Magnitude and associated factors of unintended pregnancy in Ethiopia: a multilevel analysis using 2016 EDHS data

This study was based on the nationally representative 2016 Ethiopian Demographic and Health Survey (EDHS) which was conducted from January 18, 2016, to June 27, 2016.

The survey used the Ethiopia Population and Housing Census (PHC) conducted in 2007 as a sampling frame. The frame is a complete list of 84,915 enumeration areas (EAs) created for the 2007 PHC. The sample used for the survey was stratified and selected using two stages. In the first stage, a total of 645 EAs (202 in urban areas and 443 in rural areas) were selected with probability proportional to EA size. In the second stage, 28 households per cluster were selected with an equal probability systematic selection. Detailed information was collected on issues related to the reproductive health (fertility and fertility preference, marriage, awareness and the use of family planning methods), adult and childhood morbidity and mortality as well as awareness and attitudes towards HIV/AIDS and other important public health issues from 16,650 households, 15,683 female respondents, and 12,688 male respondents. Further information regarding the sampling technique and questionnaire, in general about the survey, can be obtained from the EDHS 2016 report [31].

Our analysis was based on women’s data (IR data) set of EDHS 2016 which was accessed from the Measure demographic and health survey (DHS) program website (http://​www.​measuredhs.​com). A total weighted sample of 7590 reproductive age women (15–49 years) who gave birth in the 5 years preceding the survey was used.

The dependent variable was unintended pregnancy which includes pregnancies that are wanted no more or wanted later. For this study, the independent variables were classified as individual level and community level variables. The individual-level variables were maternal age, maternal educational status, maternal occupation, marital status, religion, parity, household size, wealth status, media exposure, ever had of terminated pregnancy and knowledge of ovulation cycle. The community-level variables in this study were the perception of distance from the health facility, residence, and region. Some of the variables were re-coded to make suitable for the study such as wealth status, re-coded as poor = 0, middle = 1, rich = 2 whereby women in the poorest and poorer category were referred to as the “poor”; those who were above the poorer wealth category but not up to those in the high category were categorized as middle whilst those in the high wealth category (rich and richest) were recoded as rich. The other variable recoded was region in which; Afar, Somali, Benishangul, and Gambela as “small peripheral regions”, Tigray, Amhara, Oromia, and Sothern Nations Nationalities and Peoples Region (SNNPR) as “large central regions” and Harari, Dire Dawa, and Addis Ababa as “metropolitans”, based on their geopolitical features consistent with a previous study from Ethiopia [32]. Other important independent variables were also recoded or recategorized to make suitable for analysis.

The extraction of data, coding, and analysis was done using Stata 14 software. The data were weighted with the available sample weight factor (v005) within the EDHS dataset to minimize the effect of sampling bias. A multi-level logistic regression analysis was used to account for the hierarchal nature of the DHS data and bivariable multilevel logistic regression analysis was performed to estimate the crude odds ratios at 95% confidence interval and those variables with p-value < 0.20 were considered for multivariable analysis. In the multivariable multilevel logistic analysis, those variables with p-value < 0.05 were considered to be significantly associated with unintended pregnancy.

After selecting variables for multivariable analysis, four models containing variables of interest were fitted. These models were model 1 without explanatory variables, model 2 which examined the effects of individual-level characteristics, model 3 examined the effect of community-level variables and model 4 which examined the effects of both individual and community level characteristics simultaneously. The random effect results were estimated using three different methods, the Intra-Class Correlation (ICC), median odds ratio (MOR), and proportional change in variance (PCV). Since these models were nested deviance was used to assess the model fitness.

Table 1 summarizes the background characteristics of the study participants. More than two-thirds of study participants were between 20 and 34 years of age with an overall mean age of 29.3(±6.8) years. The majority (63.1%) of respondents had no formal education and 43.5% of women had poor wealth status. Most, 93.7% of women were married while two-third of them had no media exposure. Regarding household size, half of the respondents were from a household size of four to six. The majority of the women were from rural areas and large peripheral regions.

In this study 73.4% [95%CI: 72.4, 74.4] of pregnancies were intended and 26.6% [95%CI: 25.6, 27.6] pregnancies were unintended (Fig. 1).

Table 2 showed the random effect model. In the null model, about 17% of the total variation on unintended pregnancy was occurred at the community level and is attributable to the community-level factors. The highest MOR value (2.17) in the null model revealed there was a variation of unintended pregnancy between clusters. Furthermore, the highest (36.4%) PCV in the final model (Model 4) indicates that 36.4% of the variation in unintended pregnancy across communities was explained by both individual and community level factors. The model fitness was checked using deviance and the model with the lowest deviance (Model 4) was the best-fitted model.

Table 3 showed the fixed effects model. In the bivariable multilevel modeling, all of the independent (both individual level and community level variables), except knowledge of the ovulation cycle, had shown statistically significant association at a p-value of < 0.20. In the multivariable multilevel logistic regression analysis, the odds of unintended pregnancy was 43% [adjusted odds ratio (AOR) = 0.57; 95%CI: 0.41, 0.79] and 32% [AOR = 0.68; 95%CI: 0.47, 0.97] lower among mothers whose age was between 20 to 34 and 35 to 49 respectively as compared to those mothers whose age was 15 to 19 years. Mothers whose religion were Muslim had 27% [AOR = 0.73; 95%CI: 0.60, 0.88] lower odds of unintended pregnancy as compared to those mothers whose religion was Orthodox Christian. Regarding marital status, married mothers had 54% [AOR = 0.46; 95%CI: 0.37–0.58] lower odds of unintended pregnancy as compared to their counterparts. Respondents with a household size of four to six and seven and above had 1.38 [AOR = 1.38; 95%CI: 1.10, 1.69] and 1.54 [AOR = 1.54; 95%CI: 1.20, 1.99] times higher odds of unintended pregnancy respectively as compared with a household size of one to three. Multiparous and grand multiparous women had 1.36 [AOR = 1.36; 95%CI: 1.10, 1.69] and 1.92 [AOR = 1.92; 95%CI: 1.47, 2.52] times higher odds of unintended pregnancy respectively as compared to primiparous women. A woman who was living in large central regions and metropolitan regions had 2.56 [AOR = 2.56; 95%CI: 2.06, 3.17] and 1.91 [AOR = 1.91; 95%CI: 1.44, 2.53] times higher odds of unintended pregnancy respectively as compared to a woman from small peripheral regions.