Patterns of social inequalities across pregnancy and birth outcomes: a comparison of individual and neighborhood socioeconomic measures

We linked data from the Canadian Maternity Experiences Survey (MES), a national, complex and unweighted sample of 6,421 (weighted sample =76,508) women sampled from the 2006 Canadian census, to neighborhood data from the same census [35]. Detailed information about the MES can be found elsewhere [36,37], but, briefly, this survey was conducted in 2006 to collect data concerning pregnancy and birth in a national sample of women [35,37]. Women were selected from the 2006 census using stratified random sampling without replacement. A total of 8,244 women met the eligibility criteria of being 15 years of age or older, having had a singleton live birth, and living with their baby at the time of data collection [37]. The response rate was 78%. After applying survey weights, the 6,421 respondents represent 76,508 women in the population [36]. Data were collected through a computer-assisted telephone interview and lasted 45 minutes. Infants were between 5 and 10 months of age at the time of the interview.

The census data were reported at the level of dissemination areas (DAs), a census unit of 400 to 700 persons [38]. Only one woman in each DA was interviewed for the MES [5,39].

The MES project was presented to Health Canada’s Science Advisory Board, Health Canada’s Research Ethics Board and the Federal Privacy Commissioner, and was approved by Statistics Canada’s Policy Committee [36]. Approvals for this analysis were obtained from the St. Michael’s Hospital Research Ethics Board and the Research Data Centre Access Granting Committee of Statistics Canada.

The demographic variables (age and parity) and SEP measures are detailed in Table 1.

The SEP measures included:

Table 2 presents detailed definitions of the MCH outcomes, including reference categories. Reference groups were chosen based on those most commonly reported in the literature or, in some cases, based on the category where there was the largest sample. As mentioned above, we organized the 25 MCH outcomes we used in the study into five groups based on areas of interest and periods around pregnancy and birth. Each of the five groups of MCH outcomes represents different health care needs of the mother and/or her infant during pregnancy and after birth. The groups of MCH outcomes were as follows:

We followed the MES Users’ Guide [41] for applying the weights [36]. Missing data were less than 10% for most of the variables. Bivariate associations between each SEP measure and MCH outcomes were calculated using estimates of the proportions and odds ratios (ORs) with 95% confidence intervals (95% CI). We then estimated ORs using multivariable logistic regression models for each of the MCH outcomes while taking into consideration potential confounders of age and parity (the condition or fact of having given birth previously). The models were as follows: Model 1, adjusted for one SEP measure only; Model 2, adjusted for one SEP measure and age; Model 3, adjusted for one SEP measure and parity; and Model 4, was adjusted for on SEP measure, age and parity. We determined that a variable (age or parity) was a confounder when the SEP estimates results changed by 10% or more from the unadjusted model (bivariate associations). However, this change was not noticed and therefore we decided to present only bivariate results in this paper. We assessed the patterns of inequalities based on the direction and magnitude of each association, while comparing the ORs of poor health category with the better health category (reference category) for each SEP measure. The level of significance was set to 10%.

Generally, the five groups of MCH outcomes showed a consistent positive gradient for most (23/25) of the health outcomes across one or more SEP measures; the higher the SEP level, the better the MCH outcome. For the significant associations, mothers with lower SEP (compared to those with higher SEP) had poorer MCH outcomes. Among the five groups of MCH outcomes, we observed more consistent association in Group A of health status indicators of the mother and infant. This was followed by Group B of parental care (Figures 2B1-6), Group C of labor and delivery experiences (Figures 3C1-4), Group E of postpartum infant care and reflection of the overall experience of health care services (Figures 5E1-4), and lastly Group D of medical care during the neonatal period, which had the least significant associations with SEP. Non-significant association was found for infant hospital readmission, admission to the NICU, and greater-than-average length of stay at the hospital (Figures 4D1-5).

While the direction of the significant associations was generally positive (OR > 1), some associations had a negative gradient (OR < 1) (Table 3). That is, those with the highest SEP were more likely to report poorer MCH outcomes than those with the lowest SEP. Those negative gradients were not always in the expected direction. For example, in Group B of prenatal care, there was a negative association between education and timing of the ultrasound. More mothers with higher education conducted the ultrasound later than the recommended time (>18 weeks or later) compared to mothers with lower education (Figure 2B2). Also, mothers with higher household income were more likely to report no HIV testing than mothers with lower household incomes (Figure 2B3). Negative associations were also observed for type of birth (Figure 3C1), and epidural for a vaginal birth (Figure 3C4). In Group D of neonatal care, all SEP measures were negatively associated with infant needing non-routine care (Figure 4D3). Lastly, in Group E of postpartum infant care, neighborhood SEP was significantly and negatively associated with any breastfeeding at 3 or at 6 months (Figure 5E3-E4).

Another important finding is that not all categories of the SEP variables had significant associations with the MCH outcomes. This inconsistency was more evident for the combined SEP measure, which did not show a steep consistent gradient for half (10/20) of the associations with the MCH outcomes. Despite the fact that some CIs overlapped, Table 3 shows that the category low individual-high neighborhood of the combined SEP measure had the highest ORs compared to other categories of this variable.

Comparing the magnitude of the MCH inequalities (the unadjusted OR (UORs) in the highest and lowest SEP categories, Figures 1, 2, 3, 4, and 5), we found that mothers’ education had the highest ORs and larger inequalities, which was followed by combined SEP, and household LICO. Neighborhood SEP had the smallest magnitudes (point estimates).

For the significant and positive associations, the unadjusted odds ratios (UORs) for education were higher compared to those in other SEP categories. We observed the largest magnitudes between the highest and lowest SEP by education in Group A of health status indicators for the mother and infant (UOR for self-rated health was 3.75, 95% CI: 2.95-5.43) (Figure 1A2), for post-partum depression UOR was 2.44 (95% CI: 1.75-3.41) (Figure 1A4), and for BMI < 18.5, UOR was 3.17 (95% CI: 2.16-4.64) (Figure 1A5). The next strongest association with education came among Group B of prenatal care variables (Figures 2B1-6). For example, the UOR for first prenatal visit was 3.52 (95% CI: 2.45-5.07) (Figures 2B1), and for received enough information UOR was 1.89 (95% CI: 1.34-2.67) (Figure 2B4). The third group in terms of strength of associations with education was Group E postpartum infant care and maternal reflections on the overall experience of health care (Figures 5E1-4). In this group, the UOR of any breastfeeding at 3 months was 2.93 (95% CI: 2.28-3.77) (Figure 5E3). The fourth group in this order was Group C of labor and delivery experiences (Figures 3C1-4), which had only two variables significantly associated with education: shaving for vaginal birth (UOR:1.60; 95%CI:1.14-2.24) (Figure 3C3) and epidural for vaginal birth, however this variable had negative associations with education (UOR: 0.74; 95% CI: 0.58-0.94) (Figure 3C4). Lastly, only one variable out of four in Group D (medical care in the neonatal period) was significantly associated with education but this was a negative association (infant needing non-routine care, UOR:0.80; 95% CI: 0.64-0.99) (Figure 4D3).

Combined SEP was the measure with the second-greatest differences between the highest and a lowest SEP level (Figures 1, 2, 3, 4, and 5). The highest UOR for combined SEP was 3.30 (95% CI: 2.42-4.50) in the association with postpartum depression (Figure 1A4). The patterns of the strength of gradients were similar for combined SEP to what we saw for education; that is, the variables grouped as A, “health status indicators for the mother and infant” had the largest gradients (Figures 1A1-6), while those grouped as B, “medical care during the neonatal period” had the smallest (Figures 4D1-5). However, as we mentioned above, the combined SEP had inconsistent gradients and the category of low individual-high neighborhood showed the poorest health outcomes. Household income was the third-strongest associated SEP measure, with the highest UOR of 2.63 (95% CI: 2.09-3.32) in the association with postpartum depression (Figure 1A4). Household income had the same pattern regarding the associations with the groups of MCH outcomes that we observed with education (Figures 1, 2, 3, 4 and 5).

The smallest magnitudes of inequalities we observed was for neighborhood SEP, with highest UOR of 1.73 (95% CI: 1.31-2.28) associated with the first prenatal care visit (Figure 2B1). Neighborhood SEP was associated with the groups of MCH outcomes in the same trend that we observed for the other SEP measures (Figure 1, 2, 3, 4, and 5). In addition, all health indicators that were associated with neighborhood SEP were also associated with the combined SEP, but the combined SEP showed inequalities for 20 MCH outcomes whereas the neighborhood SEP was only associated with 12 of these health indicators (Figure 1, 2, 3, 4, and 5).

Examining the patterns of magnitude of SEP inequalities for the positive associations with UORs > 1 within the groups of outcome variables across the different SEP measures, we found that Group A of health status indicators for mother and infant (Figures 1A1-6) showed the greatest SEP inequalities with an UOR of 3.75 (95% CI: 2.59-5.43) for the association between education and self-rated health (Figure 1A2), and lowest UOR of 1.27 (95% CI: 1.10-1.48) between household income and pre-pregnancy BMI (>24.5) (Figure 1A6). This was followed by the variables in Group B of the prenatal care (Figures 2B1-6) with highest UOR of 3.52 (95% CI: 2.45-5.07) for the association between education and first prenatal visit (Figure 2B1) and lowest UOR of 1.19 (95% CI: 1.01-1.40) for the association between weight gained during pregnancy (above recommended range) and household income (Figure 2B6). Next was Group E of postpartum infant care and maternal reflections on overall experiences of health care (Figures 5E1-4). Highest UOR in this group was 2.93 (95% CI: 2.28-3.77) for the association between education and any breastfeeding at 3 months (Figure 5E3), and the lowest UOR was 1.17 (95% CI: 0.98-1.39) for the association between household income and satisfaction at all times (Figure 5E2). The associations within Group C labor and delivery (Figures 3C1-4) were next, with highest UOR of 1.98 (95% CI: 1.46-2.69) for the association between combined SEP and shaving for vaginal birth (Figure 3C3), and lowest UOR (1.31; 95% CI: 1.12-1.54) in the association between household income and position for the vaginal birth (Figure 3C2). Group D of medical care during the neonatal period had the fewest significant associations and smallest magnitude of effects with SEP measures (Figures 4D1-5). The highest UOR in this group was 1.71 (95% CI: 1.07-2.75) in the association between combined SEP (high individual-low neighborhood category) and greater than average length of stay at the hospital after caesarean section (Figure 4D5), and lowest UOR of 1.17 (95% CI: 0.97-1.41) in the association between education and greater than average length of stay at the hospital after vaginal birth (Figure 4D4).

The magnitude for the negative associations (OR < 1) between SEP measures and MCH outcomes was highest in the association between neighborhood SEP and any breastfeeding at 6 months (UOR: 0.87, 95% CI: 0.76-1.00) (Figure 5E4), and lowest for the association between combined SEP and infant needing non-routine neonatal care (UOR: 0.57, 95% CI: 0.47-0.71) (Figure 4D3).

Overall, we found consistent socioeconomic gradients across most (23/25) of the MCH outcomes in the five groups for one or more SEP measures (individual, neighborhood or both). In general, mothers with lower education, lower household income (at or below the LICO), residing in lower SEP neighborhoods and having low individual-high neighborhood combined SEP had poorer MCH health outcomes than mothers at the higher SEP levels. These results support the already rich literature showing socioeconomic inequalities in MCH outcomes [11,17,18,20,23,42]. However, the current study also identified some groups of MCH indicators where inequalities were steeper than they were in other groups. We found that the SEP measures showed stronger associations (higher ORs) among outcomes belonging to the health status of the mother and infant (Group A), this was followed by prenatal care outcomes (Group B), postpartum infant care and maternal reflection on overall experiences of health care (Group E), labor and delivery (Group C), and, lastly, medical care during the neonatal period (Group D). The last group had the fewest and smallest significant associations with SEP.

The differences in the magnitude of inequalities between groups of variables suggest different pathways might be operating between SEP and each of the groups. For example, higher inequalities in Group A of health status of the mother and infant outcomes suggest a cumulative effect, in particular of low SEP, operating prior to and throughout the prenatal period, possibly composed of multiple social determinants of health that affect MCH outcomes and for which SEP is often an indicator (e.g., employment, housing, ethnicity etc.) [18].

While most of the significant associations between SEP measures and MCH outcomes showed positive gradients; higher SEP measure indicated poorer MCH outcomes, some of the associations with MCH outcomes were negative where women with higher SEP were more likely to experience the poorer outcome). These negative associations were more prevalent during birth and delivery (Group C) and medical care during the neonatal (Group D), for example higher-income women being less likely to have prenatal HIV testing and more likely to conduct a late first ultrasound. Reasons for these observed associations are not clear and should be the subject of future investigations. Our finding support the growing literature showing that receipt of epidural for vaginal birth is more likely among women of higher SEP [43]. However, our finding that caesarean sections occurred more frequently among those with higher income has not been reported in previous Canadian studies. One study showed that age-adjusted caesarean sections were higher in women from low-income neighborhoods [44]. However, another study from Nova Scotia, Canada, showed that neonatal medical interventions, such as induction of labor and caesarean section, did not differ by income [45]. One explanation for these differences is that obstetric practices may vary by province and that Nova Scotia differs from other provinces included in our study. Alternatively, obstetric practices may have changed since the time of the Joseph et al. study [46-48]. Results of past studies, such as Joseph et al., might be related to the universal access to these services in Canada as national public health care systems have been shown to increase equity of access to health care and result in fewer health inequalities in multicounty comparisons [46-48]. Thus, if funds and services were to be cut, we might see a change in the relationship between medical care and SEP similar to the changes observed in Australia where there was an increase in caesarean sections as a result of a shift from public towards private hospitals for delivery [49,50].

Maternal education had steeper gradients in the associations with different MCH outcomes (UOR up to 3.75, 95% CI: 2.59-3.75 in the association with self-rated health), followed by combined SEP (UOR up to 3.30, 95% CI: 2.42-4.50 in the association with postpartum depression), then household income (UOR up to 2.63, 95% CI: 2.09-3.32 in the associations with postpartum depression), and, finally, neighborhood income (UOR up to 1.73, 95% CI: 1.31-2.28 in the association with first prenatal visit). However, household income and combined SEP revealed more frequent statistically significant associations than those observed for maternal education. This result corresponds with a systematic review which indicated that income was more frequently associated with birth outcomes than education [18]. The results also suggests that education is different from income and combined SEP and should not be used interchangeably as a proxy for those measures [10,20].

The pattern we found, in which maternal education had a steeper gradient between high and low SEP than did the other measures, indicates that maternal education can reveal the depth of health disparities regarding MCH indicators. Previous studies have shown that maternal education is associated with many MCH outcomes, including preterm birth and stillbirth [14,15], low birth weight [51], inadequate use of prenatal healthcare services [17], and risky maternal behaviors during pregnancy [52]. Lower education, meanwhile, has been associated with preterm birth, small for gestational age, and other complications [11]. There may be multiple pathways linking education and maternal and infant health status outcomes. Maternal education might reflect not just individual and household economies, but other pathways, such as access to information and knowledge, problem solving, social networks and involvement, and social prestige [4].

In our study, neighborhood SEP alone exhibited the weakest magnitude of the associations with MCH outcomes. This might have been expected, as prior studies have also found significant but small effects of neighborhood measures on a variety of MCH outcomes [23,27-29,31]. Despite increased attention on the effects of the social environment on health, few studies were conducted on neighborhood SEP compared to individual-level SEP [4]. Also, our finding that MCH outcomes are more strongly associated with individual-level SEP than with neighborhood SEP might reflect the fact that neighborhood SEP in our study was an aggregate of household LICO, whereas household LICO itself might be a more accurate and direct reflection of resources available to a household. Studies using median income to measure neighborhood SEP (as opposed to neighborhood-derived variables based on education or employment) have shown consistent associations with increased risk of preterm birth [13,28].

Combined SEP showed significant associations, which were similar to those found for household income; however, the magnitude of the associations was higher for the combined SEP. Combined SEP might be a more powerful measure than individual income alone or neighborhood SEP alone in revealing social inequalities in MCH since it takes into account both the individual and neighborhood dimensions of SEP. In addition to individual achievement, it captures resources or problems at the neighborhood level (e.g., high-income neighborhoods may have more facilities and services, fewer neighborhood problems and lower crime rates that reflect larger socioeconomic and social processes and policies) [25]. For example, life-long residency in a low-income neighborhood has been associated with low birth weight, and this association was stronger among women from ethnically identified groups [53].

Most of our outcomes (23/25) were associated with combined SEP. The magnitude of the associations with combined SEP was weaker than it was for education, but stronger than that for household income and neighborhood SEP. The consistency of associations for the combined SEP variable in the current study indicates the added value for measuring multiple levels of SEP simultaneously. The combined SEP was constructed based on aggregate income in a neighborhood [39]. While individual SEP and neighborhood SEP represent different sources contributing to socioeconomic inequalities, the combined SEP measure disenables the cumulative effects of individual and contextual exposures. For example, combined SEP might capture specific mediators and pathways acting between residential neighborhoods SEP and mothers’ education or income, such as social networks, social cohesion, and/or higher use of health care services among mothers with higher SEP [18]. This allows us to consider how studies that use one level of SEP might either show higher estimation than the actual effect of a SEP measure, or over-adjust for the SEP effect when adjusting for many individual SEP measures.

However, the combined SEP did not show a consistent gradient for all associations. In 10/20 of the significant associations, the category of low individual-high neighborhood SEP showed the greatest difference from the reference category of high individual-high neighborhood SEP. This suggests that the financial and psychological strain of being low income within a high-income neighborhood contributes to adverse health [54,55] and should be the focus of futures studies to unpack the pathways by which individual and neighborhood SEP simultaneously impact health.