Multilevel Models for Determinants of Maternal Health Services Utilization in Uganda Using 2016 DHS Data

Abstract

Introduction: Studies on determinants of maternal health services such as Antenatal Care (ANC), Delivery care (that is, Facility delivery or facility-based delivery) and postnatal care (PNC) in Uganda have all applied conventional methods of data analysis such as chi-square and logistic regression analysis irrespective of the hierarchical structure of the population. This leads to biased estimates due to the clustering effect within the population. To fully understand and provide reliable statistics for policy and planning, this study investigated the determinants of maternal health services utilization (MHSU) using multilevel modeling. It further determined out the effect of four or more ANC visits on facility delivery and early postnatal care utilization using propensity score matched analysis (PSMA). Methods: Secondary analysis of data from the Uganda Demographic and Health Survey (UDHS) of 2016 among women who gave birth three years preceding the survey, aged 15 – 49 years was done. UDHS data are collected based on multistage sampling technique and thus data are clustered within primary sampling units (called communities in this study). The data analysis started from the descriptive analysis of the sample statistics, followed by bivariate analysis between the outcome variables and the determinants to identify significant associations between the outcomes and determinants that were used in multivariable analysis. Multilevel modelling began by fitting a null model and used it to check whether multilevel modelling was necessary using the Likelihood Ratio Test (LRT) and intra-class correlation (ICC). Level 1 (individual), level 2 (household) and level 3 (community) models were then fit. For the number of ANC visits (0 – 1 ANCs = 0, 2 – 3 ANCs = 1, 4 or more ANCs = 2), a multilevel ordinal logistic regression model was fit. For facility delivery (yes = 1, no = 0), the multilevel binary logistic model was fit. For timing (less than 48 hours = 1 or 48 hours or more and no PNC = 0) of postnatal care services utilization, the multilevel binary logistic model was fit. To examine the variations in individual and household level determinants on MHSU across type of place of residences, analysis was done by stratification by rural and urban residences. PSMA using radius caliper and kernel matching was used to obtain the Average Treatment Effect of the Treated (ATT). Results: The study found out that mother’s level of education (secondary or higher) [AOR = 1.37, 95% CI: 1.12, 1.68], mass media exposure [AOR = 1.27, 95% CI: 1.10, 1.47], last birth caesarean [AOR = 1.47, 95% CI: 1.08, 2.00] and high household wealth index [AOR = 1.39, 95% CI: 1.14, 1.70] were all associated with ANC utilisation. Mother’s level of education was also found to be associated with both facility delivery [AOR = 2.03, 95% CI: 1.51, 2.73] and EPNC [1.50, 95% CI: 1.13, 2.00]. Four of more ANC visits is also associated with both facility delivery [AOR = 4.87, 95% CI: 3.29, 7.21] and EPNC [AOR = 17.39, 95% CI: 13.84, 21.85]. For factors associated with utilisation of maternal health services by type of place of residence, the study found out that mother’s level of education influenced ANC utilisation in urban [APR = 1.65, 95% CI: 1.02, 2.65] and in rural [APR = 1.58, 95% CI: 1.22, 2.05] areas respectively. Four or more ANC visits was found to be positively associated with facility delivery in urban [APR = 1.34, 95% CI: 1.01, 1.78] and in rural [APR = 1.87, 95% CI: 1.54, 2.26]. Facility delivery is strongly associated with utilisation of EPNC in urban [APR = 4.40, 95% CI: 2.55, 7.62] and in rural [APR = 6.03, 95% CI: 4.98, 7.29]. The probability of facility delivery and EPNC is [ATT = 0.118, 95% CI: 0.063, 0.173] and [ATT = 0.099, 95% CI: 0.076, 0.121] respectively higher among women who have had 4 or more ANC visits, to the same women had they not had 4 or more ANC visits. The probability of EPNC was [ATT = 0.518, 95% CI: 0.489, 0.547] among women who have had facility delivery compared to those who have not. Conclusions: Different woman and household level factors such as mother’s level of education, employment status, distance to health facility, modern contraceptive use, mass media exposure, birth order number, household wealth index, and community level factors, that is type of place of residence and region are associated with maternal health services utilisation in Uganda. The results from propensity score matched analysis illustrate the need for implementation of policies towards provision of ANC services (at least four ANC visits) since it plays an effective intervention to increase facility-based delivery and ultimately early postnatal utilisation.