Background Most existing studies on unintended pregnancies tend to examine the influence of individual socio-demographic and health characteristics without sufficient attention to community characteristics. This study examines community characteristics influencing unintended pregnancies in Kenya. Methods Data were extracted from the 2014 Kenya Demographic and Health Survey (KDHS). The outcome variable was unintended pregnancy. The explanatory variables were selected individual and community level variables. The Multilevel mixed-effects logistic regression was applied. Results Findings show 41.9% prevalence of unintended pregnancies. Community characteristics such as community education, community timing for initiation of childbearing, community fertility norms, and community media exposure significantly influence the likelihood of unintended pregnancies. The Intra-Cluster Correlation (ICC) provided evidence that community characteristics had effects on unintended pregnancies. Conclusion There is evidence that community characteristics influence the prevalence of unintended pregnancies in Kenya. Community sensitisation and mobilisation should be central to all efforts aiming to reduce prevalence of unintended pregnancies.
The Republic of Kenya is the third most populous country in Eastern Africa with an estimated 2015 population of 44.3 million persons33. The country’s population grows mainly through natural increase. The current contraceptive prevalence rate for modern methods in Kenya is 53% which is above the Eastern Africa’s average of 35%33. Based on current Human Development Index, Kenya is rated as a low human development country with moderate inequality in gender, education and income34. Since 1967, the government of Kenya has implemented the national family planning programme as an integral part of the overall national development strategies, making Kenya one of the sub-Saharan African countries with the earliest national family planning programme35. With over fifty years of implementation, the family planning programme in Kenya has resulted in fertility decline and reduction in adverse reproductive health outcomes such as childhood mortality rate, maternal mortality, and HIV prevalence in the country36. Although, the first National Reproductive Health Policy in Kenya was adopted only in 200737, the country has appropriate policies and programmes to promote the sexual and reproductive health of the population. These include the 2015 National Adolescent Sexual and Reproductive Health Policy, Kenya Health Policy (2012- 2030), 2014 National Gender-Based Violence Policy, and the 2012 Population Policy for National Development38. However, in spite of the numerouspopulation and health policies, the sexual and reproductive health of men and women in Kenya remains poor with high prevalence of unintended pregnancies and clandestine abortion30. It was noted in the policy that lack of political commitment, poverty and a number of socio-cultural factors such as lack of female autonomy in decision-making, myths and misconceptions such as associating contraceptives with being promiscuous or health challenges, and negative attitudes hindered the progress of previous population and health programmes in the country35. Current efforts to improve sexual and reproductive health, particularly contraceptive prevalence in the country, include expansion of family planning service delivery points including community-based distribution, promotion of male involvement, integration of family planning with HIV/AIDS and other reproductive health services35. Data analysed in the study were extracted from the women’s data of the 2014 Kenya Demographic and Health Survey (KDHS). The 2014 KDHS was implemented by national agencies led by the Kenya National Bureau of Statistics. The sample included in the survey was drawn from 39,679 households in 1,612 clusters (primary sampling unit) with 995 clusters in rural areas and 617 urban areas. A two-staged sampling technique was adopted for the survey. Detail design of the 2014 KDHS has been published36. The study did not analyse information from all the women covered in the survey. Analysis was restricted to women who were currently pregnant or whose last child was born within the last five years preceding the survey. The weighted sample size was 6,871 women. Data were requested from MEASURE DHS through online data access facility. Authorisation to analyse the dataset was granted by the organisation. The outcome variable was pregnancy intention categorised into intended and unintended. This was generated from information on the last child born during the last five years and from current pregnancy. The pregnancy intention of women who reported that they wanted their last child or current pregnancy were categorised as ‘intended’ while the pregnancy intention of those who reported that they wanted the child or pregnancy later and those who do not want the child or current pregnancy were categorised as ‘unintended’. The explanatory variables were five individual-level variables and five community-level variables. Five individual-level variables were analysed, namely: individual education, age at first marriage, current marital status, parity and employment status. These variables were selected on the basis of their significance in previous studies34,36,22. Age at first marriage was categorised into four groups, namely,: 14 years or less, 15–19 years, 20–24 years and 25 years or older. Current marital status was grouped into two, namely: not currently married and currently married. Parity was measured by number of children ever born and was divided into three groups, namely: low parity (two or fewer children ever born), multiparity (three to four children ever born), and grand multiparity (five or more children ever born). Employment status was categorised as employed or unemployed. The five community-level variables were community media exposure, community fertility norm, community education, community type of residence, and community level of teenage motherhood. Two of these variables (education and type of residence) have been analysed in an earlier study26. Community media exposure was generated from the combined frequencies of reading newspaper, listening to radio, and watching television within a week. Community fertility norm was generated from combination of individual ideal family size, while community initiation of childbearing was generated from combination of individual age at first birth. Community education was derived from combination of individual woman education. To derive community variables, the method of aggregation was adopted by first setting a benchmark to indicate proportion of women in the community having the attribute or characteristic of interest and then aggregating the variable at the cluster (community) level. The proportions were then ranked and divided into three groups (tertiles). For instance, to derive community level of teenage motherhood, age at first birth was benchmarked at 18 years. All women who became mothers below age 18 years were combined and sorted by cluster to show the proportion of women who became mothers early. This proportion was then divided into three to indicate low, medium, or high proportions of teenage motherhood in the community. Previous studies that have explored community contexts using the DHS data have used similar method to derive community variables41–42. Four household-level variables were selected for statistical control. These are women’s autonomy, household wealth, spousal violence, and type of marriage. All the variables have shown significant influence on unintended pregnancies in previous studies40, 8. Women’s autonomy was based on responses to questions on participation in three household decisions, namely: decision on own health, purchase of large household items and visits to friends and relatives. Women who took all the decisions solely were defined as having ‘high’ autonomy, those who took part in the decision jointly with the male partner were defined as having ‘partial’ autonomy, and other women such as those who did not take part at all in the decisions were defined as having ‘no’ autonomy. Women were grouped as ‘polygamous’ or ‘monogamous’ if the partner had at least one other wife or not. The three types of spousal violence measured in the DHS (physical, sexual and emotional) were combined to form a single variable showing whether women had ever or never experienced at least one type of spousal violence. All statistical analyses were performed using Stata 12. Frequency distributions were used to describe sample characteristics and prevalence of unintended pregnancies. Simple cross tabulation of explanatory variables and pregnancy intention, and unadjusted binary logit coefficient with 95% confidence interval were used to examine the relationship between the study variables. Positive coefficient indicates positive relationship and negative coefficient indicate otherwise. The multilevel mixed-effects logistic regression analysis was applied. This method was selected for the study because of the hierarchical nature of the data43. Model Specification: Where: πij is the log of odds of unintended pregnancy (1 − πij) is the log of intended pregnancy β0 is log of the intercept β1…βn are the regression coefficients X1ij…Xnij are the individual and community characteristics included in the model u0j are the random errors at cluster level eij is the error term. The Stata xtmelogit command was used to estimate the model parameters44. Five different models were fitted. Model 0 is the empty model which did not include any explanatory variable. Model 1 included only individual-level variables while Model 2 was based on the community-level variables. In Model 3, the individual and community variables were combined. The full model (Model 4) included all explanatory and control variables. The odds ratios with 95% confidence intervals were used to measure the fixed-effects (measure of association between the explanatory and outcome variables). The Intra-Class Correlation (ICC) was used to measure the random effects (measure of variation). The ICC was calculated as: Log=[πij1−πij]=β0+β1X1ij+…βnXnij+u0j+eij where σu2 is the variance at the community level and π2/3 is equal to 3.2943. The ICC expressed in percentage showed the variation in unintended pregnancies due to community characteristics. Model adequacy was examined by the Log-likelihood Ratio test (LR test). The LR test compares the fitted model with one-level ordinary linear regression. The result will indicate if the fitted model is adequate for the data being analysed. The Variance Inflation Factor (VIF) with mean VIF of 3.68 confirmed the absence of significant multicollinearity among the explanatory variables. The statistical significance for all tests was set at p<0.05.
N/A