METHODS

The study took place in the Rhône-Alpes region, which comprises eight administrative departments and a total of 2879 communities (official municipalities, of any size). We selected only the 2872 communities with fewer than 50 000 inhabitants for the present study. Since 1978, this region has been covered by a population based birth defects registry (France Central-East registry) that is part of an international network of such registries.¹⁰ Malformations are ascertained up to the age of 1 year by reporting from maternity wards in the region, combined with an active search of the medical records of the major local obstetric and paediatric units. Stillbirths and medical terminations of pregnancy induced after prenatal detection of a major malformation have been registered since 1985. The Central-East registry collects approximately 1100 cases per year in the Rhône-Alpes region. Regular verifications are performed, and when local deficits are suspected, active data collection is instituted. Examinations of this registry and those in Sweden and California show good comparability. Ad hoc validation studies for specific anomalies (such as facial clefts or oesophageal atresia) have assessed the sensitivity of the registration as satisfactory overall (80%), although higher for malformations requiring surgery (90%).

Grouping of malformations

Malformations were first divided into four categories: minor (possible variations from normal), chromosomal, monogenic, and other major anomalies (unknown or multifactorial origin). We further subdivided the “other major anomalies” into 23 different subgroups. These subgroups were all pathogenetically homogeneous with at least 50 cases. Table 3 shows the malformations groups we chose; they account for 70% of all infants with an isolated malformation.

For each year in the 10 year study period (1988–97), census and registry data enabled us to compute the rates of congenital anomalies, stratified by type of anomaly, sex, maternal age, and mother’s community of residence at time of birth (or medical termination). The latter was the finest level of detail routinely recorded.

Inventory of MSWIs

The national agency charged with providing technological advice for waste management (Agence de l’Environnement et de la Maitrise de l’Energie, ADEME) furnished us with a list of the 70 incinerators that operated in the region for at least one year during the study period. Operators and local public authorities were asked for technical descriptions of each plant. Because very few measurements of total dusts, dioxins, and metals were available, we used expert assessments to construct a semi-quantitative estimate of the emissions for each. The team in charge of MSWI classification invited six experts to participate, representing operators (three experts), public authorities (two), and research institutions (one). First, 22 MSWI categories were defined by a broad set of variables known to influence dioxin emission. These included capacity (four categories: <1 ton/h, 1–3 t/h, 3–6 t/h, >6 t/h), type of functioning (continuous, discontinuous), dust control (present, absent), fume treatment (present, absent), and year operations began. To apply an iterative process towards consensus (the Delphi method),¹¹ the experts were each independently asked to assign a score between 0 and 9 to each category and to discuss discrepancies in scoring (outliers) until they reached a consensus for each category and each pollutant. Each of the 70 MSWIs was then rated according to its category (table 1). For the larger incinerators, scores were adjusted around the mean score for the category, by taking its precise annual capacity into account.

Dispersion models

A Gaussian plume model (POLAIR) was used to compute “immission” scores (immission is the amount of pollutant reaching a particular location as a result of—and in contrast to—the emission coming out the chimney; immission scores serve as proxies for annual ambient air concentrations of chemicals attributable to the MSWI at a given location) for each category of pollutants within 10 km of the plant; unit cells were 1 km²; parameters considered were: the semi-quantitative score obtained above from the experts (supposed to be proportional to the quantities emitted), chimney height, and local weather data.¹² The latter came from the national meteorological agency (Météo France) for all but 14 MSWIs, for which statistics on local winds were judged unreliable. The missing information for these 14 was replaced by the hypothesis of equal distribution of wind directions in the area. Pollutant concentrations attributable to an MSWI were considered negligible (dioxin concentration below 0.02 pg/m³ in the air) outside a 10 km zone. As a consequence, any immission score estimated from the model and falling below the maximum obtained at 10 km from all MSWIs in the region was considered to be zero (unexposed zone).

Exposure of communities

The immission score assigned to each of the 2872 communities was that estimated at the geographic point of highest population density. We calculated two different exposure indexes for each community, according to different hypotheses about the mode of exposure: (1) the immission score alone represented exposure from inhalation only; (2) this score multiplied by the number of years the plant had operated took into account the total amount of pollutants accumulated since the start of the activity. We present the results according to the latter index, which showed a dose-response relation with disease risk. When a community was within 10 km of several MSWIs, the exposure index was estimated as the sum of the estimated indexes of each. Among exposed communities, only births occurring during the years following the start of the MSWI were considered in the analysis.

Further information about each town, such as population density (inhabitants per square kilometre) and average yearly family income (€/year), came from the 1990 and 1999 national censuses.

According to ADEME, in 1997 the principal sources of dioxins in the atmosphere in France were metal industries (between 270 and 2300 g TEQ/year), followed by MSWIs (400 g TEQ/year), high temperature industrial processes (cement production for instance, 23 g TEQ/year), road traffic (1–5 g TEQ per year), and incineration of industrial waste (2 g TEQ/year). Identification of the presence of these sources for each community was extremely tedious, which is why it was done only among exposed communities. The presence of metal industries was defined by the percentage of the working population employed in smelting or metallurgy plants according to the 1990 census (>5%). Traffic density (vehicles/day) within 500 m of the city centre was based on the 1999 statistics (or those for the last year the incinerator operated) from the relevant technical centres. We also identified three industrial waste incinerators among the exposed communities.

Statistical analysis

Analysis focused mainly on the dioxin concentration estimates, since the other pollutant emissions were highly correlated with them (r_Spearman (dioxins, metals)  =  0.64). There were 194 communities considered exposed and 2678 unexposed. Three classes of increasing dioxin exposure were defined by tertiles of births (with the constraint that all births in the same community belonged to the same class). We fitted a Poisson regression model to the rate of congenital anomalies, assuming that the count of observed events (O) followed a Poisson distribution (a standard hypothesis for counting rare events).

Two different models used two types of denominators for the rate. We began by fitting the model with an offset as the logarithm of the number of births (N):

X_{i k} were covariates, and

was interpreted as an estimated rate. The relative risk (RR1) associated with X_k was computed as the exponential of the coefficient α_k.

We then chose the offset as the logarithm of the expected number of events E, standardised for maternal age, year of birth, and department of birth. This model is written as:

Again Z_{j k} were covariates and

was therefore interpreted as an estimated SMR; accordingly, the relative risk RR2 associated with Z_k was computed as the exponential of the coefficient β_k. Wald confidence intervals were estimated based on the asymptotic normality property.

The first model allowed us to compare the exposed and unexposed populations by a method equivalent to indirect standardisation, with adjustments for maternal age (<20, 20–24, 25–29, 30–34, 35–39, 40–44, ⩾45), year of birth, and department of birth, and with additional adjustments for population density (four classes: <100 inhabitants/km², 100–500, 500–1000, >1000) and family income (four classes (in €/year): 11 400, 11 400–13 700, 13 700–16 000, >16 000) since both affect access to medical services and quality of the housing environment. Categories of maternal age and year of birth were reduced to three classes each in models involving subgroups of major congenital anomalies. Because the year of birth is included in the covariates, the data tables are constructed by year, which may lead to dependence between the observations.

The second model, by definition, applies only to the exposed population, since E is computed from the unexposed reference population. It adjusts for population density, family income, and the supplementary information about road traffic, which was available only for the exposed population. Because additional adjustment for the presence of other environmental sources of dioxin exposure did not modify the risk estimates, and since some values were missing, this variable was not included in the models. The three exposure levels made it possible to estimate the dose-response relation between exposure to MSWI emissions and the risk of congenital malformations. To prevent overdispersion, which often occurs in Poisson models, we decided to correct by adjusting all statistics for the dispersion parameter, computed as the deviance divided by the number of degrees of freedom. These Poisson models were computed with the SAS GENMOD procedure. Linear trend tests were computed for categorical exposure variables.¹³

Despite efforts at exhaustiveness, registry coverage was less complete for pregnancy terminations than for births. In a complementary analysis, an additional parameter indicative of the quality of the ascertainment—that is, the percentage of medical terminations of pregnancy among reported birth defects, was included in the model.

RESULTS

Communities considered to be exposed to MSWI emissions were on average larger, more densely populated, and wealthier than unexposed communities (table 2). Almost half the births (47.6%) in the exposed communities occurred in those that were also traversed by heavy traffic (more than 50 000 vehicles per day). Seven exposed communities, accounting for fewer than 10% of the births, hosted metallurgy plants or industrial waste incinerators, but this parameter was not included in further analysis. We observed significant correlations between community size, population density, income, and traffic density, but in the exposed communities, none of these variables was significantly correlated with the quantitative exposure index (all correlation coefficients were below 0.07).

Overall, the rate of congenital anomalies was higher in the exposed than the unexposed communities (RR = 1.09) (table 3). This excess was explained mainly by anomalies of unknown or multifactorial aetiology (other major) (RR = 1.12, p<0.05). Additional adjustment for population density and average family income reduced the estimates, and the relative risk for other major anomalies was no longer statistically significant (RR = 1.07, 95% CI 0.98 to 1.16). The relative risks of specific major anomaly subgroups, including facial clefts (RR = 1.30, 95% CI 1.06 to 1.59) and renal dysplasia (RR = 1.55, 95% CI 1.10 to 2.20), were higher than one.

For no category of anomalies did the risk in the exposed communities increase significantly with the exposure index (table 4). An increase in the risk of other major malformations was observed in relation to heavy traffic. The risk of chromosomal anomalies was strongly associated with population density.

Table 5 presents the dose-response trends for the risk of some subgroups of major anomalies—those for which we had sufficient numbers (>30)—in the exposed communities. A trend towards increasing risk with increasing exposure was observed for obstructive uropathies (anomalies of renal pelvis, ureter, urinary bladder, and urethra) (p = 0.07). Dose-response associations with road traffic were observed for the risks of cardiac anomalies (conotruncal or others), obstructive uropathies, and skin anomalies (30% were large angiomas or naevi, and the rest, generalised skin anomalies such as incontinentia pigmenti, bullous epidermolysis, or ichthyoses).

DISCUSSION

Our study of malformation rates around MSWIs showed that overall congenital anomalies were not significantly more frequent in exposed than non-exposed communities. An excess number of some specific groups of major anomalies (facial clefts and renal dysplasia) was observed. In exposed communities, a dose-response relation linked obstructive uropathies to MSWI exposure. Risks of cardiac anomalies, obstructive uropathies, and skin anomalies increased linearly with road traffic density.

A major drawback of this study is the extent of exposure misclassification. Emissions and soil measurements were not available for all 70 MSWIs in the region. A rating was attributed to each MSWI by the experts, based on their overall knowledge, on emissions measurements furnished by operators or public authorities, and on the sparse data available for soil measurements and concentrations in cows’ milk. In a clear improvement over studies using circles around point sources of pollution,¹⁴ a dispersion model took into account local meteorological data to define the exposed area. Due to the level of administrative units used, however, we had to allocate the same exposure to all births occurring in a single community: if the town centre (defined as the point of highest population density) was within the plume of an incinerator, all births there were considered exposed. Depending on the territory covered by a community, this inevitably classifies as “exposed” births to mothers living outside the plume or assigns incorrect exposure levels. Conversely, some births just outside town limits may wrongly be considered unexposed. In addition, this ecological classification obviously does not take into account individual characteristics (for example, time-activity patterns and the variations of exposure in time relative to pregnancy). Diet is a major contributor to exposure in exposed areas, but distribution of food production outside polluted areas may help dilute the contrast between exposed and non-exposed areas. Migration of a small proportion of mothers (<6%) between the beginning of pregnancy and birth is an additional source of misclassification. The same type of misclassification applies to road traffic.

Trying to infer the direction of the ensuing bias is not easy. In larger cities, where the incinerator plume is less likely to cover the entire territory than in smaller communities, more births were probably erroneously considered to be exposed. Population density, which was strongly correlated with population size (r_Spearman = 0.69), was included in all models, together with family income, to adjust for all factors associated with urbanisation including this possibility of misclassification. Cities of 50 000 or more inhabitants were excluded from the analysis to reduce misclassification of exposure.

In all models, other risk factors, such as maternal age, were taken into account, and the possible association with socioeconomic status¹⁵ was dealt with by including family income in the models. We decided to correct for overdispersion of data when present since this situation could underestimate the variance.

Population density and family income also affect access to medical care, including prenatal diagnosis. The probability of ascertainment of cases of congenital malformations may vary according to the severity of the anomaly and the completeness of the access to information regarding pregnancy terminations. We recognise the possibility that a proportion of the milder cases may be more easily recorded in bigger cities but there is no correlation between the quantitative index of exposure to MSWI and any of the indicators of population size or density. The percentage of terminations of pregnancy following prenatal diagnosis, for chromosomal anomalies in particular, was significantly correlated with family income in our data. The registry coverage for pregnancy terminations was less exhaustive than for births, and most probably better in the larger cities. This explains the association with population density for chromosomal anomalies in particular; this association disappears once the percentage of pregnancy terminations among anomalies registered was included in the models but their association with exposure was reinforced (RR = 1.37 (95% CI 0.91 to 2.07) in the intermediate exposure category; RR = 1.10 (0.75 to 1.60) in the high exposure category). The risk estimates for the other main groups of anomalies did not change. The dose-response relations between subgroups of major anomalies and both the exposure index and road traffic density were not modified by the inclusion of the percentage of pregnancy terminations in the models.

MSWI emissions are a mixture of various compounds, the most toxic of which are dioxins and metals, such as cadmium, lead, and mercury.³ In our study, the estimate of dioxin emissions can be considered as a tracer for the whole mixture, since expert estimates for dioxins and metals were highly correlated. Our expert ratings were recently confirmed a posteriori: an MSWI with the highest exposure rating was recently closed down because dioxin levels in local animal products exceeded permissible levels (up to 25 pg/g of fat for dioxins (I-TEQ) in meat). The significant association that we found between exposure to MSWIs and such specific anomalies as obstructive uropathies and cleft palate are biologically plausible, as hydronephrosis, cleft palate, genital anomalies, and impaired dentin and enamel formation are the main consequences of developmental toxicity of 2,3,7,8-TCDD in animals.¹⁶

Knowing the close ties between exposure to road traffic and factors associated with ascertainment of congenital anomalies (population density, family income) or other risk factors, can we still interpret the results related to road traffic? Some of the associations seen with road traffic are for malformations that may be diagnosed after birth (cardiopathies, renal anomalies): better late ascertainment in bigger cities may thus explain the associations found. Comparison between the French Central-East registry and registries in Sweden and California showed that the French registry had good sensitivity for kidney malformations;¹⁷ this study also showed that approximately 22% of the reported ventricular septal defects could be considered mild.¹⁸ If ascertainment bias occurred it most likely affected only this fraction.

The models controlled for several variables, but residual confounding remains a possibility. On the other hand, since higher population size generates higher pollution levels, controlling for population density may in some instances result in overadjustment when studying the effects of road traffic pollution. The plausibility of the association with road traffic has recently been reinforced by the observation of higher risks of cardiac anomalies related to ambient air pollution in Southern California,¹⁹ where vehicular traffic is the major source of air pollution. In summary, although both MSWI emissions and road traffic are plausible candidates for explaining the excess risks found, the ecological design of our study limits its conclusions, and we cannot exclude several alternative explanations, including exposure misclassification, ascertainment bias, or residual confounding. It should also be stressed that some of the effects observed in relation to exposure to MSWI emissions should, if real, be attributed to old-technology MSWI and the persistent pollution they generate.