Individual and community level factors associated with unintended pregnancy among pregnant women in Ethiopia

Unintended pregnancy is among the major challenges of public health and a major reproductive health issue, due to its implications on the health, economic and social life of a woman and her family mainly in low and middle-income countries, particularly sub-Saharan Africa. The study aimed to assess unintended pregnancy and associated factors among pregnant women in Ethiopia using multilevel analysis from the EDHS 2016. We used the data from the 2016 Ethiopian Demographic and Health Survey, comprised of 1122 pregnant women. The prevalence of unintended pregnancy was determined through descriptive statistics and multilevel logistic regression was performed to identify factors associated with unintended pregnancy. Variables with a p-value < 0.05 in the selected model were considered as significantly associated and an adjusted odds ratio was used to determine the strength and direction of the association. The prevalence of unintended pregnancy was 29.7% (CI 27.0%, 32.4%), of which 20.4% were mistimed and 9.3% unwanted. Being multi-para and fertility preference to have no more child were associated with a higher risk of unintended pregnancy whereas husbands' polygamy relation, having no women autonomy, and living in Afar and Somali regions showed a less likely risk of experiencing an unintended pregnancy. This study showed that the proportion of women who experienced unintended pregnancy is considerably high. Parity, fertility preference, polygamy relation, women autonomy, and region were identified factors associated with unintended pregnancy. Therefore, policymakers at all levels, reproductive health experts, and concerned organizations should emphasize minimizing unintended pregnancy targeting the regional variation at large. Researchers have to explore the regional variations through a qualitative study.

www.nature.com/scientificreports/ Various global and local studies identified that maternal age, marital status, education, occupation, partnerrelated characteristics (age, education, and occupation), residence, number of children, family size, income, wealth status, and obstetric characteristics were found to be associated factors of unintended pregnancy 8,[13][14][15][16][17][18][19][20][21][22][23][24] . In Ethiopia, several studies assessed the prevalence of unintended pregnancy and associated factors in different perspectives and levels like in the national level or local levels but none of them tried to address the determinants of unintended pregnancy in the individuals and community level using multilevel analysis. It is essential to identify factors associated with unintended to design targeted intervention. Hence, this study aimed to identify individual and community-level factors associated with unintended pregnancy among pregnant women in Ethiopia using data from the 2016 Ethiopian Demographic and Health Survey (EDHS).

Methods and data
Data source and population. We used data from the nationally representative (9 regions and 2 city administrations) cross-sectional study EDHS 2016 25 . The sampling procedure of the EDHS consists of a twostage stratified sampling procedure. In the first instance, the regions were stratified in urban and rural areas, and each stratum was clustered into enumeration areas based on the 2007 Ethiopian housing and population census 26 . The first sampling involved the selection of clusters through probability proportional size allocation. Finally, households were selected from each cluster using the systematic probability sampling method.
For the current analysis, 1122 pregnant women were selected from the total of 15,683 reproductive age women interviewed in the 2016 EDHS (Fig. 1).

Study variables and definitions.
The outcome variable of this study is unintended pregnancy which represents mistimed (the pregnancy that occurs earlier than desired time) or unwanted (the pregnancy that occurs when no more children are desired) types of pregnancy 1 . The EDHS assessed it through a question inquiring from women whether the most recent pregnancy is wanted or not and those who didn't want the pregnancy were further requested whether they wanted later or not wanted 25 .
Community-level factors. Residence, region, community women education, community poverty status, community media exposure, community women autonomy, and community modern contraceptive methods awareness 32,33 . Residence and region were taken directly from EDHS, however, other community-level variables namely community women education, community poverty status, community media exposure, community women autonomy, and community modern contraceptive methods awareness were created from individuallevel factors by aggregating them using the median due to the asymmetric distribution of the data. In aggregating the community level variables, initially respective individual-level variables were dichotomized into yes or no by referring to previous studies and those clusters with a median of 0.5 and 1.0 were considered as clusters having high proportion, and those with the median value of 0.0 were taken as having low proportion 34 .
Community-women education. The proportion of women who completed primary and above educational level in the clusters 35 .
Community poverty status. The proportion of women who have poorer and poor wealth status in the clusters 35 .
Community media exposure. The proportion of women who had exposure at least for television or radio or newspapers in the clusters 35,36 . www.nature.com/scientificreports/ Community women autonomy. The proportion of women who make decisions alone or jointly on the woman's own health care, major household purchases, and visits to the woman's family or relatives in the clusters 37 .
Community awareness on modern contraceptive methods. Refers to the proportion of women in the clusters who have ever heard about modern contraceptive methods.
Data processing and analyses. The data were accessed in SPSS format from the Demographic Health Survey Archive, and managed and analyzed using R version 3.6.1 (variable recoding and the descriptive statistics) and STATA version 14 software. Due to variation in sample representation, sample weight was considered and the weight variable was created by dividing the individual women sample weight variable by 1,000,000 to make the data representative enough. Then all the descriptive and analytical analyses took this weight into account. Descriptive statistics were computed to characterize the study population in terms of relevant socio-demographic and obstetric characteristics both at the individual and cluster levels.
We chose relevant factors associated with unintended pregnancy through reviewing previous studies and discussion with experts in the subject matter. We conducted a bivariate analysis using cross-tabulation, to select variables eligible for multivariable analysis. Due to the nature of the EDHS data, being hierarchical, we used multilevel logistic regression to identify variables independently associated with unintended pregnancy using a series of four models.
Model I (empty model). In this model unintended pregnancy was analyzed by the cluster variable to test the random effect of between-cluster variability. The intra-class correlation (ICC) was estimated to determine the effect variability justified by the cluster.

Model II. This model was run between the dependent variable and individual-level independent variables.
Model III. This model was used to examine the association of community or cluster level variables with unintended pregnancy.
Model IV (combined model). Finally, both individual and community-level variables were run together to examine the combined effect on unintended pregnancy.
The intra-class correlation was calculated for each of the models and the Proportional Change in Variance (PCV) was computed for Model II, III, and IV with respect to the variance in the empty model to show the power of the factors in the model to explain unintended pregnancy by using the formula PCV = (Ve − Vmi)/Ve where Ve is variance obtained in the empty model and Vmi is variance in successive models. PCV was calculated by subtracting the variance of each model from the null model. AIC and BIC were also computed for each of the models and the model with the lowest AIC value (the combined model) was selected to identify individual and community-level factors associated with unintended pregnancy 38 . The fixed-effect sizes of individual and community-level factors on unintended pregnancy were expressed using the Odds Ratio (OR) and the community effect sizes were estimated using the 95% Confidence Interval (95% CI).
Ethics approval. The data were accessed from Demographic Health Survey Archive through a formal request and permission to analyze the data and dissemination of the results has been obtained.

Results
Individual-level characteristics. The median age of participants was 26.0 with a range of 15-48 years.
Nearly half (49%) of the respondents were under the age group 25-34 and almost all (97.1%) were married. Slightly higher than half (53.3%) of them and 43.9% of the respondents' husbands had no formal schooling. Nearly one-fifth (20.2%) were primigravida and 9.7% had terminated pregnancy at least once. One-tenth (9.8%) were in a polygyny relationship. (Table 1).

Community-level variables.
Above two-fifths, (41.9%) of the respondents were from Oromia region and 85.8% were rural residents. Nearly half of the respondents were from clusters with higher women's education (52.7%) and high poverty (53%) ( Table 2).
Factors associated with unintended pregnancy. In the multilevel logistic regression four models had been fitted. The random effect analysis in the null model was used to examine the cluster effect on unintended pregnancy. The result depicted a significant variability in unintended pregnancy across the clusters (ICC 0.193) which further indicates that the cluster accounted for a 19.3% variance in unintended pregnancy. The ICC values for models II, III, and IV respectively are 0.114, 0.005, and 0.043.
The fixed effect analysis indicated that parity, fertility preference, polygamy relation, and women autonomy from individual-level factors, and region from community-level factors were found to be significantly associated with unintended pregnancy ( www.nature.com/scientificreports/
In the multilevel analysis; parity, fertility preference, polygamy relation, and women autonomy and region showed a significant association with unintended pregnancy. Multiparous and grand multiparous women had a more likely chance of experiencing as compared to nulliparous women. Similar findings were obtained by studies conducted in Ethiopia (the National Survey, Addis Zemen hospital, Maichew Town, Gelemso General Hospital, Duguna Fango District, and Arsi Negele Woreda) 8,13,[17][18][19]24 , Senegal 23 , in rural Ghana 14 , in the Amazon basin of Ecuador 44 and by A Narrative Review of studies in Developing Countries 16 . The increased likelihood of experiencing unintended pregnancy with increased parity might have resulted from reporting differences that mean primiparous women might accept the pregnancy and report it as wanted whereas multiparous women report it properly.
Women who have no more desire to have children had a more likely chance of experiencing unintended pregnancy. A research finding which was also conducted in Jimma town, southwest Ethiopia 27 also showed a similar association stating that women who desire less number of children were more likely to have unintended pregnancy. This might be due to the difference in reporting, women who have a desire to have more children may accept the pregnancy and later on consider it as wanted and report it as an intended pregnancy. A significant association between husband fertility desire and unintended pregnancy has been reported by several studies in Ethiopia 19,21,22 .
Women whose husbands were in a polygamy relationship were less likely to experience unintended pregnancy. No previous study established a relation between polygamy relationship and unintended pregnancy. Thus, this finding will help future researchers to determine the relationship through the application of advanced study designs that indicate cause and effect relationships.
Regarding women's autonomy, women who were autonomous in household decision-making had less likelihood of having unintended pregnancies as compared to women who were not autonomous. Similar findings were obtained by numerous studies conducted in Ethiopia 16,20,24 and Senegal 23 . Women's autonomy in household decision-making has a positive impact on maternal healthcare service utilization 15,[45][46][47][48] . Unintended pregnancy can be prevented through the utilization of the right contraceptive method at the right time, which is among the core maternal health care services.
This study also identified that region of residence has a significant association with unintended pregnancy. Women living in Afar and Benshangul Gumuz regions tend to have less likelihood of having unintended pregnancies as compared to women living in the Tigray region. This might be attributed to fertility desire difference because Afar and Benshangul Gumuz are underdeveloped regions whereas Tigray is among developed regions in the Ethiopia context. A similar geographic difference in the prevalence of unintended pregnancy was reported by a multilevel analysis of demographic health survey conducted in the amazon basin of Ecuador 44 . Table 3. Factors associated with unintended pregnancy were identified through multilevel logistic regression. Signif. codes: '***' < 0.001 '**' < 0.01 '*' < 0.05. www.nature.com/scientificreports/ Strengths and limitations. As a strength, the study is representative at the national level and it is the analysis of the data from pregnant mothers which helps to reduce recall bias. However, the design of the study is limited to establish cause and effect relationship.

Conclusion and recommendation
This study revealed that a significant proportion of pregnant women experience unintended pregnancy. Women who were multi-para and multipara, fertility preference to have no more child had a higher risk whereas women whose husbands had polygamy relation, have no autonomy, and Afar and Somali region residents had a less likely risk of experiencing unintended pregnancy. Designers and implementers of projects targeting to avert unintended pregnancy have to consider the regional variations. Various stakeholders have to work in promoting women's autonomy in the household decision-making role. The authors would like also to recommend for researchers investigate the impact of polygyny relationships on unintended pregnancy using various designs. Researchers should explore the reasons for regional variation in terms of unintended pregnancy prevalence.

Data availability
The study used the 2016 Ethiopian Demographic and Health Survey. The data is available and can be obtained from the DHS (https:// dhspr ogram. com/ Data/) upon request. www.nature.com/scientificreports/