Ninth Annual Maternal and Child Health Epidemiology Conference / December 10-12, 2003
DR. CATHERINE CUBBIN: Thank you and good morning. I’m very pleased to be on this panel of speakers today and to have the opportunity to present an overview of our study, which his purpose was to inform the measurement of SCS in racial ethnic disparities in maternal and infant health. So a study was published in public health reports in 1991. It was for a special issue on racial ethnic disparities, and I just want to acknowledge my co-authors Dr. Paula Braveman, who wasn’t able to come to this meeting. She is actually doing some work in Geneva with WHO, Kristen Markey, Susan Edeger and Gill Chavez. I have a couple copies of the paper if anyone is interested and I’m happy to sent anybody else copies if you just give me your Email address or address. So the purpose of this study was to inform how SCS is measured and interpreted and studies of racial ethnic disparities, and four important maternal and infant health indicators focusing on the two SCS measures; education and income, used the most often in the United States. We wanted to address a general audience of researchers who might conduct or interpret studies of racial ethnic disparities and health or who might include race ethnicity or SCS variables in other studies. We had conceptual and impure concerns about SCS measurement. Although the term socio economic status and socio class appear throughout the health literature, they are rarely defined.
Diverse measures are used to reflect these concepts including income, educational attainment, economic assets and occupation. However, justification is rarely provided for the measures selected, how and when it is measured or its interpretation. Several studies have shown for different populations and outcomes that correlations among different SCS measures can be modest and that associations between different SCS measures can vary both the measure and with the health outcome. Based on these findings, conclusions about racial ethnic disparities might vary not only by whether one measure was used rather than another to reflect SCS, but also by how the variable was specified, and for example, in the absence of adequate measurement of SCS, racial ethnic disparities and health are often interpreted either explicitly or implicitly as reflecting inherent biological and cultural differences and we think that’s very problematic.
So by SCS we are referring to wealth and the power and prestige closely associated with it, which may be reflected in occupational status, educational attainment and other measures. Despite prevailing social science theory and substantial epidemiologic evidence that SCS is multidimensional, the health literature is replete with studies that claim to have control for SCS when they have used a single or inadequate measures. So for this study, we used a face-to-face statewide representative survey of over 10,000 English or Spanish speaking women who gave birth in California during 1994 and 1995. The surveys were then linked with birth certificate data. So let me just define some terms I’m going to be using. By dimensions of SCS, I’m referring to different constructs, such as education or income. We distinguished different dimensions of SCS from different specifications of a given SCS dimension. So examples of different specifications would include specifying income as just income in dollars or as a percentage of the federal poverty level or in relation to family size specifying a continuously or categorically and if the latter how it is grouped, say in dequintiles, certiles, 100 percent increments of the Federal poverty level, et cetera, and for education, examples include whether we’re measuring maternal, paternal or household education and whether continuously or categorically.
For this study, we used only individual or household level measures of before tax family income. We examined several ways of specifying income that are listed here. So a percent of the Federal poverty level divided into five categories continuously as just the log of income in dollars as income in dollars grouped into cortiles, as income divided by family size as a continuous measure and then as income divided by family size grouped into cortiles. We six different measures of education. Birth certificates gave us information on both maternal and paternal education in years. The survey included only maternal education, which was grouped in levels according to completed credentials. We also examined educational level by converting the number of years from the birth certificate into levels according to credentials, and by household education we mean the education of the most educated parent. We conceptualize race ethnicity as reflecting the large geographic region of family origin of the mother, which her effect her experiences and/or her responses to them. We had four mutual exclusive categories that are listed here. There are very few black Latinas in California, and there were too few Native Americans or other groups to analyze separately. Our dependent variables were low birth weight, prenatal care starting after the first trimester or no prenatal care, unintended pregnancy and whether a woman intended to breastfeed. So in order to assess how different SCS measures could make a difference in conclusions, we first examined correlations among different dimensions and specifications of SCS overall and within each racial ethnic group. We then examined the associations between each SCS measure and each of the four health indicators.
First unadjusted overall and again in each racial ethnic group, then adjusted for age, parity and race ethnicity. We used logistic regression for all of our models. So examining unadjusted odds, the odds adjusted for age imparity and then adding in different SCS measures, we were able to determine whether different SCS measures made a difference in conclusions regarding racial ethnic disparities. So I’ll show you some findings. In this matrix shows for the entire sample, correlations among seven different SCS measures. Now, I’m just going to point out a few things. So if you look at the first three rows, which was shown in blue, you’ll see that the three different specifications of income examined here correlated highly with each other, but looking at rows four through seven in columns one through three, which are shown yellow, you’ll see that correlations between different income and education measures were only moderate between .54 and .6. Correlations between maternal and paternal education were also moderate, which are shown here in green. So just a glance at this, this is for the Latina women only. You’ll note that the correlations among income measures are still strong, but less though than the overall sample, and looking at rows four the seven, columns one through three, you’ll see that the correlations between the three income measures and the four education measures were weaker than in the overall sample. Only .3 to .4 versus .5 to .6 in the whole sample.
So these are correlations for the African-American women, and we also found weaker correlations here, but they were not as weak as for the Latina women. Okay, so the basic story of our models is that everything varied. Just keep that in mind, vary, vary, vary. We found that the associations between SCS and the dependent variables depending on the SCS measure we used both by dimension and how it was specified and on the dependent variable we were looking at. We also found that the magnitude and the statistical significance varied. We saw these variations in both the unadjusted and adjusted models. So for example, this table shows the adjusted associations between breastfeeding intention on the one hand, and other, three different income and two different education measures. The far left column shows you how the SCS variables were categorized. The SCS measure themselves are listed at the top of the columns.
Looking at the last two columns, you would conclude that SCS was significantly associated with breastfeeding intention if you had used either of the two education measures shown, but if you used at least two of the three income measures, you would conclude the opposite. Neither the log of income nor income in coretiles were associated with breastfeeding intention, and using income as a percent of the poverty level, you might have missed the relationship with absolute poverty if you had used larger income categories. Similarly, everything varied for the race ethnicity associations. For the unadjusted associations between racial ethnic group and each dependent variable, we found that they varied by the racial ethnic group and by the dependent variable. In the adjusted models, we found that the associations varied by the SCS measure we use, again both by dimension and specification and again, the magnitude and significance of the associations often varied. So here’s another example.
These are the associations between race or ethnic group and delayed or no prenatal care. Each column gives the odds ratios for racial ethnic disparities compared to European-Americans. These were obtained from series of models including different SCS variables as co-variants. So column one is adjusted only for age and parity. You would reach different conclusions about disparities for Asian Pacific Islanders and Latinas compared to European-Americans based on which income specification you chose. So in summary, for the dimensions, we found that the correlations between measures of income and education were modest overall and low among of women of color, particularly the Latinas. Thus, although education itself is in a very important socio economic factor, it is not an acceptable proxy for income, at least among child-bearing women and particularly among ethnically diverse women because it would misclassify women of color to a greater extent than European-American women.
Summarizing what we found about different specifications of income and education, different income specifications were strongly correlated with each other overall and across racial ethnic group. So it may not always be necessary to use multiple different specifications. It is also reassuring that maternal education for birth certificates and from the survey were strongly correlated, suggesting that relying on education from the birth certificate may work very well. The findings overall show that one could reach different conclusions about the roles of both race ethnicity and socio economic status, depending on the SCS measures, health indicator and groups being examined. Both the dimension of SCS and how it is specified could matter. So what are some practical implications of this work?
Researchers studying racial ethnic or socio economic disparities or using race ethnicity or SCS as co-various in other relationships should try different SCS measures overall and in racial ethnic subgroups and see how the results might change. We should study and acknowledge how results vary by outcome, SCS measure and social group. We should acknowledge the limitations of the SCS measures we use, both in terms of the dimensions covered, for example, we didn’t examine occupation or accumulated wealth or neighborhood environments or any other environmental factors. We didn’t look at early life SCS factors in relation to adult outcome. So those are the sort of limitations that we need to start including in our inner papers, and then also in terms of the detail specifications used. So if you only have poor versus not poor, acknowledge the limitation of that. So you’re not going to be looking at the full range of income. You’re lumping all the non-poor together and it sort of ignores the socio economic gradient. Furthermore, the choice of socio economic measure should not be a knee jerk or one size fits all reflex. SCS measures should be chosen based on the considering the potential causal pathways through which socio economic factors may operate for a given health indicator and population.
So finally, researchers must discuss whether SCS was measured adequately for the health indicators and groups being examined and studies of either socio economic or racial ethnic disparities. For any given outcome, it is impossible to measure all SCS factors that could be relevant to all possible pathways. Therefore, one can claim that a racial ethnic disparity is independent of SCS or that SCS does not influence a given health indicator using SCS in a global sense. So lastly, the concept I’m presenting aren’t new at all. It’s in the social epi literature, but our contribution is to illustrate these concepts concretely using a large population base and diverse sample and to make practical recommendations for use by the general community of health researchers. Thank you for your time and attention.