Determinants of maternal healthcare utilisation among pregnant women in Southern Ethiopia: a multi-level analysis

Background: Despite efforts to make maternal health care services available in rural Ethiopia, utilisation status remains low. Therefore, this study aimed to assess maternal health care services’ status and determinants in rural Ethiopia. Methods: The study used quasi-experimental pre- and post-comparison baseline data. A pretested, semi-structured, interviewer-administered questionnaire was used to collect data. A multilevel, mixed-effects logistic regression was used to identify individual and communal level factors associated with utilisation of antenatal care (ANC), skilled birth attendance (SBA), and postnatal care (PNC). The adjusted odds ratio (AOR) and corresponding 95% confidence intervals (CI) were estimated with a p-value of less than 0.05, indicating statistical significance. Results: Seven hundred and twenty-seven pregnant women participated, with a response rate of 99.3%. Four hundred and sixty-one (63.4%) of the women visited ANC services, while 46.5% (CI: 42–50%) of births were attended by SBA, and 33.4% (CI: 30–36%) had received PNC. Women who reported that their pregnancy was planned (aOR = 3.9; 95% CI: 1.8–8.3) and were aware of pregnancy danger signs (aOR = 6.8; 95% CI: 3.8–12) had a higher likelihood of attending ANC services. Among the cluster-level factors, women who lived in lowlands (aOR = 4.1; 95% CI: 1.1–14) and had easy access to transportation (aOR = 1.9; 95% CI: 1.1–3.7) had higher odds of visiting ANC services. Moreover, women who were employed (aOR = 3.1; 95% CI: 1.3–7.3) and attended ANC (aOR = 3.3; 95% CI: 1.8–5.9) were more likely to have SBA at delivery. The likelihood of being attended by SBA during delivery was positively correlated with shorter travel distances (aOR = 2.9; 95% CI: 1.4–5.8) and ease of access to transportation (aOR = 10; 95% CI: 3.6–29) to the closest healthcare facilities. Being a midland resident (aOR = 4.7; 95% CI: 1.7–13) and having SBA during delivery (aOR = 2.1; 95% CI: 1.2–3.50) increased the likelihood of attending PNC service. Conclusions: Overall, maternal health service utilisation is low in the study area compared with the recommended standards. Women’s educational status, awareness of danger signs, and pregnancy planning from individual-level factors and being a lowland resident, short travel distance to health facilities from the cluster-level factors play a crucial role in utilising maternal health care services. Working on women’s empowerment, promotion of contraceptive methods to avoid unintended pregnancy, and improving access to health care services, particularly in highland areas, are recommended to improve maternal health service utilisation. We analysed data from a baseline survey from a quasi-experimental pre- and post-comparison study. Ten villages (kebele) were selected randomly from 29 villages of Arba Minch zuria district after stratifying the district in climatic zones in to high land and low land areas. Arba Minch town, Gamo Zone’s capital, is located 502 kms south of Addis Ababa, Ethiopia’s capital (Fig. 1). The total population of the district for 2017 was 195,858, with 50% of the population being female. Nine of the 10 villages included in the study were from the Arba Minch Health Demographic Surveillance site (AM-HDSS). This population primarily engages in subsistence agriculture, crop farming and small-scale animal rearing. Map of the study area Similar to the rest of Ethiopia, Gamo Zone uses a three-tier health service delivery system, comprising primary, secondary and tertiary levels of care. The primary health care unit comprises a health centre with up to five health posts attached to it. The district has seven health centres and 40 health posts. At the health centre level, basic emergency obstetric care services are offered. Typically, health officers, midwives and nurses staff each health centre, and when they are faced with obstetric complications, they refer cases to Arba Minch General Hospital, where comprehensive emergency obstetric care is offered. The target population comprises pregnant women from Arba Minch Zuria district in Gamo Zone who had at least one birth in the past five years preceding the survey. Three maternal health care service utilisation were considered – having received ANC during the pregnancy of the most recent birth, having given birth to the previous baby at a health facility and receiving any postnatal care following the previous childbirth – which were viewed as primary outcomes. We assessed predictors of each outcome separately, with respect to the previous birth. Explanatory factors were found at three levels: individual; household; and community. The use of ANC, BPCR awareness and BPCR practices were included as explanatory factors while examining factors linked with SBA. Similarly, utilisation of ANC and SBA, along with the other variables, was included in the analysis of PNC utilisation. Each explanatory variable’s coding is provided in Table 1. Description of the variables and measurements for the multilevel logistic regression analysis of the determinants of maternal health care utilization in southern Ethiopia The sample size was estimated based on quasi-experimental study design. SBA’s prevalence in an Ethiopian rural setting was estimated to be 43% from EDHS 2019 [9]. The sample was based on 80% power to detect a change of 10% (with a 5% error level) and a design effect of 1.5. A sample size of 50 pregnant women per cluster was required, and accordingly, the final calculated sample size after considering a 10% loss to follow-up was 392 subjects per group (784). The study included all nine villages of the Arba Minch-HDSS and one village from the Arba Minch Zuria district. Table ​Table22 provides the characteristics of the 10 clusters/kebele included in the study. AM-HDSS was modelled after a stratified (agro-ecology), two-stage, cluster-sampling technique. We conducted a census to identify all eligible pregnant women from the selected villages and identified 1,447 women who were pregnant in 2017 from 10 villages (< 27 weeks of gestation). Altogether, 732 who fulfilled the inclusion criteria were interviewed for the study. Individuals who did not have a plan to move from where they were living until the end of the follow-up period were included in the baseline study (727), and they were all married. The character of the eleven clusters /kebele included in the study, Gamo zone 2020 *Number of Households per cluster The data were collected using a pretested, semi-structured, interviewer-administered questionnaire comprising five main components: socioeconomic and demographic variables; obstetric characteristics; birth preparedness and complication readiness (BPCR) practices; maternal health care characteristics (places of births, number of ANC and PNC visits); and attitudes towards the importance of skilled birth attendance and maternity waiting homes. The EDHS suggested that household-asset questions be used to construct the household wealth index [9]. Face-to-face interviews were conducted in the local language by trained data collectors who received three days of training. Eleven data collectors and six site supervisors working in the district participated in the data collection. Data were collected on a smartphone-based application, Open Data Kit (ODK). After a completeness check, the data were uploaded to a secure database in ODK Aggregate, and ODK briefcase software was used to retrieve the data in Comma Separated Variables file format. Data were cleaned and analysed using Stata (StataCorp, Version 16, College Station, TX, USA). The questionnaire was written in English, then translated into Amharic and back to English to ensure that the translation accurately represented the original meaning. Before data collection, the questionnaire was pre-tested in a village in the Arba Minch Zuria district, whose demographic characteristics resembled those of the sample population, then modified based on the results. Inter-item consistencies between the variables were tested for the variables creating each component of items to measure participants attitude towards skilled birth attendance using the Cronbach’s alpha (all variables were > 0.86). Sociodemographic, economic, and obstetric data were presented using descriptive statistics. Bivariate analysis was also used to look into the relationship between the outcomes and the explanatory variables. For the multi-level analysis, we included variables with a p-value of 0.05 in the bivariate analyses with 95% CI. A separate multi-level model was constructed for every one of the three outcome variables. Due to the clustering of the data—individuals were nested within households, and households were nested in communities—we have chosen a multi-level analysis instead of conventional logistic regression. Given that no two communities, or kebeles, have the same variance, the likelihood of MHC uptake differs significantly among communities. Therefore, a four-level multilevel model was fitted based on these attributes of the data. The first level was the null model (Model 0), which had no exposure variables and was designed to look for community variance and provide evidence for assessing the random effects of MHC utilisation at the community level. Model I was a multivariable model that was adjusted for characteristics at the community level; Model II incorporated factors at the individual level; and Model III, the final model, was fitted by taking this datum characteristic into account both individual and community-level factors as the outcome variable. A forward stepwise approach was followed until we reached the final model. The measure of association (fixed effects) was estimated and expressed as an aOR with a 95% CI. Regarding the measures of variation (random effects), community-level variance, intra-cluster correlation coefficient (ICC), and median odds ratio (MOR) were used. The ICC, which quantifies the proportion of observed variation (variance partitioning) in the outcome that is attributable to the effect of clustering and was computed using the formula ICC = Va/(Va + π2/3), where: Va community level variance and the unobserved individual variable follows a logistic distribution with individual level variance equal to π2/3 (i.e., 3.29) [22]. We have also calculated the median odds ratio (MOR = exp( √2 × Φ–1 (0.6745))) [23] to quantify the variation between clusters by comparing two persons with the same covariates from two randomly chosen, different clusters. Where Φ(·) is the cumulative distribution function of the normal distribution with mean 0 and variance 1, Φ–1(0.75) is the 75th percentile, and exp(·) is the exponential function. We computed the MOR using a Stata command: nlcom exp(sqrt(2*_b[/var(_cons[kebele_BaselineM])])*invnormal(0.75)), cformat(%9.2f) and the result was greater than 1 which showed a considerable between-cluster variation to proceed with the multilevel analysis [24]. In addition, the Akaike information criteria (AIC) and loglikelihood model were used in comparison with other models to estimate models goodness-of-fit (Table ​(Table55). Multilevel logistic regression analysis of the determinants of MHC services utilization among pregnant women in southern Ethiopia,2022 Dependent variable: ANC/SBA/PNC; (ii) cluster variable: kebele (9 in number); Model 1: cluster level variables included; Model 2: individual level variables included; Model 3: full model (all the cluster level and individual level variables included; * p < 0.05 shows statistically significance; VPC = Variance Partition Coefficient; AIC = Akaike Information Criterion Collinearity was assessed using the means of variance inflation factors (VIFs) as a post-estimation procedure, following for regression analysis because this study used several explanatory variables that might be correlated to each other (such as mother’s education, husband’s education and the household wealth index). The multi-collinearity between both parents’ education levels was found to be significant, so the husband’s education was excluded from the regression models (Variance Inflation Factor > 5).