Predictors of utilisation of skilled maternal healthcare in Lilongwe District, Malawi

Background: Despite numerous efforts to improve maternal and child health in Malawi, maternal and newborn mortality rates remain very high, with the country having one of the highest maternal mortality ratios globally. The aim of this study was to identify which individual factors best predict utilisation of skilled maternal healthcare in a sample of women residing in Lilongwe district of Malawi. Identifying which of these factors play a significant role in determining utilisation of skilled maternal healthcare is required to inform policies and programming in the interest of achieving increased utilisation of skilled maternal healthcare in Malawi. Methods: This study used secondary data from the Woman’s Questionnaire of the 2010 Malawi Demographic and Health Survey (MDHS). Data was analysed from 1126 women aged between 15 and 49 living in Lilongwe. Multivariate logistic regression was conducted to determine significant predictors of maternal healthcare utilisation. Results: Women’s residence (P =.006), education (P =.004), and wealth (P =.018) were significant predictors of utilisation of maternal healthcare provided by a skilled attendant. Urban women were less likely (odds ratio [OR] = 0.47, P =.006, 95% CI = 0.28-0.81) to utilise a continuum of maternal healthcare from a skilled health attendant compared to rural women. Similarly, women with less education (OR = 0.32, P =.001, 95% CI = 0.16-0.64), and poor women (OR = 0.50, P =.04, 95% CI = 0.26-0.97) were less likely to use a continuum of maternal healthcare from a skilled health attendant. Conclusion: Policies and programmes should aim to increase utilisation of skilled maternal healthcare for women with less education and low-income status. Specifically, emphasis should be placed on promoting education and economic empowerment initiatives, and creating awareness about use of maternal healthcare services among girls, women and their respective communities. This study used Andersen’s Behavioural Model of Health Services Use as its conceptual basis. Andersen’s model is one of the most widely acknowledged multilevel conceptual frameworks that incorporate both individual and contextual determinants of utilisation of health services.38 Andersen’s model has been used extensively in health services research to investigate factors that lead to utilisation of health services,38 and to evaluate the extent to which health services are equitably distributed or accessed.39 The model has been used, for example, to analyse determinants and patterns of healthcare utilisation,40,41 and to understand health-seeking behaviors.42 It has also been used to assess disparities in healthcare utilisation.43 Aday and Andersen argue that utilisation of health services is determined by characteristics of the population at risk. These characteristics can be conceptualised as comprising 3 components, ie, predisposing, enabling, and need components. The predisposing component includes variables that describe the “propensity” of individuals to use services such as age, gender, race, and religion.44 The enabling component describes the “means” available to individuals for the use of services. Examples of enabling factors include income, family support, access to health insurance, and attributes of the community in which the individual lives (urban-rural settings). The need component refers to illness level, which is the most immediate cause of health service utilisation. The need for healthcare may be either that perceived by the individual or that evaluated by the delivery system. Data analysis was conducted using secondary data from the 2010 Malawi Demographic and Health Survey (MDHS).45 The 2010 MDHS is a nationally representative sample comprising 27 000 households and involving 24 000 female and 7000 male respondents, to provide data to policy-makers, planners, researchers, and programme managers.45 This study used data from the Woman’s Questionnaire of the MDHS. The Demographic and Health Survey (DHS) comprises a wealth index. The DHS wealth index is calculated using households’ cumulative living standards, classifying households according to 5 wealth quintiles; calculated using data on households’ ownership of selected assets, including televisions and bicycles, and types of access to water and sanitation facilities.46 The 5 wealth quintiles comprise “poorest,” “poorer,” “middle,” “richer,” and “richest.”47 The study conducted a secondary data analysis of 2010 MDHS data from 1126 women aged 15-49 in Lilongwe district. As Malawi’s capital city, Lilongwe was chosen as a case study as it is the country’s largest and fastest urbanising city with a diversity of cultures, ethnicities, and healthcare practices, therefore reflecting the diversity at national level. The population of Lilongwe in 2017 was estimated as 1.1 million.48 The sample was selected using a stratified two-stage cluster design, with enumeration areas (EAs) being the sampling units for the first stage and households as a second stage of sampling, as detailed in the MDHS.45 The sampling frame was based on EAs of the 2008 Malawi Population and Housing Census. The sampling frame was stratified into 27 districts in the country; and within each of the districts, EAs were further stratified by urban and rural areas. A fixed number of 20 households were selected in urban and 35 households in rural primary sampling units. In the selected households, a total of 23 748 women aged 15-49 were eligible, and 23 020 women (97%) were interviewed. Among these, 1126 women were from Lilongwe, thus the sample size for this study. Figure presents a flow diagram of the sampling process. Demographic characteristics of respondents are presented in Table 1. Flow Diagram of Sampling Process. Abbreviation: MDHS, Malawi Demographic and Health Survey. The study focused on several key questions in the MDHS in relation to use of ANC, delivery and postnatal care services. In the MDHS, women who had given birth in the 5 years before the survey were asked questions regarding their care. Specifically, women reported if they received ANC, delivery care and postnatal care from a skilled attendant (ie, doctor, clinical officer, nurse or midwife) for their most recent live birth within the 5 years preceding the survey. Furthermore, women reported the number of visits made to the ANC clinic, and the timing (trimester) of the first ANC visit during the most recent pregnancy within the 5 years preceding the survey. Respondents also reported the place of delivery during their most recent birth in the 5 years preceding the survey. Moreover, women reported the timing after delivery of their first postnatal check-up. Permission to use the data was obtained from the DHS Programme website. Ethical approval for this study was granted by the Health Policy and Management/Centre for Global Health Research Ethics Committee at Trinity College Dublin, Ireland, and the College of Medicine Research Ethics Committee (COMREC) at the University of Malawi. The study comprised eight independent variables, ie, age, marital status, residence (urban/rural), education, work status, economic status (wealth), ethnicity and religion. Selection of independent variables was guided by Andersen’s Behavioural Model of Health Services Use and a review of the literature. The variables were categorised according to the MDHS dataset. Dependent variables comprised utilisation of ANC, delivery care, and postnatal care by a skilled attendant. A maternal healthcare index was created to indicate the continuum of maternal healthcare, and was generated by combining data for utilisation of ANC, delivery and postnatal care provided by a skilled attendant. The outcome variables were coded as 1 if the woman received maternal healthcare from a skilled attendant and as 0 if she did not receive maternal healthcare from a skilled attendant. The response category was collapsed to create dichotomous dependent variables with only 2 categories, ie, “yes” and “no,” on the basis of whether or not the woman had received maternal healthcare from a skilled health attendant. Variable categories with small sample sizes were combined with another category to enable meaningful analysis. For example, the sample sizes for some ethnic and religious groups were very small, and as such they were combined with others. If the respondent reported more than one person providing assistance during delivery, the most qualified person was used for analyses. Adebowale and Udjo similarly used a maternal healthcare index in their study of the relationship between infant mortality and a maternal healthcare services access index in Nigeria, using data from the 2013 Nigerian DHS; whereby their maternal healthcare services access index was created using variables including antenatal visit, antenatal attendance, tetanus injection during pregnancy, place of delivery, and birth attendance.49 Descriptive analyses were conducted of demographic characteristics of respondents and the use of maternal healthcare services (as presented below in the Results section). Pearson chi-square tests were conducted to explore the relationship between demographic characteristics and use of skilled maternal healthcare services. Statistical significance was established at P values of <.05. Direct logistic regression was conducted to determine predictors and factors that were significantly associated with utilisation of maternal healthcare services provided by a skilled health attendant. Analysis was conducted using IBM SPSS Statistics 22. Multivariate logistic regression was conducted to determine predictors of maternal healthcare utilisation. Logistic models were developed using a three-step approach. The first step comprised running a series of univariate binary logistic regression models in which the relationship of each independent variable with respective dependent variables was examined. Independent variables that were significant (P < .1) were considered important for inclusion in the next step of the process. The second step comprised collinearity diagnostics for independent variables chosen from the first step, as logistic regression is sensitive to multicollinearity. A cut-off value of 10 for the Variance Inflation Factor was used to determine the presence of multicollinearity.50 No multicollinearity was identified. Third, binary logistic regression models were run using variables from the first step, having been assessed for multicollinearity. Models used the ENTER method. Of the eight independent variables, six variables (age, marital status, residence, education, work status, and wealth) were treated as potential confounders and were included in binary logistic regression models for use of skilled attendance for maternal healthcare, whether or not they had attained statistical significance in the first step. Ethnicity and religion were excluded in the regression models as potential confounders, based on a review of the literature. This stage represented the multivariate model. Statistical significance for the final models was set at P <.05. The regression model was tested for ‘goodness of fit’ using the Hosmer-Lemeshow test. A good model fit is indicated by a Hosmer-Lemeshow test statistic value of P >.05. The Nagelkerke R Square was used as an indication of the variation in the dependent variable that could be explained by the model.50 Statistical significance was established at P values of <.05.