In this study, using individual-level estimates of black smoke exposures during pregnancy, associations were seen between black smoke and birth weight, gestational age and birth weight standardized for gestational age. However, not all associations were significant and, regardless of whether significant, effect sizes tended to be fairly small, in particular for gestational age.
In contrast to most previous studies which have been ecological in design, with relatively crude measures of exposure, this study utilised individual-level exposure estimates on a large unselected birth cohort. Completeness of the PAMPER database for both the number of births and collected information for each birth is one of the evident strengths . The study area covers a clearly defined conurbation with high quality records of land use, including historical records of industrial usage and numbers of residential and industrial chimneys, temperature and season, all of which was included within the exposure estimation modelling . Although from a small geographical area, due to the long time period covered, the study included large variations in black smoke levels, ensuring an additional contribution to statistical power to that given by the large number of births. The black smoke recordings used in this study were collected routinely over the study period, although not all monitors were in place throughout the study period, creating likely geographical differences in uncertainties surrounding the estimated exposures. In addition, monitoring procedures naturally reflect the best practice at the time and, thus, while they are the best estimates available, their accuracy may have varied over time. It is also possible that those with shorter gestations will have more variability in exposure estimation for the whole of pregnancy and the last trimester. In contrast to other studies, we investigated non-linearity of the observed associations. In finding these, it remains to be seen if non-linearity is a true phenomenon, as a result of our data (as the fractional polynomial process is data driven), or due to residual confounding or due to the high level of correlation between black smoke and calendar year.
Residential mobility during pregnancy may be associated with exposure misclassification and therefore may introduce bias . Whilst residential mobility data for the study cohort are not available, there is published, indirect, evidence of population stability for children and older women [21, 22]. Further, only 9% of women in the north of England, in which this study is based, moved during pregnancy between 1985 and 2003, with a median moving distance of only 1.4 km . Data on daily mobility were not available.
While many studies investigate influences on birth weight, or influences of birth weight on later health, many do not account for the impact of gestational age, a major determinant of birth weight. In this study, in addition to crude birth weight, a standardized measure was also used to give a better measure of fetal growth than birth weight alone. Different methods for gestational age assessment (based on the last normal menstrual period or early ultrasound measurements) throughout the study period may introduce bias in gestational age estimation over time. Gestational age in this study was made as objective and accurate as possible by accepting gestational age calculated from the recorded estimated date of delivery (i.e. last menstrual period based) for the majority of births, rather than by using gestational age as recorded in the neonatal notes or birth records. However, these methods are still prone to bias. Most home births in this study period were excluded due to a lack of outcome data. As the majority of home births were in the earlier decades of the study, they would have been associated with relatively high black smoke exposures compared to births in more recent years. However, due to the lack of information from birth ledgers, it was not possible to assess whether this introduced any bias.
Adjustment for year of birth in long-term studies of air pollution and birth outcomes is crucial. However, year of birth and black smoke exposure during pregnancy are highly correlated. While it is possible that adjustment for year of birth may reduce the strength of the association between black smoke and birth weight, it is also possible that residual confounding by year of birth or by other factors affecting birth weight and related to year of birth, for example increase in maternal overweight and improvement in quality of pre-natal care over the study period, may remain. There was a striking decrease in gestational age over the study period due to a number of reasons unrelated to air pollution. For example, the higher prevalence of obstetric interventions, including caesarean section, will account for much of the temporal decrease in gestational age.
The composition of black smoke is known to have varied over the study period , reflecting the diminishing importance of coal and increasing importance of vehicle traffic. From the late 1970s in Newcastle upon Tyne, there was a shift from domestic coal use for heating and cooking to smokeless fuel. However, there was no significant difference in the adjusted association between exposure to black smoke and birth weight between 1962-77 (the period before the implementation of smokeless zones in the area) and 1978-92. Although historical data from the north of England were used in this study, the levels as described still occur in many developing cities of the world. For example, the annual average PM10 concentrations in New Delhi were reported in 2005 to be above 150 μg/m3, and in Beijing above 100 μg/m3 .
Two considerations are of relevance when comparing estimates of associations between black smoke or PM10 and birth outcomes: the nature of black smoke measurements means that their conversion factor to PM10 varies both over time and by composition . For the geographical and temporal setting of this study, a study comparing black smoke and PM10 in the context of health studies concluded that daily average black smoke was a reasonable predictor of daily average PM10 . In the present study, the non-linear relationship between black smoke and birth weight further complicated comparisons with previous studies that used linear models, therefore linear estimates were also reported as well as giving estimated effects at different percentiles of exposure to aid understanding of the functional forms used. The adjusted linear estimate of a 1.7 g (95% CI 0.9 to 2.4) decrease in birth weight per 10 μg/m3 increase in average weekly black smoke exposure over the whole pregnancy period was lower than previous estimates (5, 8-10), but the estimates were more comparable in the non-linear models.
There was a significant interaction between community-level socio-economic status and black smoke on birth weight, with increasing estimated decreases in birth weight per 10 μg/m3 with increasing socio-economic deprivation. Industrial processes in the UK cluster in areas of socio-economic deprivation, but attributing such unequal distribution to unequal pollution impact is complicated . Other covariates that are known to affect birth weight and may be related to black smoke exposures, which we did not have access to, were ethnicity, maternal obesity and smoking. The ethnic minority of primary relevance in Newcastle upon Tyne are from the Indian subcontinent (4% in 1991), therefore the potential for confounding by ethnicity is likely to be small.
A pooled analysis showed that passive smoking reduced birth weight on average by 31 g, a meta-analysis reported a 28 g reduction [33, 34]. The estimated effect of black smoke exposure on birth weight in this study is, therefore, of a similar order of magnitude as the effect of passive smoking, but is, on the other hand, much weaker than the effect of active smoking . Smoking levels in women in the UK never exceeded 45%. For the 1990s we know that smoking was more prevalent in lower socio-economic groups, whereas in the 1960s the prevalence of smoking was high across all social strata. During the 1970s and 80s smoking rates fell sharply in the non-manual occupational groups, leading to the still widening gap between socio-economic groups that exist today. Adjusting for neighbourhood deprivation controls well for smoking to some extent, at least since the 1980s .
Although the association between black smoke and gestational age was not statistically significant overall, it was for each trimester. Given the inconsistent directions between the trimester associations, it is likely that they cancelled out an association for the whole of pregnancy, although the estimated effect sizes were very small. Given the associations between black smoke and both birth weight and standardized birth weight were stronger, it is more likely that the observed associations point towards intrauterine growth restriction as a potential causal pathway for the black smoke effect.
Slama et al.  summarised in detail the potential mechanisms for the association of air pollution on intrauterine growth restriction. These include alterations of umbilical and utero- placental blood flow, and deterioration in the transport of glucose and oxygen to the fetus, all of which can influence fetal growth. Air pollution has been previously associated with increased risks of pre-eclampsia and pregnancy induced hypertension . Data on these outcomes were not available to this study.