METHODS

Data for this case-control study were drawn from a cohort of all live births at Moffit Hospital, University of California at San Francisco (UCSF) between 1980 and 1990 (17 146 deliveries). Human subjects approval was granted by the University of California at Berkeley and the USCF. We selected all African American and white singleton preterm deliveries as cases and two random selections of singleton term deliveries as controls, stratified into African American and white infants. We excluded two groups from the analysis: (1) women with non-singleton pregnancies, infants with congenital anomalies, and induced deliveries as the aetiology of preterm deliveries in these situations differs substantially from spontaneous preterm singleton deliveries,^4,38,39 and (2) women who were transported for special care from other hospitals or regions because they did not represent the population base for this study.

The eligible analytical cohort included 417 African American and 1244 white women. We used standard methods to abstract data from the medical records as previously described.⁶ Each woman’s address was linked to census tract data from the 1980 and 1990 US census through geocoding techniques. Census tracts contain about 4000 people and were originally defined to encompass neighbourhood areas.⁴⁰

We created 12 census tract level neighbourhood variables from the 1980 and 1990 US census data. These variables, detailed in table 1, were selected to characterise the neighbourhood social environment by aggregating characteristics that have been used to examine individual level SES, as well as those that might reflect racial segregation. We examined the change in variables between 1980–1990 to identify neighbourhoods in flux and assess trends in neighbourhoods.

The outcome of interest in this study, preterm delivery, was defined as delivery before 37 completed weeks of gestation.⁴¹ The main individual level determinant of interest, cigarette smoking, was measured as cigarettes per day, self reported at the first prenatal care visit. Individual level covariates of the effect of cigarette smoking on preterm delivery considered in the analysis, included insurance coverage as Medi-Cal (California public insurance) compared with other (private insurance or payment by patient), maternal age at delivery in years, maternal height in centimetres, pre-pregnancy weight in kilograms, and parity.

All analyses were conducted separately for African American and white women. In the neighbourhood level analysis we used regression models with generalised estimating equations to generate parameter estimates accounting for both the individual and neighbourhood levels of the data.⁴² We modelled the risk of each neighbourhood variable alone on preterm delivery, controlling for Medi-Cal insurance coverage to adjust for individual level socioeconomic status. Neighbourhood variables were included with squared terms to accommodate the non-linear functional form of their relations with preterm delivery in both descriptive and multivariable analysis. The neighbourhood level variables with 95% confidence intervals on the combined main effect and squared terms that did not overlap the null were presented. The neighbourhood variables included in the initial models were included together in one model for African American women and one for white women.

We used logistic regression for the analysis of the effect of smoking on preterm delivery, including cigarette smoking and potential confounders. Cigarette smoking was modelled as cigarettes per day in all models.

After estimating the effect of neighbourhood and smoking independently on preterm delivery we added cigarettes per day to each of the neighbourhood models to determine whether this would change or remove the effect of the area level variables. We examined all potential cross level interactions of cigarettes per day and Medi-Cal insurance coverage with each neighbourhood variable in the final models for African American and white women. Plots of the odds ratios corresponding to values of the neighbourhood variables over the range of the data were constructed; these plots illustrate the relations between the neighbourhood variables and preterm delivery given the complexity of interpreting squared area level terms and interaction terms from parameter estimates alone.

RESULTS

The bivariate relations between demographic characteristics and preterm delivery for African American and white women are presented in table 2. Among African American and white women, the preterm cases had indicators of lower socioeconomic status and smoked more cigarettes per day, compared with controls. The African American women delivering preterm also had slightly lower non-pregnant weight, and slightly higher parity.

The first contextual model for the African American women contained three neighbourhood characteristics: median tract income in 1990, proportion of unemployed men in the tract in 1990, and the change in the proportion of African American residents between 1980 and 1990 (table 3). These neighbourhood characteristics had significant squared terms, indicating non-linear U shaped relations between the variables and preterm delivery. The first contextual model for white women had only one neighbourhood variable: the change in the proportion of unemployed men in the tract between 1980 and 1990 (table 3). This neighbourhood characteristic had a significant squared term, indicating a U shaped relation between the change in unemployment and the risk of preterm delivery. These associations have been demonstrated in another publication by the authors.⁶

We examined the risk of preterm delivery attributable to smoking in the single level multivariable logistic regression analysis, controlling for Medi-Cal insurance, parity, maternal height, and non-pregnant weight. The model for African American women demonstrated that smoking is a risk factor for preterm delivery. The odds ratio for preterm delivery for every 10 cigarettes smoked per day was 1.77, 95% CI (1.12 to 2.79) (table 4). Smoking also increased the risk of preterm delivery among white women. The odds ratio for preterm delivery for every 10 cigarettes smoked per day was 1.25, 95% CI (1.01 to 1.55) (table 4).

Examining the combined effect of neighbourhood and smoking among African American women, adding cigarettes per day to the neighbourhood model only slightly changed the parameter estimates for median income, proportion unemployed, and change in the proportion of African American residents (table 3). There was no evidence of interactions between cigarette smoking and these neighbourhood variables. However, there was significant interaction between Medi-Cal insurance and proportion of unemployed male residents in 1990. Among white women, the association between change in unemployment and preterm delivery remained significant when adjusting for cigarette smoking. The parameter estimates for change in unemployment were slightly greater in this model (table 3). There was no evidence of cross level interaction between cigarette smoking or Medi-Cal status and change in unemployment for white women. Odds ratios corresponding to values of each neighbourhood variable for selected percentiles are presented in graphs, along with 95% confidence intervals (fig 1A–E). The risk of preterm delivery for African American women was increased in neighbourhoods with low median income (fig 1A). Among African American women who were not covered by Medi-Cal, the risk of preterm delivery was low in neighbourhoods with low unemployment and high in neighbourhoods with greater proportions of unemployment (fig 1B). However, African American women who were covered by Medi-Cal were at high risk of preterm delivery at low levels of unemployment, with some reduction in risk at higher levels of unemployment (fig 1C). In neighbourhoods with increasing proportions of African American residents, African American women were at increased risk of preterm delivery, and the relation was reversed in neighbourhoods with decreasing proportions of African American residents (fig 1D). White women were at increased risk of preterm delivery in neighbourhoods with both increasing and decreasing levels of unemployment (fig 1E).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Odds ratios and 95% confidence intervals for preterm delivery by census tract characteristics among African American and white women, UCSF cohort 1980–1990. (A) African American women: ORs for preterm delivery by median tract income. (B) African American women without Medi-Cal insurance: ORs for preterm delivery by proportion of unemployed men. (C) African American women with Medi-Cal insurance: ORs for preterm delivery by proportion of unemployed men. (D) African American women: ORs for preterm delivery by change in the proportion of African American residents. (E) White women: ORs for preterm delivery by change in the proportion of unemployed men.

DISCUSSION

Our analyses suggest that neighbourhood socioeconomic characteristics may have an impact on preterm delivery for African American and white women that is largely independent of cigarette smoking and may be modified by individual socioeconomic characteristics. Smoking was a risk factor for preterm delivery in adjusted models and had a stronger impact on the likelihood of preterm delivery in African American women than in white women. Neighbourhood socioeconomic factors did not interact with cigarette smoking. Among African American women, the effect of neighbourhood level unemployment was different depending on whether the women had Medi-Cal insurance coverage or not; women with Medi-Cal coverage had a high risk of preterm delivery with some reduction in risk at higher levels of unemployment, while women without Medi-Cal coverage had an increasing likelihood of preterm delivery at higher levels of neighbourhood unemployment.

In our analysis of cigarette smoking as an individual level risk factor, smoking was associated with preterm delivery among both African American and white women. The association between smoking and preterm delivery was stronger among African American women than it was among white women, which is consistent with previous studies.^4,25,43

Recent research indicates that smoking may interact biologically in women with recurring bacterial vaginosis (BV), increasing the likelihood of an episode of BV, which is a risk factor for preterm delivery.^44–50 Recurring BV is also far more common among African Americans.^45,49–51 Although the reasons for more prevalent BV among African Americans are unknown, it may explain part of the differential impact of smoking on preterm delivery in this racial/ethnic group. This finding merits further exploration with data on smoking, BV and preterm delivery.

Smoking could have explained some of the association between neighbourhood factors and preterm delivery as maternal smoking may be influenced by normative behaviours and stresses in neighbourhoods.³² Research on the effects of stress and smoking on pregnant women offers a biological explanation for the increased risk of preterm delivery among women who both smoke and live in poor neighbourhoods (possibly through the synergistic effects of smoking and stress on raising catecholamine levels).^33–37 However, our results suggest that the effects of neighbourhood characteristics were not affected by cigarette smoking.

Individual level Medi-Cal insurance coverage modified the impact of neighbourhood unemployment on preterm delivery. For African American women who were not covered by Medi-Cal insurance and had more economic resources the risk of preterm delivery was low when neighbourhood unemployment was low and increased when unemployment was higher. This is plausible because women who may not have experienced major individual level socioeconomic hardship may still have experienced a negative impact on their birth outcomes if the neighbourhood was in poor economic condition.¹⁷ However, the relation was quite different for women who were covered by Medi-Cal insurance and had fewer economic resources. These women were already at high risk of preterm delivery with low unemployment in the neighbourhood. The risk of preterm delivery among women with Medi-Cal insurance was somewhat lower in neighbourhoods with higher unemployment. These women may have experienced the highest risk in areas with the lowest unemployment because they were impoverished in relation to the others in that neighbourhood.¹⁴ This finding is consistent with O’Campo’s study of low birth weight, which found that the increased risk of low birth weight for women on Medicaid was decreased in neighbourhoods with low per capita income compared with neighbourhoods with high per capita income.¹⁰

There are several limitations to this study. The findings of a hospital based study are not easily generalised to broader populations. If white women from more prosperous areas who thought they were at high risk of preterm delivery chose Moffit Hospital because of its tertiary care capacity, the increased risk of preterm delivery in neighbourhoods with decreasing unemployment (a potential indicator of improving social conditions) could be an artefact. Census data are only available for the beginning of each decade, so it is difficult to know how well this information represents the conditions of the tracts throughout the decade. In addition, census data do not capture all aspects of neighbourhoods that may affect health,⁵² and do not include information about the other communities (for example, workplace, school) to which these women may belong. Neighbourhood factors have not been studied sufficiently to determine which factors best characterise social conditions; thus, the neighbourhood characteristics examined in this study are necessarily exploratory. Our measure of smoking, self reported cigarettes per day at the first prenatal care visit, is limited because some women may not accurately report smoking. It is most probable that some smokers are misclassified as non-smokers and not vice versa creating a conservative estimate of the effect of smoking on preterm delivery. Smoking at the beginning of pregnancy may not represent smoking throughout pregnancy, particularly in the third trimester when smoking has the greatest impact on preterm delivery.^26,53 Additionally, the small sample size of the African American subgroup and the modelling of interaction led to imprecise estimates of the effect of neighbourhood unemployment.

Notwithstanding these limitations, we demonstrated associations between neighbourhood socioeconomic characteristics and preterm delivery for both African American and white women that are independent of cigarette smoking and modified by individual socioeconomic characteristics. In addition, we found that cigarette smoking increases the risk of preterm delivery particularly among African American women. A better understanding of the differential influence of factors such as cigarette smoking on African American women may additionally help elucidate racial disparities in preterm delivery. Examining neighbourhood and individual socioeconomic factors in combination with behavioural and biological characteristics may lead to a better understanding of the complex causes of preterm delivery.