1In France, perinatal health has been a public health priority since the early 1970s, with a key objective of reducing social inequalities in healthcare access and health outcomes. It has been widely shown that levels of perinatal mortality and morbidity are influenced by individual social characteristics (Kaminski et al., 2000). Among married couples, for example, the perinatal death rate of babies born to parents in managerial and higher-level occupations was 7.1‰ compared with 11.7‰ for unqualified manual workers in the late 1990s (Dinh, 1998). It is difficult to monitor perinatal health inequalities in France however. To study rare events, such as perinatal mortality and very preterm births, data on large numbers of births are required. But in France, there are no information systems which include data on both the socioeconomic characteristics of the parents and the indicators of mortality and morbidity for all births (Bréart et al., 2003).
2Another way to measure social inequalities is to use socioeconomic data on neighbourhoods of residence. Studies in France have revealed geo-graphical disparities in overall mortality (Rican, Salem and Jougla, 1999; Rican, Jougla and Salem, 2003) and in infant mortality (Barbieri, 1998) on the scale of the region and the département, but none have been carried out at local level. Small-area-based studies are common in Anglo-Saxon countries however. They show that the risks of preterm birth and low birthweight are higher in neighbourhoods with multiple socioeconomic disadvantages (high unemployment, low educational level, etc.) than in neighbourhoods qualified as affluent (Guildea et al., 2001; Krieger et al., 2003a).
3For the 1999 census in France, INSEE created a new infra-communal census zone called the IRIS 2000. This analysis explores social inequalities in perinatal health using socioeconomic data on the neighbourhoods of pregnant women living in different IRIS 2000 zones. We use as an example the risk of very preterm birth, a serious complication of the perinatal period. It is defined as birth before 33 weeks of gestation or before the eighth month of pregnancy. Very preterm babies incur a high risk of neonatal death (around 15%) and of severe short- and long-term morbidity (Larroque et al., 2004). Our aim is both to show that analysis of the demographic and socioeconomic characteristics of neighbourhoods of residence provides a means to identify social inequalities in health, and to present a methodology that could be used to track these inequalities using census-based data combined with data on individual patients.
I – Theoretical approach
4We used an approach based on living conditions to analyse social inequalities. This approach was developed by Townsend in 1987, and is founded on a long tradition of analyses of social inequalities in health (Townsend, 1987). It aims to identify various types of difficulty, disadvantage and deprivation associated with living conditions which are summed up in a deprivation score. In 1991, Carstairs and Morris devised a deprivation index based on a similar approach to that of Townsend. These two indices are the most widely used in British and American studies of social inequalities in health, though other methods also exist (Jarman, 1983; Krieger et al., 2003b). The deprivation scores are determined on the basis of variables relating to material deprivation (average income, personal assets, etc.) and social deprivation (single-parent families, persons living alone, etc.)
5The neighbourhoods are classified according to their score, which measures aggregate deprivation: the lower the score, the more socioeconomically advantaged the neighbourhood; the higher the score, the greater the level of deprivation (more unemployment, more manual workers, more people with a low educational level, etc.). The deprivation score is a continuous variable (Carstairs, 1995) which provides an aggregate measure of the socioeconomic levels of neighbourhoods using data from the population census.
6The social indicators constructed from neighbourhood data provide a means to determine inequalities at this collective level. The characteristics of an area being the sum of the individual characteristics that influence health, the observed inequalities may reflect the effect of individual characteristics on health. This approach is used to reveal the existence of inequalities when no social data on individuals is available. However, studies have shown that the fact of living in a poor neighbourhood may have a negative impact on health independently of individual characteristics (O’Campo, 2003). This effect is linked to healthcare accessibility, insecurity and exposure to other environmental factors. In the absence of data on individuals, analysis of inequalities using collective data does not enable us to distinguish between individual effects and aggregate or contextual effects (Geronimus and Bound, 1998; Courgeau and Baccaïni, 1997).
II – Sources and methods
7Our data sources comprise both a survey of very preterm births – the 1997 “Épipage” epidemiological study of very preterm babies (Étude épidémiologique sur les petits âges gestationnels) – and the 1999 general population census. The Épipage study enabled us to characterize the neighbourhoods of residence of very preterm infants via the geocoded addresses of their mothers, while the population census data served to characterize the neighbourhoods on the basis of the IRIS-2000 infra-communal zones and to study the neighbourhoods of residence of newborns in the general population.
Geographical scale
8The geographical scale is that of the IRIS-2000 zones: “2000” corresponds to the year of their creation and to their theoretically minimal population size. The IRIS zones are homogeneous in terms of population size, types of dwelling and urban development, and they follow the historical neighbourhood boundaries (Bolusset, 2000). Using the IRIS zones, it is thus possible to conduct very detailed studies at communal level: “(…) the smaller the reference area, the more homogeneous the population, the lower the risk of classification error and the larger the number of major health inequalities detected” (Pampalon and Raymond, 2000). Our study concerns Paris and its inner suburbs [1], divided into 2,715 IRIS zones. The area counts more than six million inhabitants, with an average of almost 2,300 inhabitants per IRIS zone (Table 1). The population size of each zone is very variable, but the number of inhabitants is generally somewhere around 2,000.
Presentation of the IRIS zones of Paris and the inner suburbs
Presentation of the IRIS zones of Paris and the inner suburbs
9Most of the IRIS zones cover a small surface area (0.3 sq.km on average) and have a high population density (20,000 inhabitants per sq.km). Some – qualified as “miscellaneous” – have very few inhabitants and cover large areas because they include particular features such as woodland, parks, railway stations, cemeteries, etc.
Constructing the deprivation score
10To determine the socioeconomic level of the IRIS zones, we chose to measure living conditions in terms of aggregate deprivation, as is often the case in the literature. For the deprivation score, we selected variables which best capture the various dimensions of deprivation: employment and working conditions, schooling, housing conditions, household assets and demographic characteristics. The SOC variable (occupations and socio-occupational categories) indicates social status, and the unemployment level is an indicator of economic insecurity. Educational capital is measured by level of schooling, but covers the population aged 20-40 only, since certain IRIS zones have a high proportion of elderly people, and hence a generally low educational level due to a generation effect. The number of persons per dwelling measures the degree of overcrowding. Household car ownership measures the mobility of its members. Last, we included the proportion of single-parent families whose head of household is a woman, since these families are more often in situations of severe deprivation (Algava, 2002).
11To identify the variables most appropriate to our study, we performed a principal components analysis (PCA). The first axis of the PCA identified the variables associated with the most deprived neighbourhoods. This first axis explained 42% of variance. The following variables were selected: proportions of manual workers, of persons in higher-level occupations, of unemployed persons, of persons with permanent work contracts, of persons aged 20-40 with primary-level schooling, average number of persons per room, proportion of households with no car, of single-parent families and of persons born abroad (Table 2).
Census-based variables used to study socioeconomic inequalities
Census-based variables used to study socioeconomic inequalities
12The selected variables are strongly correlated, justifying the use of a composite index. The correlation coefficients range from 0.7 to 0.8 between variables relating to schooling level and employment, and from 0.4 to 0.7 for the number of persons per room, the percentage of persons born abroad and of single-parent families. The proportion of households with no car is not systematically correlated with the other variables however (the coefficient ranges from a minimum of 0.01 with the percentage of manual workers to 0.68 with the proportion of persons without a permanent work contract).
13After identifying the score components via the PCA, we defined a deprivation score which represents the sum of the values of each variable standardized as a function of its mean and its standard deviation (Carstairs, 2000).
The study population: very preterm babies
14The Épipage study was conducted in 1997 on all very preterm babies (born between 22 and 32 weeks’ gestation) in nine regions of France (Larroque et al., 2004). Its main purpose was to monitor the progress of very preterm babies through regular follow-up to the age of eight. For our analysis, we selected the women who lived and gave birth in Paris and its inner suburbs. Data was collected from February to July 1997 in this region. The study population is limited to singleton births before 33 weeks’ gestation to mothers living in Paris and its inner suburbs. We excluded multiple births because the causes of prematurity in such cases are different and there is no proven link with the parents’ socioeconomic characteristics (Rolett and Kiely, 2000). We also excluded stillbirths and neonatal deaths in the delivery room, for which the mothers’ addresses were not recorded, and infants born to mothers whose addresses were incorrect or incomplete. Out of the total number of singleton births before 33 weeks’ gestation, 12.5% were stillbirths, 3% died during delivery and 4% died in the delivery room.
15To identify the neighbourhoods of residence of the women in the Épipage database, we geocoded the address of each mother using an INSEE file listing all the addresses in each IRIS zone. The second stage involved matching the data for each mother (individual socio-demographic data, place where she gave birth, information on birth and pregnancy outcome, etc.) with the socioeconomic data of her neighbourhood of residence taken from the census.
16Altogether, out of all singleton births transferred to the neonatal ward, 90% of addresses in Paris and the inner suburbs were matched to an IRIS zone [2]. Our sample thus included 303 newborns.
17To study the link between the socioeconomic level of the neigh-bourhoods of residence and the women’s individual characteristics, we used the main individual socio-demographic variables included in the Épipage survey (country of birth, marital status, educational level, occupational status, social category, partner’s educational level and employment status). The social data were obtained from the women’s medical records and from an interview following delivery. Not all women participated in this part of the study however: only 80% of our sample of 303 women answered the questionnaire. The remaining 20% did not take part for a variety of reasons: concern about their baby’s health, personal health problem, refusal to answer the questionnaire, language problem, etc.
The control population: newborns enumerated in the 1999 census
18To constitute a control population, we used census data on children aged “0 years” enumerated in 1999, i.e. children born between 1 January and 7 March inclusive. The 1999 population census counted a total of 15,414 newborns over this period living in Paris and the inner suburbs and who thus served as our control population.
19The mean number of births per IRIS zone is 5.68 (the interquartile distribution ranges from 3 to 8). We were able to check the reliability of the census data on births by checking them against the civil records for the number of live births to women living in Paris and the inner suburbs between 1 January and 7 March 1999. The census data do not appear to be exhaustive, with a 6% shortfall in the number of births compared with the civil records (Table 3). These missing data may be due to collection problems, in Paris especially, where census coverage was least comprehensive due to difficulties in entering residential buildings (growing use of access codes and intercoms) and in contacting residents (persons living alone, atypical working hours, etc.).
Births in the Paris region according to data sources by département of residence
Births in the Paris region according to data sources by département of residence
III – Analysis strategy
20We conducted our analysis in several stages.
21We began by comparing the individual characteristics of the women followed in the Épipage survey (i.e. mothers of very preterm babies) by quartile of neighbourhood deprivation score using the Chi [2] test.
22We then compared these women’s neighbourhoods of residence with those of women in the general population (from census data) using the quartile distribution of the different socioeconomic indicators of each group. Lastly, for the deprivation score, we present the results in quartiles and in quintiles so that our findings can be compared with those of other studies. A statistical comparison of the study and control populations was performed using trend tests and odds ratios, with confidence intervals calculated from logistic regressions.
IV – Results
23Our deprivation score has a very wide distribution, with a mean of 0 (by construction) and a range from –15 to +32. Map 1 clearly highlights the major social inequalities that exist in Paris and its inner suburbs.
Deprivation score in the IRIS zones of Paris and its inner suburbs in 1999.
Deprivation score in the IRIS zones of Paris and its inner suburbs in 1999.
24The first quartile of the deprivation score represents the most affluent neighbourhoods of Paris and its inner suburbs, and the last quartile the most deprived. The spatial distribution of the deprivation score reveals high levels of deprivation in Seine-Saint-Denis, the north of Paris, the northern part of Hauts-de-Seine and the south-west of Val-de-Marne.
25A relationship exists between the individual socioeconomic characteristics of mothers of very preterm babies and the deprivation score of their neigh-bourhood of residence (Table 4): 79% of women living in the affluent neigh-bourhoods (first quartile of the deprivation score) were born in France; 60% have higher- or intermediate-level occupations and 60% have an educational level above that of the high-school diploma. Conversely, in the most disadvantaged neighbourhoods (last quartile of the score), half of the women were born abroad, fewer than 2% have higher-level occupations and only 8% an educational level above the high-school diploma. There is no significant relationship between the marital status of mothers and the neighbourhood deprivation score however. The same results are obtained for the partner’s characteristics.
Individual characteristics of mothers of very preterm babies in Paris and its inner suburbs by characteristics of their neighbourhood of residence (%)
Individual characteristics of mothers of very preterm babies in Paris and its inner suburbs by characteristics of their neighbourhood of residence (%)
26Table 5 compares the distribution of very preterm births with that of the control population by neighbourhood characteristics, presented in quartiles. The risk of premature birth varies according to the socioeconomic characteristics of the neighbourhoods, whatever the characteristic examined. However, the trend test indicates that certain variables are more closely correlated with the risk of very preterm birth, such as the proportion of manual workers and the percentage of persons aged 20-40 with only primary-level education. The number of persons per room and car ownership are less strongly associated with severe prematurity.
Distribution of total births and very preterm births in Paris and the inner suburbs by socioeconomic characteristics of neighbourhoods (%)
Distribution of total births and very preterm births in Paris and the inner suburbs by socioeconomic characteristics of neighbourhoods (%)
27The deprivation score is significantly linked to the risk of preterm birth (Table 6). We note that 31% of mothers who gave birth prematurely live in the most disadvantaged neighbourhoods (last quartile of the score), while 21% live in the most affluent neighbourhoods. This excess risk corresponds to an odds ratio of 1.5. When the deprivation score is analysed in quintiles, the difference between the two extremes is even larger: 26% of women with a very preterm baby live in the most deprived neighbourhoods versus 16% in the most affluent ones, giving an odds ratio of 1.65. By comparison with each indicator analysed separately, the deprivation score clearly demonstrates the existence of a social gradient.
Distribution of total births and very preterm births in Paris and the inner suburbs by neighbourhood deprivation score (%)
Distribution of total births and very preterm births in Paris and the inner suburbs by neighbourhood deprivation score (%)
V – Discussion
28Our findings for Paris and its inner suburbs tie in with those of numerous studies which have shown that the socioeconomic environment is linked to the risk of preterm birth: the more deprived the environment, the higher the risk. The deprivation score based on census data enabled us to identify the geographical areas with multiple socioeconomic disadvantages and to study the births in these areas, both very preterm births and all other births.
29The variables used in our deprivation score are all closely correlated, so we can only study the impact of the overall context. Though the individual variables are linked to the risk of preterm birth and the associations vary from one variable to another, it is difficult to interpret these associations and their differences. For example, we cannot separate the effect of the level of unemployment on the risk of very preterm birth from the effect of the proportion of single-parent families. In the census data analyses, each variable taken separately serves as a sort of proxy for the full set of variables that are “markers” of deprivation (Geronimus and Bound, 1998). This provides justification for constructing a composite indicator to measure these inequalities.
30Our study is a case-control study and the regression models used for the analysis generate odds ratios and not relative risks. However, as very preterm birth is a rare event, the odds ratio is a good approximation of relative risk. Based on the assumption that the proportion of very preterm births is 1%, the odds ratios obtained imply a proportion of 0.8% very preterm births in the most advantaged quintile and 1.3% in the most deprived quintile.
31Our findings thus confirm those of other perinatal health studies based on a deprivation score for the neighbourhood of residence. In the United Kingdom, two studies of low birthweight babies have been conducted, one based on the Townsend score and the other on the Carstairs score. The relative risks between the fifth and first quintiles of these scores are 1.59 and 1.53 respectively (Pattenden et al., 1999; Spencer et al., 1999). A study of perinatal and infant mortality produced an odds ratio of 1.53 between the first and fifth quintiles of the Carstairs score (Guildea et al., 2001). An American study found a larger difference in the risk of low birthweight – around 2 – using the Townsend score (Krieger et al., 2003a). Moreover, the odds ratios are also close to those found in studies based on individual characteristics – around 1.5 between the groups with the highest and lowest socioeconomic indicators (Ancel et al., 1999; Barbieri, 1998; Kaminski et al., 2000; Leon et al., 1992).
32But the limits of this approach must be taken into account when interpreting the results obtained. The first limit results from the missing cases in both the control and the study populations. For the controls, we have no information on the populations not included in the census. Are they the most disadvantaged individuals, with no social security coverage and whose access to antenatal care is limited or non-existent? The study population is not exhaustive either, since not all very preterm infants are included. Certain addresses were absent or incomplete so could not be attached to an IRIS zone. For certain women, the absence of an address indicates a highly deprived socioeconomic situation (women who are homeless or living in hostels). Hence, if the women with no address are also the most deprived, we are liable to under-estimate the effect of social characteristics. In addition, we have addresses for infants transferred to neonatal units, but not for stillbirths or infants who died in the delivery room. These infants represent around 20% of very preterm births. If there is inequality of access to care at the time of birth, then the social gradient may be even greater for these cases. Nevertheless, the population covered by our study – very preterm babies admitted to a neonatal unit – is interesting in itself. These infants, of whom 85% leave hospital alive (Larroque et al., 2004), have a high risk of developing sequelae linked to severe prematurity. The fact that they more often live in disadvantaged neighbourhoods is an important factor for assessing health inequalities among children and for healthcare resource planning.
33Another limit is linked to the timing of data collection: the control population consists of infants born in 1999 and the study population of infants born in 1997. When census data are used, the second data source often does not cover the same period. However, a study in the United States has shown that the association between health indicators and socioeconomic data does not vary significantly by date of census data used: the authors conclude that data on neighbourhoods, even if measured at ten-year intervals, are closely correlated (Geronimus and Bound, 1998). With the new census method introduced in France from 2004, data will become available at shorter intervals.
34Our methodology has only been tested for urban areas. The IRIS-2000 zones are only obligatory for communes of more than 10,000 inhabitants and recommended for those of at least 5,000 inhabitants. In rural areas, the “commune” is the smallest census zone. We still need to determine whether using census data at commune level is an effective means to reveal social inequalities.
35Lastly, this methodology cannot be used to analyse explanatory mechanisms since the census data are analysed at community level only. It is thus not possible to obtain results adjusted for other confounding factors such as mother’s age or parity. High maternal age is a risk factor for premature birth and a part of the difference between the quartiles may be due to an uneven age distribution. However, since women of high socioeconomic status often give birth at older ages, the observed effect may also be reinforced if age is taken into account.
36More generally, we cannot distinguish between the different possible explanations for the existence of social inequalities. The excess very preterm births in deprived neighbourhoods could be explained by the effect of individual characteristics of the population or by an effect linked to the local environment, independently of the inhabitants’ characteristics. To estimate the specific effect of the environment, it must be possible to adjust the model for the inhabitants’ characteristics, such as their educational level or their occupational status, using multilevel analysis methods. Numerous studies have revealed an association between the neighbourhood of residence and the risk of very preterm birth or low birthweight (Ahern et al., 2003; Boardman, 2004; Buka et al., 2003; Matteson, Burr and Marshall, 1998; O’Campo et al., 1997). The main explanatory mechanisms suggested by these authors are the stress of living in a deprived neighbourhoods or in a community with high unemployment, the absence of support from family or community networks, harmful environmental exposure, and the absence of public services and of high-quality health and social services.
37Despite these limitations, this methodology has several key advantages. As only the addresses of persons with the health problems under study are required, the methodology can be used to monitor social health inequalities on the basis of registers of diseases or health problems, such as registers of cancer patients or of persons with disabilities, and of specific epidemiological surveys. Patients’ addresses are often available in medical files, whereas information on their socioeconomic status is not. The method could also be applied to other health problems for which a control population, representing the population at risk of developing the disease in terms of age or sex, could be established using census data. Lastly, the use of these data for multilevel analyses, enabling more exhaustive analysis of explanatory mechanisms, is also possible if individual level data are available for the control population. In conclusion, this methodology would provide a means to take the social dimension into account on a regular basis in health studies, and would thus provide the information needed to implement and assess public policies and to estimate community healthcare needs.
Notes
-
[*]
Recherches épidémiologiques en santé périnatale et santé des femmes, Inserm U149, Paris.
-
[**]
Centre de recherche populations et sociétés (CERPOS), Université Paris X, Nanterre.
-
[***]
Épipage groupe Paris – Petite couronne : Pierre-Yves Ancel, Gérard Bréart, Michel Dehan, Monique Kaminski, Christiane du Mazaubrun, Michel Vodovar, Marcel Voyer, Véronique Zupan-Simunek.
Translated by Catriona Dutreuilh -
[1]
The inner suburbs comprise the first ring of départements surrounding Paris intra muros, i.e. Seine-Saint-Denis, Val-de-Marne and Hauts-de-Seine.
-
[2]
10% of addresses could not be matched to an IRIS zone. In 70% of cases, the address was missing because the mother refused to take part in the survey or had no permanent address. Only 12 addresses could not be matched to an IRIS zone because they were incomplete.
