Study design and study setting
This was a cross-sectional survey carried out to provide estimates of basic demographic and health indicators for the Malawi nation. The study was nationally representative, with all 28 districts included in the sample, stratified by residential area (urban and rural). It yielded 56 sampling strata.
Weighted sample was selected in two stages. In the first stage, 850 standard enumeration areas (SEAs) also known as clusters (173 SEAs in urban areas and 677 in rural areas) were selected with probability proportional to the SEA size. The list of households in each selected SEA served as the sampling frame for the second stage. Some of the selected SEAs were large, hence to reduce the work of household listing, each large SEA (more than 250 households) was segmented. From each of the segmented SEAs, one segment was selected for the survey with probability proportional to the segment size. In the second stage, a fixed number of 30 households per urban cluster and 33 per rural cluster were randomly selected from each cluster’s household listing. A total of 27,516 households were selected, of which 26,564 were occupied. Among the occupied households, 26,361 were interviewed (99% response). From the interviewed households, 25,146 eligible women were identified for individual interviews and only 24,562 women were successfully interviewed (98% response). All women aged 15-49 years old were interviewed if they were permanent residents of, or had stayed in the household in the previous night before the survey in the period between October 2015 and February 2016. A detailed methodology has been published elsewhere [46].
Study population
The target population was all Malawi resident women aged 15-49 years with live births one-year-old or less preceding the survey, between the years of 2015 and 2016. The study was based on data from the 2015-16 Malawi Demographic and Health Survey. Malawi National Statistical Office conducted the survey from 19 October 2015 to 18 February 2016
Inclusion and exclusion criteria
The analysis included all women age 15-49 with live children born at least in August 2015, because the implementation of the updated policy was nationally full-fledged in October 2014. On the other hand, study participants who had missing data on at least one of the key variables used in the analysis were excluded from the analysis.
Sample selection
Out of 24, 562 women who completed the survey interviews, 6,586 had live birth in the 2 years before the survey. Of the 6,586 women, 1,219 had children who were born after the month of July 2015. About 120 women were excluded from analysis because of lack of information on at least one variable used in the analysis. The final sample size used in the analysis was 1, 084 (unweighted) and 1,069 (weighted) (Figure 1).
Insert Figure 1 here
Data extraction and variables used in the analysis
Guided by literature review, data were extracted for analysis based on variables that were theoretically and empirically linked to uptake of IPTp-SP as follows: (a) the outcome variable was IPTp-SP uptake, categorised as two doses or less (≤2 doses) and three doses or more (3+ doses, also known as optimal dose); (b) explanatory variables were woman’s residential area, woman’s level of education, woman’s age, woman’s occupation, wealth, marital status, region, parity, timing of the first ANC visit, and number of ANC visits (Table 1). The data covering these variables were extracted from the 2015-16 MDHS women dataset. The data were collected using the woman’s questionnaire.
Table 1: Variables used in the study
|
Outcome variable
|
Definition
|
Category
|
IPTp-SP uptake
|
Two or less (≤2) doses is incomplete and three or more (3+) is optimal
|
≤2 doses
|
3+ doses
|
Explanatory variable
|
|
|
Residence
|
Area of woman’s residence (urban or rural)
|
Rural
|
Urban
|
Education
|
Level of education of woman
|
No formal education
|
Primary
|
Secondary or higher
|
Age
|
Age group of a woman
|
15-19
|
20-24
|
25-29
|
30-34
|
35-39
|
40+
|
Occupation
|
Woman’s occupation
|
Unemployed
|
Self-employed
|
Employed
|
Wealth
|
Household’s wealth from which a woman is an occupant
|
Poorest
|
Poorer
|
Middle
|
Richer
|
Richest
|
Marital status
|
Marital status of a woman
|
Married
|
Divorced/separated/ widowed
|
Never married
|
Region
|
|
Southern
|
Central
|
Northern
|
Parity
|
Number of birth that a woman had after 20 weeks gestation
|
One child
|
Two children
|
Three children
|
Four children
|
Five or more children
|
Timing of 1st ANC visit
|
Age (in months) of the pregnancy a woman visited antenatal care clinic (ANC) for first time
|
1st trimester (1-12 weeks)
|
2nd trimester (13-26 weeks)
|
3rd trimester (27+ weeks)
|
Number of ANC visits
|
Number of antenatal care visits a pregnant woman made during her gestation period
|
4+
|
3
|
1-2
|
Data management
Data extraction, cleaning, and analysis were done using Stata version 16 (Stata Corp, College Station, Tx, USA).
Data analysis
Proportions and frequencies were used to summarize categorical variables in descriptive analysis. On the other hand, bivariate and multiple logistic regressions were used in analytical analysis.
Four stages of logistic regression modelling of survey data were applied as specified by Heeringa, West and Berglund [47]; and Hosmer, Lemeshow and Sturdivant [48]. First, bivariate analyses of the relationship of outcome to individual explanatory variable candidates were performed. Second, explanatory variables that had a bivariate association with the outcome at significance were selected as candidates for the main effects in a multivariate logistic regression model. An initial model-building process using multivariate logistic regression analysis was done to further examine the association (measure of effect) between the outcome and each explanatory variable while controlling the effects of other explanatory variables. The model estimated adjusted odds ratios (AOR). The level of significance used was 5% (0.05), two-tailed at 95% confidence interval (CI). Third, the contribution of each explanatory variable to the multivariate model was evaluated using Wald test at 5% significant level (Table 2). Table 2 shows one of the six adjusted Wald tests is statistically significant and the predictor is “number of ANC visits”, in this initial model. This suggests that the parameters associated with the number of ANC visits in this logistic regression model are significantly different from zero and that the variable may be an important predictor of uptake of at least three doses of SP when adjusting for the relationships of the other predictor variables with the outcome. Therefore, at this stage in the model-building process, only the ‘number of ANC visits’ variable was retained of all of the candidates’ main effects. Thus, the second model-building process included the ‘number of ANC visits’ predictor only, which had similar odds ratios as the adjusted one. Lastly, scientifically justified interactions among the explanatory variables were also checked and there were no significant interactions observed.
The statistical analysis took into account the complex characteristics of the survey sample design by allowing adjustments for stratification, clustering and weighting for unequal selection probabilities.
Table 2: Design-Adjusted Wald tests for the parameters associated with categorical predictors in the initial multiple logistic regression model
|
Predictor
|
F-Test Statistic
|
P-value
|
Education level
|
F(2,563) = 0.30
|
0.7399
|
Age group
|
F(5,560) = 0.86
|
0.5088
|
Occupation
|
F(2,563) = 0.14
|
0.8736
|
Wealth status
|
F(4,561) = 0.92
|
0.4527
|
Parity
|
F(4,561) = 0.37
|
0.8316
|
Timing of 1st ANC visit
|
F(2,563) = 1.23
|
0.2925
|
Number of ANC visits
|
F(2,563) = 25.01
|
<0.001*
|
* Design-adjusted Wald tests significant at the 0.05 level
|
Ethical considerations
The Malawi Demographic and Health Survey protocol was reviewed and approved by Malawi’s National Health Sciences Research Committee and Inner City Fund (ICF) Institutional Review Board [46]. Interviewers informed prospective participants about the purpose of the study, procedures required of them if recruited, and that they had the right to volunteer whether or not to participate in the study [46]. Informed consent was obtained from each participant before administering the questionnaire and the respondents were assured of privacy and confidentiality [46]. To access the survey datasets, the author obtained permission from the Demographic and Health Surveys (DHS) Program. The datasets received were treated as confidential and no effort was made to identify any household or individual respondent interviewed in the survey. In addition, no ethical clearance was sought from Malawi’s National Health Sciences Research Committee because the research material collected for MDHS were not used differently in this study as stipulated in National Health Sciences Research Committee guidelines.