). The above categorizations are in accordance to World Health Organisation (WHO) and United Nations Children’s Fund (UNICEF) Joint Monitoring Programme (JMP) for Water and Sanitation [10].
Furthermore, type of floor material was grouped as improved (tablets/wood/planks, palm/bamboo, mat, adobe, parquet/polished wood, vinyl/asphalt strips/ floor mat, linoleum, ceramic tiles/mosaic, cement, carpet, stone, bricks); and unimproved (earth, sand, clay, mud, dung). These categorizations conform with the Demographic and Health Surveys (DHS) programme [11].
Data analysis
Prevalence rates of diarrhoea were initially examined among the explanatory variables and the chi-square test of independence was used to identify significant associations. Only independent variables that had a p-value less than 0.25 at the chi-square test were considered for the logistic regression. Applying hierarchical logistic regression modelling procedures, categories of significant factors associated with diarrhoea were ascertained. To finally examine the interactive nature of the outcome, exposure and explanatory variables, a single multilevel logistic modelling approach was applied. Thus, logistic regression analyses were performed to assess the effect of explanatory variables that were considered as potential confounders on the association between breastfeeding practices and childhood diarrhoea. Essentially, the various explanatory variables were nested in the exposure variable to estimate their effect on each of the outcome variable (diarrhoea). Probability values (p-values) of 0.01, 0.05 and 0.001 were used to establish significant associations. In generating these models, clustering effects were accounted by prefixing each STATA command with svy to estimate for robust standard errors. A survey design weight of v005, primary sampling unit of v021 and strata of v023 were used in the survey set statement. Results were presented in odds ratios and 95% Confidence Interval (CI) STATA (version 12.0) was used to process the data.