Inequalities in reproductive, maternal, newborn and child health in Vietnam: a retrospective study of survey data for 1997–2006

We used publicly available data collected at four points in time: 1997 and 2002 cross-sectional data from the Demographic and Health Surveys (DHS) and 2000 and 2006 cross-sectional data from the Multiple Indicator Cluster Survey (MICS) [20, 21]. The DHS and the MICS collect comparable, nationally representative data on health and nutrition outcomes, intervention coverage, demographic information and a wealth index. Using data from both DHS and MICS to analyze trends in health outcomes and service coverage is common in the literature [22‐25]. Each survey contains information provided by women of reproductive age (15–49 years old) from at least 7,000 households (1997: 7,001; 2000: 7,628; 2002: 7,048; 2006: 8,355; each with a response rate of at least 98%). As a consequence of the stratified sampling survey design used by both DHS and MICS, different observations have different probabilities of selection. We adjusted for this by applying weights to each observation equal to the inverse of the probability of being sampled [26].

DHS and MICS contain information required for estimating infant and under-five mortality rates, malnutrition and prevalence of common child illnesses (see Table 1 for a list of the health and nutrition outcome indicators included in the study).

The two data sources we use in this study call for two different methods of estimating infant and under-five mortality. DHS surveys collect full birth histories from mothers, which enables direct estimation of mortality [27]. Using this approach we constructed life-tables with survival times, which were then used to calculate mortality rates. MICS surveys collect incomplete birth histories, which include data on the number of children born and the number of children surviving, but no information on dates of these events. Incomplete birth histories require indirect estimation of mortality, which involves applying a model life table to the available data, i.e. number of children ever born and number of children still alive [28]. We used the QFIVE software to calculate mortality rates with the indirect estimation method [29].

We calculated prevalence of pneumonia and diarrhea, which remain the leading killers of children globally; in 2008 they together accounted for one-third of deaths among children less than five years of age [30]. To measure inequalities in malnutrition, which contributes to one-third of child deaths globally, we calculated prevalence of underweight, stunting and wasting [31]. Only the 2000 MICS survey included data on these three indicators, which prevented a trend analysis. However, all four surveys included data on low birth-weight, which enabled us to study trends in an indicator considered an important predictor of a newborn’s survival, growth, and long-term health and psychosocial development, as well as being associated with a pregnant woman's overall health status [32].

Although the pathways to child mortality and morbidity are complex and influenced by economic development and other social determinants of health, interventions in the health system contribute to improving health outcomes [33]. A recent systematic review of the evidence has identified 56 interventions that are key to improving RMNCH outcomes [34]. It has been estimated that if these interventions were universally available, they could reduce global child mortality by two-thirds and maternal deaths by at least half [35, 36]. It is therefore of interest to study inequalities in coverage of these interventions (see Table 2 for a list of the intervention coverage indicators for which data were available in our data sources, and their definitions, included in the study).

We measured inequalities by several stratifying variables. Direct measures of living standards such as income, expenditure and consumption are rarely collected in health surveys such as DHS and MICS. However, both DHS and MICS collect data on living conditions and household assets, which can be used to compute an indirect or proxy measure for living standards called the asset or wealth index [37, 38]. The index is calculated for each household in a survey to divide the sample into quintiles: five groups of equal size, from poorest to richest.

We also estimated inequalities by education of mother, ethnicity, region, urban versus rural residence, and sex of the child. The two surveys use a different number of categories for the highest level of education obtained by the mother. MICS uses seven categories in the 2000 survey and five in 2006. DHS uses six categories in both 1997 and 2002. To address this we constructed four categories for education of mother: no education, primary education, lower secondary education, and upper secondary and higher education.

Since there are more than 50 ethnic minority groups in Vietnam, comprising 14% of the total population, we combined them into one group and compare their outcomes with those of the majority ethnic group [39]. We also assessed inequalities by urban versus rural residence and by region. MICS uses eight regions, while DHS uses seven regions. To adjust for this we obtained the list of provinces included in the MICS regions and then allocated those provinces in the same manner as was done in the DHS, leaving us with seven regions for analysis in both surveys.

We used several complementary methods to study inequalities in MNCH, all of which have found broad acceptance in the health inequality literature [40]. We tabulated health and nutrition outcomes and intervention coverage by stratifying variable, for example under-five mortality by wealth quintile and immunization coverage for the seven regions. We calculated ratios for health and nutrition outcomes and intervention coverage, such as poorest vs. richest quintile, ethnic minority vs. majority, and rural vs. urban place of residence. These measures of inequality provide a broad descriptive analysis of inequalities in MNCH. However, while tabulations and ratios are intuitive in their presentation, they have certain limitations. For example, ratios ignore patterns in the middle of the wealth and education distributions and tabulations do not shed light on variations within quintiles or education levels. A more comprehensive picture of inequality across the full distribution is provided by the concentration index (CI), which is a measure of the magnitude of inequality that can be compared across time, country and region [41].The CI has been used to compare inequality in child mortality [42], child immunization and child malnutrition [43]. The CI is defined as twice the area between the concentration curve and the line of equality.^a The CI varies from −1 to 1, with 0 indicating perfect equality. The index takes a negative value if it lies above the line of equality, which indicates a disproportionate concentration of the variable among the poor. If the CI of mortality is negative it means that mortality is higher among the poor. It takes a positive value if it lies below the line of equality, indicating that the variable is disproportionately concentrated among the rich. If the CI for immunization is positive, it means that the rich benefit from higher coverage of this intervention. The CI requires data that can be ranked in a meaningful way. This requirement holds for measures of living standards, such as the wealth index, and education levels, but there is no way to meaningfully rank individuals or households by ethnicity, geography, or gender. We were therefore limited in generating concentration indices for health and nutrition outcomes and intervention coverage indicators for the living standard and education stratifying variables. To examine trends in inequality over time, we tested the null hypothesis that there was no change between 1997 and 2006.

With the exception of mortality estimates from MICS data, which were computed using the QFIVE software, Stata version 10.0 was used for all analyses of this study.

The mean infant mortality rate (IMR) decreased from 32.0 deaths per 1,000 live births in 1997 to 18.8 in 2006, while the mean under-five mortality rate (U5MR) decreased from 41.2 to 23.0. There were inequalities in mortality rates by living standards (see Table 3). For example, in 1997 a child in the poorest 20% of the population was almost three times (2.8) more likely to die before his or her fifth birthday compared to his or her peers in the richest 20% of the population. These inequalities increased over time; in 2006 the poor/rich ratio was 4.5. Similar inequalities were identified for the IMR.

The largest reductions for both IMR and U5MR in 1997–2006 were recorded in the richest and poorest quintiles; the middle quintiles recorded smaller reductions. Within each survey the mortality rate generally decreased from the poorest quintile to the 2^nd quintile and so on, with a few exceptions (see Figure 1).

The CI confirmed the living standards inequalities identified through tabulations and ratios. The CI in 1997 was −0.1639 for IMR and −0.1828 for U5MR. By 2006, disparities had increased to −0.1896 for IMR and to −0.1875 for U5MR.

While smaller in magnitude than for living standards, there were inequalities by the other stratifying variables as well (see Table 3). In 2006, the U5MR for children whose mother had no education was 41.0 deaths per 1,000 live births, while the rate for children whose mother had completed upper secondary education or higher was 6.7. In 1997, the U5MR of ethnic majority and minorities groups were 34.7 and 72.5 deaths per 1,000 live births, respectively (a ratio of 2.1). This inequality increased slightly over time; in 2006 the ratio was 2.3.

The analysis of differences by sex of the child found that boys were more likely to die before their fifth birthday in 1997, 2000 and 2002.The U5MR for boys was 49.1 deaths per 1,000 live births in 1997; for girls it was 32.6. In 2006, inequalities ran in the opposite direction; the rate was 17.5 for boys and 25.5 for girls. There were also geographic inequalities. The IMR and U5MR were about twice as high for children living in rural versus urban areas in 1997–2002; in 2006 it was about six times as high. In 2006, mortality rates were highest in the Northern Uplands and Red River Delta. The largest reductions in under-five mortality between 1997 and 2006 occurred in the Mekong River Delta.

Inequalities in the three nutrition outcomes for which we only had data from the 2000 MICS survey were identified for all stratifying variables (see Table 4). The poor/rich ratio was 2.75 for stunting, 2.41 for underweight and 1.49 for wasting. The CI in 2000 was −0.1067 for underweight, -0.1216 for stunting, and −0.0967 for wasting. Almost half (46.5%) of children with mothers without education were underweight, while the figure for upper secondary or higher education was 20.0%. Nutrition inequalities were less pronounced by ethnicity, urban versus rural residence, and sex of child.

Inequalities by living standards for low birth weight were considerable, although they decreased over time. For example, the CI for low birth weight declined from −0.2810 to −0.1236 between 1997 and 2006. Inequalities by education (CI = −0.0040) and ethnicity (minority/majority ratio = 1.11) had almost disappeared by 2006.

Mixed results were found for child illness inequalities by living standards as measured by the CI (see Table 5). The CI for proportion of children with suspected acute respiratory infection decreased from −0.0794 in 1997 to −0.0231 in 2006, while the CI for proportion of children with diarrhea increased from −0.0603 in 1997 to −0.1414 in 2006. We also found inequalities by the other stratifying variables, but they were generally less pronounced. We did not identify any strong patterns of change over time in inequalities in prevalence of child illness by education, ethnicity, region, urban versus rural residence, or sex of child.

The results for intervention coverage are presented in Table 6. We found small inequalities for family planning interventions in 1997–2006. Almost all maternal and newborn health indicators displayed decreasing inequalities by both living standards and education over time. For example, for at least one antenatal care contact the living standard CI decreased from 0.2833 in 1997 to 0.0587 in 2006, while the education CI decreased from 0.0765 to 0.0463. Inequalities were more pronounced for interventions requiring multiple service contacts. For example, while pregnant women in the richest quintile were 1.5 times more likely than pregnant women in the poorest quintile to have been attended at least once during pregnancy by skilled health personnel in 2002 (CI = 0.0776), they were almost five times as likely to have been attended at least four times (CI = 0.3170). In 2006, pregnant women in the poorest quintile were about half as likely as the rich to have delivered in a health facility (CI = 0.1182) and to have had their births attended by skilled health personnel (CI = 0.1025). These indicators exhibited considerable inequalities by other stratifying variables as well: the CI by education for delivery in facility and skilled birth attendance were 0.0669 and 0.0612, respectively, in 2006. The ethnic minority/majority ratios for these two interventions were 0.43 and 0.47, respectively.

Two measures exhibited pro-poor inequalities: early initiation of breastfeeding and exclusive breastfeeding. For example, in 2006, women in the poorest quintile of the population were more than twice as likely as women in the richest quintile to exclusively breastfeed their children until they reached six months of age (CI = −0.1472). A similar pattern was found for education (CI = −0.1151) and ethnicity (minority/majority ratio = 1.68).

We found inequalities for interventions targeted to children less than five years of age, but they were less pronounced than for maternal and newborn health interventions. Almost all child health indicators displayed decreasing inequalities by both living standards and education over time. For example, for measles vaccination the living standard CI decreased from 0.0698 in 1997 to 0.0341 in 2006, while the education CI decreased from 0.0575 to 0.0211. Interventions that require multiple service contacts displayed larger disparities. For example, in 2006, inequalities were larger for DPT immunization (living standard CI = 0.0641; education CI = 0.0539), which requires three doses, than for BCG (living standard CI = 0.0164; education CI = 0.0121), which only requires one. Care-seeking for pneumonia displayed greater inequalities (poor/rich ratio of 0.70 in 2006) than care-seeking for diarrhea (0.89).

While using data from two different types of surveys enabled the study of trends over a 10-year period, it is also a source of some limitations of the study. Although DHS and MICS are similar in key respects and measure most of the same indicators for MNCH, and use the same methodology to calculate the wealth index, there are differences. In general, inequalities were found to be larger in the two MICS surveys compared to the two DHS surveys. For mortality this may be due to the fact that two different methods to estimate mortality had to be applied. The methods may suffer from different errors, for example random errors in sample surveys or systematic errors due to misreporting. However, we have not been able to reach a conclusion as to why MICS generally present larger inequalities for health and nutrition outcomes and intervention coverage. There were some differences in the definitions of the education, ethnicity and region variables. We explained in the methods section how we addressed these differences.

Another limitation of this study is that the measure of living standards - the wealth index - is correlated with other stratifying variables, such as education, ethnicity, and rural or urban place of residence, which raises another important point of discussion. We believe that differences in education, ethnicity, and residence have a direct impact of inequalities in health and nutrition outcomes and intervention coverage. However, one plausible alternative hypothesis is that the predominant source of inequality is living standard status and that the other stratifying variables are proxies for living standards. For example, inequality by ethnicity could be a reflection of the fact that ethnicity may proxy for being poor, less educated, and living in environments less conducive to good health outcomes.

Finally, in recent years there has been a growing recognition that there are potential problems with the CI as a tool to measure inequality [44]. For example, one study found that different rankings could be obtained if the measure of inequality is inversed, for example if inequalities in ill health are measured rather than inequalities in health (sometimes referred to as the “mirror problem”) [45]. This would be a particular problem if analysis of the development of the CI over time would show different trends depending on which of the two indicators would be used. We have tested our data for this possibility, and have not observed any differences in trends.