Application of a modified Poisson model in identifying factors associated with prevalence of pregnancy termination among women aged 15 – 49 years in Uganda

Background Abortion in Uganda is illegal, only permitted when it places the pregnant mother at risk. This study aimed to apply the modified Poisson model in identifying factors associated with the prevalence of pregnancy termination among women of reproductive age in Uganda. Methods The 2016 Uganda Demographic Health Survey (UDHS) data were used in this study. More than 18,000 women of the age of 15 – 49 years participated in this study. A modified Poisson model that incorporated sampling weights was used to establish the factors associated with pregnancy termination. Results In Uganda, 18,506 (18.1%) had ever had a pregnancy terminated. The results revealed that, the woman's age [APR = 3.15, 95% CI: 2.72–3.63], being married [APR = 1.55, 95% CI: 1.40–1.71], mass media exposure [APR = 1.18, 95% CI: 1.08–1.29], working status [APR = 1.21, 95% CI: 1.09–1.35], and having visited a health facility [APR = 1.20, 95% CI: 1.10–1.31] were positively significantly associated with likelihood of pregnancy termination. Conclusion There exists a significant proportion of women who have had their pregnancies terminated in Uganda. It is observed that woman's age, marital status, mass media exposure, having visited a health facility in the last 12 months and working status were main predictors. Based on these results, researchers concluded that the emphasis should be put on improving access to post-abortion care, contraceptive use and media exposure.


Introduction
Pregnancy termination, otherwise known as induced abortion 1 or elective abortion and therapeutic abortion is removal of the fetus and placenta or pregnancy tissue from the uterus. In developing countries especially in Africa such as Uganda, induced abortion is unsafe and a major public health issue 2 . Unsafe abortion is defined as "a procedure for terminating an unwanted pregnancy by persons lacking necessary skills or in an environment lacking minimal standards or both" 3 Globally, about 20 million unsafe abortions occur each year with 95% of these in developing countries and about 67% of maternal deaths are related to these abortions making it the leading cause of maternal death 3,4 . Of the abortions that occur, 19 -20 million abortions are conducted by individuals without the necessary skills, and are thus associated with severe complications such as hemorrhage, sepsis and peritonitis, trauma to the cervix, vagina, uterus and abdominal organs and death [4][5][6][7][8] . According to 9 , unsafe abortion explains up to 20% of all deaths during pregnancy. The proportion of unsafe abortions are significantly high in highly restrictive abortion laws regions than with less restrictive law regions 8 such as Uganda.
Developing countries accounted for approximately 99% of global maternal deaths in 2015 with sub-Saharan Africa (SSA) accounting for 66% of these 10 . This high maternal mortality is attributed to poverty, lack of information about abortion and other maternal health services, distance to the health center, inadequate maternal health services, cultural practices, residence status where women in remote regions do not usually access quality health care. According to 3 , Africa is ranked second in terms of induced abortion rate with 29 occurring per 1000 women aged 15 -44 years. Almost all the induced abortions are unsafe, with 5.5 million unsafe abortions per year 2 . In East Africa, induced abortion rate is 38 per 1000 women per year 11 .
In Uganda, legal abortion is permitted only for the safety of the woman 7 . This restriction compels the perpetuation of the practice in secrecy and often in unsafe conditions 7,12,13 . The prevalence of pregnancy termination and factors associated with such prevalence in Uganda are not known. Literature highlights different factors associated with pregnancy termination. Such factors include socio-economic and demographic variables, sexual violence, and interruption of family planning, health seeking behavior, contraceptive use, marital status, and birth order 3,4,14,15 . In this study, the researchers aimed to apply the modified Poisson model in identifying factors associated with the prevalence of pregnancy termination among women of reproductive age in Uganda based on 2016 Demographic Health Survey (DHS) Data.

Methods
The study used secondary data collected by the Uganda Bureau of Statistics (UBOS) during the 2016 Uganda Demographic Health Survey (UDHS). The sampling frame for this survey was the same as that of the Uganda National Population and Housing Census (NPHC) conducted by UBOS in 2014. The participants in the 2016 UDHS were a random sample selected nationally from 20,880 households, from 15 sub-regions, stratified and selected in two stages. The first stage involved selecting 697 enumeration areas (EAs) of which 162 EAs were selected from urban areas and 535 from rural areas. From the 20,880 selected households, 18,506 women were interviewed. All women aged 15 -49 years in the selected households were eligible to be part of the study 16 . An approval letter to use DHS data was obtained from the DHS program.
Statistical analysis focused on women of reproductive age 15 -49 years. The outcome variable was termination of pregnancy, a binary categorical variable, that is, ever had pregnancy terminated or not. The exposure variables considered for this study included age of the mother, marital status, mother's level of education, child's birth order, contraceptive use, frequency of listening to radio, smoking status, and distance to the health facility, among others. The analysis followed three levels starting with descriptive exploration of the study variables, both the response variable and the predictors. Bivariate analysis level was conducted by performing bivariate modified Poisson regression analysis. Only variables that were associated with pregnancy termination at p-value of 0.20 at bivariate analysis were considered in multi-variable model. Backward elimination technique was applied to remove predictors that were not significant at multi-variable level, one at a time and only variables that were significant were considered as main predictors. Researchers tested for confounding of predictors on age of the woman. Prevalence ratios and their corresponding 95% confidence intervals are presented. The statistical models for binary outcomes suggested in literature include binary logistic regression model, modified Poisson regression model and log-binomial regression model 17-19 each used based on some definitive features of the outcome variable. The commonest of these regression models is binary logistic model (20) given by are predictor variables and = 0 is the probability of experiencing the outcome. Odds ratio is exp(β). The link function for logistic model is the logit link function. The major weakness of logistic regression is that it tends to over-estimate the risk if the outcome of interest is not rare [20][21][22] .
However, modified Poisson regression model is preferred in cross-sectional studies when the outcome of interest is not rare because it approximates the risk ratios or relative ratios better than binary logistic regression model 17,20,23 . The link function of modified Poisson regression model is log link. The modified Poisson regression model is the Poisson regression of binomial data using robust error variance (18). Its functional form is in equation (ii) below.
where π_i is the probability of experiencing the outcome of interest for subject i, β's is the mean of the ith subject and approximates relative ratios as exp(β).

Results
A total of 18,506 women participated in this study. Results in Table 1   The likelihood of pregnancy termination among women who had secondary or higher level of education was 17% lower in comparison to those who had no education [APR = 0.83, 95% CI: 0.73, 0.94], and among women who had access to media was 18% higher compared to those who were not exposed to mass media [APR = 1.18, 95% CI: 1.09, 1.27]. For women who were not cigarette and tobacco smokers, the likelihood of pregnancy termination was 24% less compared to those who were smokers [APR = 0.76, 95% CI: 0.63, 0.92]. Perceived distance to a health center was negatively associated with the likelihood of pregnancy termination. Women who perceived this distance as not being a big problem, the likelihood of pregnancy termination was 11% less compared to those who perceived the distance as being a big problem [APR = 0.89, 95% CI: 0.83, 0.96]. In addition, the likelihood of pregnancy termination among working women and those who visited the health facility in the last 12 months was 21% [APR = 1.21, 95% CI: 1.09, 1.35] and 20% [APR = 1.20, 95% CI: 1.10, 1.31] respectively, compared to women who are not working and had not visited the health facility in the last 12 months. Geographical region was found to be negatively associated with pregnancy termination. The likelihood of pregnancy termination was [APR = 0.83, 95% CI: 075, 0.91] among those women from Eastern Uganda, [APR = 0.80, 95% CI: 073, 0.89] among women from Northern Uganda, and [APR = 0.62, 95% CI: 0.68] among women from Western Uganda compared to women from central Uganda. Total number of children ever born confounded the relationship between pregnancy termination and age of the mother which was considered a primary exposure in this study.

Discussion
In this study, it was revealed that the overall prevalence of pregnancy termination was 18.1%. This finding is congruent with the findings in Batu Town, Ethiopia 24 , in Kampala Uganda 25 , in Nepal 26 and in Ghana 27 who all found a slightly high prevalence level. However, the findings in Ethiopia 4 , in Hamedan, Iran 28 and in Mozambique 28 showed slightly lower prevalence (less than 10%). High prevalence in Uganda can be attributed to low contraceptive use and high rates of unintended pregnancies 7,15 .
Significant association was observed between age of the mother and pregnancy termination. It was found out that women aged 35 -49 years were highly likely to terminate their pregnancies than those aged 15 -24 years, followed by those aged 25 -34 years, compared to those aged 15 -24 years. Similar findings are observed in 4,27-29 . However, they are in disagreement with the findings of 30 and 24 . The study also found that marital status, which is, being married (including living with a partner) was statistically associated with pregnancy termination. Studies by 4,30,29 , and 27 agree with this finding. However, the findings are not congruent with the findings of 24 . This is attributed to low levels of modern contraceptive use among mar-rieds15 and culture that encourages women to give birth when married or living with a partner.
In the present study, exposure to mass media was found to be significantly associated with pregnancy termination. The study revealed that the odds were higher among those who listen to radio at least once a week. This concurs with finding of 27 in Ghana and 4 in Ethiopia.. This might be because women who have access to media get information about where to get maternal health services. The study also revealed that women who do not smoke have lower odds of pregnancy termination compared to those who smoke. This could be due to the fact that, those who do not smoke likely have healthier lifestyle than those who do. These health lifestyles might also include better health seeking behavior. Regarding distance to the health facility, whether it's a big problem or not while seeking medical help, the odds of pregnancy termination were lower among women who find the distance not a big problem than those who find it a problem. Contrary to this finding, 4 observed that distance to the health centre being a big problem was insignificantly associated with pregnancy termination. In this study, this could be due to the fact that, when distance to the health facility is not a big problem, health seeking behaviors in terms of family planning might be better, reducing the odds of pregnancy than those who find a big problem.

Conclusions and Limitations
The prevalence of pregnancy termination in Uganda is 181 per 1000 women. Mother's age, mother level of education, marital status, mass media exposure, distance to the health facility, smoking status, woman's working status, visiting of health facility and region of residence were found to be significantly associated with pregnancy termination. Researchers recommend that emphasis should be put on improving access to post abortion care since there is a strong evidence of pregnancy termination, family planning services, and use of media to sensitize and inform women. However, there are questions of interest that were not asked in this study. Whereas it was found that 181 in every 1000 women had ever had pregnancy terminated, the conditions upon which these pregnancies were terminated are not covered in this study. Whether they were unsafe or safe and the consequences from terminated pregnancies-if unsafe are also not known and cannot be presented in this study.