3.1 Data set

The data in this study come from the AZDHS. The AZDHS is a part of the long-standing worldwide demographic and health surveys program, which is developed to assist developing countries in collecting data on fertility, reproductive health, maternal and child health, nutrition and HIV/AIDS. The AZDHS is conducted by the State Statistical Committee of Azerbaijan, the national statistical authority, with the technical assistance of Macro International Inc., USA (SSC and Macro International Inc., 2008). Funding was provided by the USAID (United States Agency for International Development) and the UNICEF (United Nations Children’s Fund). The AZDHS is a nationally representative survey of 8444 women aged 15–49 years. It was implemented from July to November 2006 with a household gross response rate of more than 90%.

The AZDHS employs a multistage stratified clustered sampling design based on a sampling frame that is from the Population Census of the Azerbaijan Republic conducted in 1999, as provided by the State Statistical Committee of Azerbaijan. In this sampling frame, stratification was achieved in two stages. In the first stage, the country was stratified by economic regions and in the second stage the regions were divided into rural and urban strata. In this way, the 10 regions of Azerbaijan were stratified into 20 sampling strata and the capital city of Baku, which has only urban areas (see Figure 1).

Figure 1

Azerbaijan and its socio-economic regions. Notes: 1 = Aran, 2 = Yuxarı Qarabağ, 3 = Lankaran-Astara, 4 = Ganca-Qazax, 5 = Şəki-Zaqatala, 6 = Absheron, 7 = Dağlıq Şirvan, 8 = Quba-Xaçmaz, 9 = Kalbacar-Laçın, 10 = the Autonomous Republic of Naxçıvan.

Although officially there are 10 economic regions in Azerbaijan, 2 regions were not covered by the AZDHS. First, the Kalbacar-Laçın region is currently occupied by Armenian forces and was excluded from the sampling frame as well as the survey, with a loss of approximately 5% of the Azerbaijan total population (SSC and Macro International Inc, 2008). Second, the Autonomous Republic of Naxçıvan did not participate in the survey due to the blockade with a loss of approximately 4.5% of the Azerbaijan total population. This leaves the AZHDS with 17 strata including 16 rural/urban strata for the remaining eight economic regions of Azerbaijan and one urban stratum for the capital Baku.

For clustering purposes, samples were selected independently in each of 17 strata using a two-stage procedure. In the first stage, 350 census EUs were selected with a probability proportional to the EU size, where the EU size denotes the number of households in the EU. In each selected EU, a household listing operation was conducted to develop a full list of all households living in the EU. The developed lists of households served as the sampling frame for the second-stage selection of households.

Some of the selected EUs were large in size and to minimize the task of household listing, EUs with more than 300 households were further segmented. Only one segment in each of these large EUs was selected for the survey with a probability proportional to the segment size. The household listing for that EU was conducted only in the selected segment. Consequently, all clusters in the survey are either an EU or a segment of a large EU. In the second stage of selection procedure, 24 households were selected in each cluster, using equal probability systematic sampling.

As a result, a total of 318 clusters (178 urban and 140 rural) in Baku and eight economic regions were selected for the AZDHS implementation, resulting in a final sample of 7619 households. After the spreadsheet indicating the selected households for each cluster was developed, only the preselected households were interviewed. No replacements or changes to the preselected households list were allowed in order to avoid possible selection bias.

The AZDHS collects a wide range of detailed information on demographics, education, household wealth, maternal and child health and antenatal healthcare utilization. In particular, the pregnancy and postnatal section of the survey contains comprehensive information on the antenatal healthcare utilization of women who gave birth in 2001 or later. The questions about antenatal healthcare utilization concern only the most recent births delivered during this period. In addition to asking women whether they see anyone for antenatal care for pregnancy during the last birth, the AZDHS records information about the frequency and timeliness of antenatal care utilization as well as about the quality of utilized services.

3.2 Research questions and selection of dependent variables

The discussion in the previous sections of this paper has motivated us to examine the socio-economic determinants of antenatal healthcare utilization in Azerbaijan. To shed light on antenatal healthcare utilization in the country, the following four research questions are investigated in this study:

1. 1. What are the determinants of antenatal healthcare utilization?

2. 2. What are the determinants of frequency of the utilization?

3. 3. What are the determinants of timing of the first utilization?

4. 4. What are the determinants of the quality of the utilized care?

Inasmuch as this study seeks to answer the four specific research questions posted above, we studied four distinctive dependent variables. To answer the first posted question, we examined the answers to the following question asked by the AZDHS: “Did you see anyone for antenatal care for this pregnancy?” The respondent had the opportunity to answer by selecting between ‘someone’ and ‘no one.’ The definition of ‘someone’ in this question encompasses a wide range of possible antenatal care providers such as: doctor, nurse/midwife, feldsher, community/village health worker and others. Therefore, this question clearly identifies those women who did not receive any antenatal health care during their last pregnancy.

For the women who did receive care, the AZDHS asked the follow-up question: “How many times did you receive antenatal care for this pregnancy?” The responses to this question varied from 1 visit to 20 visits. These responses serve as a measure of the frequency of antenatal care utilization. Another follow-up question asked by the AZDHS was: “How many months pregnant were you when you first received antenatal care for this pregnancy?” Since it is common to measure the timing of the pregnancy by trimesters, the responses are grouped into three categories: visits during the first trimester, visits during the second trimester and visits during the third trimester. Consequently, the responses to this question varied from one to three. These responses serve as a measure of timing of antenatal care utilization.

Lastly, the series of follow-up questions were aimed at assessing the quality of delivered antenatal care. Specifically, the AZDHS asked whether each of the following was carried out at least once during antenatal care during the last pregnancy: (1) the woman was weighed; (2) her blood pressure was measured; (3) her blood sample was taken; (4) her urine sample was taken; (5) she was told about the signs of pregnancy complications; (6) she was told where to go if she had any of the pregnancy complications. The response for each question was coded ‘1’ if the specific service was provided and ‘0’ if otherwise. These responses were employed to compute the additive index for the quality of services (Kohler and Kreuter, Reference Kohler and Kreuter2008). The index varied from 0 to 6. If women received no service, then the index took a value of 0. If women received only one type of service, for instance, their blood pressure was measured, then the index took the value of one, while the index took value of six if women received all the aforementioned services.

3.3 The conceptual framework and selection of antenatal healthcare utilization determinants

The rapid increase in scholarship on determinants of antenatal healthcare utilization in developing and transitional countries prompts us to base the conceptual framework of this paper on the two most recent systematic literature reviews on the topic. These reviews were conducted by Simkhada et al. (Reference Simkhada, van Teijingen, Porter and Simkhada2008) and Say and Raine (Reference Say and Raine2007). The former examined 28 peer-reviewed studies published since 1990, while the latter examined 30 studies for approximately the same period. Both reviews converged in the main findings. The major determinants of antenatal healthcare utilization in developing countries are: (1) women’s socio-demographic characteristics (e.g. current age, educational attainment and age at first birth); (2) current pregnancy characteristics (e.g. birth order, gender of child, multiple births and wanting the last child or the current pregnancy); (3) affordability (e.g. socio-economic status and insurance coverage); and (4) accessibility of services (e.g. place of residence such as rural/urban and region, travel time to health facilities and state/private facilities). Several other factors, such as ethnicity, religion and culture, were also found to be important determinants of antenatal care utilization.

As suggested by the conceptual framework, women with higher education have a higher likelihood of utilizing antenatal care since they may have better knowledge of the benefits associated with antenatal healthcare utilization. Women with higher educational attainment may also be better informed about how the healthcare system functions and hence can get better access to it. In contrast, older women, women who experienced early pregnancies in their life, and women who have a higher birth order have a lower likelihood of antenatal care utilization due to the experience and confidence gained from previous pregnancies. Therefore, several independent variables were selected in this study to test the relationship between the abovementioned characteristics and antenatal care utilization. Women’s educational attainment is denoted by a continuous variable indicating the years of schooling. Likewise, women’s current age and age at first birth are also denoted by continuous variables.

Among the current pregnancy characteristics, the literature suggests that having been pregnant with a male child or multiple births, as well as wanting the last child or current pregnancy, may strongly and positively be associated with the probability of antenatal healthcare utilization. Hence, our study includes dummies for being pregnant with a male child or multiple births as well as a dummy indicating the desire for the last child or the current pregnancy to test these relationships.

The literature indicates that the ability of women to participate in decision-making about their own healthcare utilization is positively associated with the likelihood of antenatal care utilization. To capture this effect, we include a dummy variable indicating if women alone or women together with their husband or partner have the final say on their own health care.

To capture the effect of affordability, we use the wealth index quintiles provided in the AZDHS data set.Footnote 2 There are six wealth quintiles in the data set ranging from the first, the poorest, to the sixth, the richest. We select for use the bottom two quintiles, which represent the poorest and the poorer households, and the top two quintiles, which represent the richest and the richer households. The middle, the third quintile, is omitted as a benchmark category for the purpose of regression estimation.Footnote 3 Given the inequalities in healthcare utilization discussed in the background section, we expect that living in the poorest and poorer households will have a negative effect on the likelihood of antenatal care, while living in the richest and the richer households will have a positive effect.

Two variables were employed to capture the accessibility perspective. First, we use a categorical variable, rural, that denotes women living in rural areas, while the urban category is omitted as a benchmark for the purpose of regression estimation. We hypothesize that living in rural areas will have a negative effect on the likelihood of antenatal healthcare utilization as compared with living in urban areas. Second, we use dummies for the eight regions of the country, whereas the capital Baku is omitted as a benchmark for the purpose of regression estimation (see Figure 1 for details). Among the regions, Aran is the centrally located part of Azerbaijan. It is mostly an agricultural region with mainly industrial conglomerates in the southwest and northwest. Yuxarı Qarabağ is mainly an agricultural region that borders Aran on the east, Iran on the south and the occupied territories on the west. Aran and Yuxarı Qarabağ are the two regions that suffered the most from the war with Armenia. On the south of Aran, there is the Lankaran-Astara region surrounded by Iran and the Caspian Sea. This region is specialized in the agriculture, fishing and forestry sectors and in the processing of products from these sectors. On the west of Aran, there is the Ganca-Qazax region that borders Georgia and Armenia. The region is the second largest economically developed part of the country after Baku. On the northwest of Aran, there is the Şəki-Zaqatala region that borders Russia and Georgia. The key sectors of the economy are agriculture, the ‘light’ industry and food processing sectors. On the northeast of Aran, there is the Absheron economic region, which, together with Baku, is the largest contributor to the country’s domestic production. It is the second most densely populated region of Azerbaijan that is situated very favorably on the territory surrounding Baku, the most developed part of the country. Another region, Dağlıq Şirvan, is surrounded by Absheron, Aran and Şəki-Zaqatala. It is the least densely populated region in the country, which specializes in viticulture, winemaking, cattle-breeding and traditional handcrafting. Finally, the Quba-Xaçmaz region is located in the northeast of the country and borders Russia in the north and the Caspian Sea in the east. The major sectors of the region’s economy are agriculture, manufacturing and tourism. Although there are some differences between the regions in terms of socio-economic development, these differences are not significant as compared with differences between the regions and Baku (Government of Azerbaijan, 2003, 2004; World Bank, 2003). Therefore, we expect that living in the regions will have a negative impact on prenatal care utilization as compared with Baku. Variation between the regions is also expected.

It must be highlighted that at the preliminary stage of analysis, we investigated a large number of independent variables widely known to affect antenatal healthcare utilization, as suggested by the conceptual framework. However, some of these variables, for instance, perception of service quality or travel time to health facilities, were not available in the AZDHS. Other variables turned out to have a low level of variation in distribution, for example, religion (99% are Muslim) and ethnicity (95% are Azeri). However, a number of other variables were not relevant for the current conditions of the country. Thus, the share of maternity care received in private facilities is <1%, while the share of women and men covered by health insurance is about 1% and 5%, respectively, since the system of health insurance and private healthcare facilities is not well developed in Azerbaijan.

In addition, some other variables are found to be highly multi-collinear with each other. We identified the severe multi-collinearity between women’s age at the last birth, birth interval, the current age of the women, birth order, women’s education, wealth status of the household and residency. We found that a younger age for women at their last birth and shorter birth intervals were strongly multi-collinear with a younger current age for women, a larger birth order, lower educational attainments and wealth status as well as with living in a rural area. The effect of multi-collinearity was so strong that it changed the sign of regression coefficients to the opposite direction. These results are not surprising since severe multi-collinearity between these variables was also reported in the previous studies of the determinants of antenatal health care (Habibov and Fan, Reference Habibov and Fan2008). Under these circumstances, we followed the approach suggested by Fan and Habibov (Reference Fan and Habibov2008). We ran a separate regression between each predictor and outcome variable and recorded the sign of the regression coefficient. Then, we ran a full model with all predictors and compared the sign of the estimated regression coefficients. Only those predictors, whose sign did not change in the full model as compared with separate regression, remained in the final model reported in this paper and hence variables for birth intervals and age of women at last birth have been excluded. It must be noted that the same approach to detecting and avoiding multi-collinearity has already been used in several previous studies (Fan and Habibov, Reference Fan and Habibov2008; Habibov and Fan, Reference Habibov and Fan2008; Habibov, in press). The descriptive statistics for the remaining independent variables (after exclusion of multi-collinear variables) are provided in Table 1.

Table 1

Descriptive statistics for independent variables

Notes: Data are rounded up.

Data are for unweighted sample of women with the last births in the 5 year preceding the time of the Azerbaijan Demographic and Health Survey (AZDHS) in 2006.

Omitted categories are excluded. For instance, four wealth quintiles are included in the regressions and described in this table, while the fifth quintile is omitted from the regressions and hence not described. Likewise, eight regional dummies are included in the regressions and described in the table, while Baku is omitted and hence not described.

Source: AZDHS (2006).

3.4 Analytical strategy

For each of the four research questions, we developed a separate model. In each model, we commence with the examination of the descriptive distribution of the dependent variable under investigation and discuss the implications of this distribution. Multivariate analysis is followed. The same set of independent variables is used in all multivariate models. Such a multistage sequential modeling is based on the hypothesis that utilization of antenatal care is a multistage process in which decisions are sequential. The decision of using or not using antenatal care is made in the first stage. The binomial logit regression model is used due to the binomial nature of the outcome variable, namely, utilized prenatal care or not. During the subsequent stages, other series of decisions are made regarding the frequency, timing and quality of the usage. Negative binomial regression (NBREG) is employed to estimate determinants of frequency of the utilization. Ordered logit regression is employed to estimate the determinants of the timing of the first utilization and determinants of the quality of the received care. The computation in the three abovementioned regression models was restricted to those women who actually utilized prenatal health care.

Although at all of the stages the decision is affected by the set of factors described in the conceptual framework, we hypothesize that the effect of these factors can be different at different stages. Hence, for instance, some variables can be significant predictors of antenatal care utilization, but at the same time be insignificant predictors of the frequency of utilization or its timing. In fact, recent studies in transitional countries have confirmed the sequential nature of decision-making in maternal care utilization (Habibov and Fan, Reference Habibov and Fan2008; Sepehri et al., Reference Sepehri, Sarma, Simpson and Moshiri2008; Habibov, Reference Habibov2009).

4.1 Results of the model for antenatal healthcare utilization

The result of answering the question in regard to the usage of antenatal healthcare utilization shows that a significant number of women, about 22%, did not receive any antenatal care during their last pregnancy, while about 77% of women received some care. This figure is extremely high even in comparison with other countries with a similar level of economic development or similar socio-cultural background. To provide reliable comparison, we can use the data from the recently conducted comparable DHS surveys. For instance, in Moldova, another low-income transitional country, the share of women who did not receive any antenatal care is only 1.8% (National Scientific and Applied Center for Preventive Medicine (NCPM), 2006). Turkey, which borders Azerbaijan, has similar religious and cultural traditions; however, this figure is no higher than 18% (Hacettepe University Institute of Population Studies (HUIPS), 2004). In another neighboring country, Kazakhstan, the share of women who did not receive antenatal care is also only 6% (Academy of Preventive Medicine (APM), 2000).

The results of the binomial logit-regression model estimation on the probability of antenatal healthcare utilization are reported in Table 2. In total, the effect of 8 variables out of the 22 presented in the model is found to be significant. Among current pregnancy characteristics, two factors, birth order and wanting last child or current pregnancy, are found to be the significant predictors of antenatal care utilization. For a one-unit increase in the birth order, the odds of utilizing antenatal care decreased by 28.2%. Likewise, the odds of utilizing antenatal care are 39.8% lower for women who did not want the last child or current pregnancy. Affordability of service also plays an important role in determining prenatal utilization. The odds for utilizing prenatal care are 54.4% lower for women from the poorest households. Several variables are significant accessibility of service variables. Living in rural areas decreases the odds of prenatal care utilization by 35.1%. Similarly, living in the Lankaran-Astara, Quba-Xaçmaz, Aran and Dağlıq Şirvan regions significantly decreases the odds of antenatal care utilization. In fact, with the decrease in the odds of prenatal utilization between 73.5% and 59.4%, the regional residencies are the most powerful predictors of antenatal care utilization in Azerbaijan. Living in other regions also reduces the odds of utilization, but the variables are not significant in a strictly statistical sense. With respect to other explanatory variables in the estimated model, all of them have the expected direction, although the effect is not statistically significant.

Table 2

Multivariate analysis of determinants of antenatal care utilization (binomial logit)

Notes: % change denotes percentage change in odds for unit increase in an independent variable. *10% significance level; **5% significance level; ***1% significance level.

Data are rounded up.

The results of estimations are adjusted to a complex survey design though Taylor’s linearization series and are nationally representative.

Source: Azerbaijan Demographic and Health Survey (2006).

4.2 Results of the model for frequency of utilization

The result of answering the question about the frequency of antenatal healthcare utilization is graphically illustrated by Figure 2. As shown, the number of utilization events increased from one time during pregnancy – for about 12% of women, to three times – for about 16% of women and then steadily declined to 20 visits – for <1% of women. The distribution in Figure 2 can be assessed from the perspective of the national and international normative standards in antenatal health utilization. For a normal pregnancy, a pregnancy without high risk for antenatal complications, the Ministry of Health of Azerbaijan recommends monthly visits until the third trimester and twice monthly visits in the third trimester. However, <1% of the pregnant women received the required antenatal care. WHO uses less stringent guidelines by recommending at least four antenatal care visits for a normal pregnancy. Nevertheless, only approximately 10–11% of all women in Azerbaijan received the required volume of antenatal care.

Figure 2

Distribution of the number of times women utilized antenatal health care. Notes: Data are rounded up. Missing responses are excluded. n = 1259. Source: Azerbaijan Demographic and Health Survey (2006).

The results of the NBREG estimation on the frequency of prenatal care utilization are reported in Table 3. In the reported results, the effect of 3s variables out of the 22 presented in the model is found to be significant. Women’s socio-demographic and affordability perspectives are important predictors of utilization frequency. In terms of women’s socio-demographic perspective, for every additional year of schooling, the expected number of prenatal utilization visits increase by 4.4%. In terms of affordability, living in a richer household increases the expected number of visits by 20%, while living in the richest household increases the expected number of visits even more – by 30%. It can be concluded, therefore, that affordability of services is the most important factor explaining the frequency of antenatal care utilization in Azerbaijan. Other determinants presented in the estimated model are not statistically significant, although they have the expected direction.

Table 3

Multivariate analysis of determinants of frequency of utilization (NBREG)

Notes: % change denotes percentage change in expected count for unit increase in an independent variable. *10% significance level; **5% significance level; ***1% significance level.

Data are rounded up.

The results of estimations are adjusted to a complex survey design by Taylor’s linearization series and are nationally representative.

Source: Azerbaijan Demographic and Health Survey (2006).

4.3 Results of the model for timing of the first utilization

The result of answering the question about the timing of the first antenatal healthcare utilization shows that most women in Azerbaijan, more than 67%, utilize antenatal care during the first trimester. Nevertheless, a considerable share of women utilizes antenatal care later – approximately 25% during the second trimester and approximately 7% during the last trimester only. This finding shows that the timing of the first antenatal care utilization in Azerbaijan is roughly comparable with that of Moldova and Turkey – 72% and 66%, respectively (HUIPS, 2004; NCPM, 2006). However, the timing indicator in Azerbaijan is still worse than that in neighboring Kazakhstan where most women, 84%, have already utilized antenatal care by the beginning of the third month of pregnancy (APM, 2000).

The results of ordered logit estimation on the likelihood of timing of the first utilization are reported in Table 4. In total, two variables, women’s socio-economic characteristics and accessibility, are found to be significant predictors of the timing of the first prenatal visit. The increase in women’s educational attainments is negatively associated with the likelihood of the late prenatal visit. More specifically, for each additional year of women’s education, the odds of having the first antenatal visit in the third trimester are almost 14% smaller than in the first semester. In contrast, living in the Quba-Xaçmaz region is positively associated with the likelihood of late antenatal care utilization. The odds of having the first prenatal visit in the third trimester are approximately 93% higher for the women living in this region. Hence, the regional variables are most important in understanding the timing of prenatal care utilization in Azerbaijan. Other variables in the estimated model are not statistically significant, although they have the expected direction.

Table 4

Multivariate analysis of timing of utilization (ordered logit)

Notes: % change denotes percentage change in odds for unit increase in an independent variable. *10% significance level; **5% significance level; ***1% significance level.

Data are rounded up.

The results of estimations are adjusted to a complex survey design though Taylor’s linearization series and are nationally representative.

Source: Azerbaijan Demographic and Health Survey (2006).

4.4 Results of the model of quality of delivered care

The result of the quality index distribution is illustrated in Figure 3. The figure demonstrates the high polarization in the quality of delivered care in Azerbaijan. Approximately 40% of all women do not receive any quality care during their antenatal visits – the quality index is 0. By sharp contrast, around 20% of women received the highest quality of care – the quality index is 6 and approximately 22% received a high quality of care – the quality index is 5. In contradistinction, a much smaller number of women received mediocre or average quality of care with the quality of indexes varying from 2 to 4.

Figure 3

Quality of received antenatal health care for the last pregnancy. Notes: Data are rounded up. Missing responses are excluded. n = 1259. Source: Azerbaijan Demographic and Health Survey (2006).

Since the dependent variable, quality index, is ordinal and rank-ordered, the ordered logit regression model is fitted to estimate the probability of receiving a higher quality of antenatal services. The results are reported in Table 5. In the reported results, 10 variables out of the 22 presented in the model are found to be significant predictors. Among women’s socio-economic variables, the increase in women’s educational attainments is strongly associated with the rising likelihood of receiving a higher quality of care. For a 1-year increase in women’s educational attainment, the odds of receiving a higher quality of antenatal care increased by more than 10%. Likewise, having a higher body mass index is also associated with the improved likelihood of receiving better care, but the effect is negligible. In contrast, the accessibility perspective is most important in shaping the determinants of the quality of antenatal services received. All regional dummies in the model are significant, signaling that the quality of antenatal care in the region is much lower than in Baku. For example, the odds of getting all six types of prenatal care are 92% smaller for those who live in the Yuxarı Qarabağ region and 84% smaller for those who live in the Ganca-Qazax region.

Table 5

Multivariate analysis of quality of received antenatal care (ordered logit)

Notes: % change denotes percentage change in odds for unit increase in an independent variable. *10% significance level; **5% significance level, ***1% significance level.

Data are rounded up.

The results of estimations are adjusted to a complex survey design though Taylor’s linearization series and are nationally representative.

Source: Azerbaijan Demographic and Health Survey (2006).