We observed a slight reduction in estimated mean birth weight among term infants exposed in utero to the 2003 California wildfires. The strongest estimated effect was observed for second-trimester exposure, followed by third-trimester exposure.
Climate change scientists predict that wildfires will increase in frequency and magnitude as global temperatures increase and rainfall patterns change (Westerling and Bryant 2008
; Westerling et al. 2006
). These increases in wildfire events are projected to add to atmospheric pollution in the western United States under various climate change scenarios (Spracklen et al. 2009
). In California, smoke impacts are already a required consideration in the planning and execution of preventive wildfire management activities, such as prescribed burns. For example, forest management professionals are required to assess the likely direction of smoke plumes and gauge their potential for impact on smoke sensitive areas (State of California 2001
). Kochi et al. (2010)
make the case that optimal wildfire management policy should explicitly include estimates of health-related and economic costs of wildfire smoke exposure.
Potential etiologic pathways. At least two categories of etiologic pathways plausibly link maternal wildfire exposure with lower birth weight: biological (exposure to air pollution from the fires) and psychosocial (stress caused by direct or indirect consequences of wildfires). A combination of the two is also plausible. We cannot differentiate the contributions of these two pathways here, because we were not able to quantify daily air pollution exposures for each birth, or individual or ecological indicators of maternal stress. Nevertheless, our results may reflect the potential conjoint effect of these two pathways.
Among the biological mechanisms hypothesized as having a possible effect on intrauterine growth rate are hypoxia and/or oxidative stress resulting from exposure to woodsmoke constituents, including carbon monoxide and PM (Siddiqui et al. 2008
), alteration of maternal–placental exchanges, endocrine disruption, and oxidative stress pathways leading to alteration of maternal host–defense mechanisms and subsequently higher risk of infections (Slama et al. 2008
). Reviews on the topic have found limited applicable research from animal and toxicological studies to distinguish these possible mechanisms (Ritz and Wilhelm 2008
; Slama et al. 2008
; Woodruff et al. 2009
). Human studies of the acute effects of wildfire smoke exposure on firefighters have demonstrated inflammatory responses and pulmonary function test declines (e.g., Swiston et al. 2008
). Human experiments in which healthy nonsmokers were exposed to woodsmoke under controlled conditions, with concentrations of PM2.5
> 240 µg/m3
for up to 4 hr, resulted in elevated levels of blood and urine biomarkers indicating oxidative stress and pulmonary inflammation in the lower airways (Barregard et al. 2006
; Sällsten et al. 2006
). More recent experiments confirmed biomarkers of systemic and pulmonary inflammation in blood and lavage, but found no effect on pulmonary function or self-reported symptoms, and minimal effects on indices of heart rate variability (Ghio et al. 2012
Psychosocial aspects of wildfire exposure may also contribute to adverse health outcomes, although this is an understudied topic (Kumagai et al. 2004
). Several studies have observed signs of fetal stress and adverse birth outcomes in the aftermath of disasters such as earthquakes (Weissman et al. 1989
), shipwrecks (Catalano and Hartig 2001
), and terrorist attacks (Catalano et al. 2005
). Plausible causes of stress in the wake of wildfires include loss of property, shelter, money, and other basic individual resources; physical incapacitation or injury; and disruption of sharing and support networks (Fowler 2003
Further analyses with methods that better characterize individual-level pollution exposures and psychosocial stress would be required to distinguish the relative contribution of the two pathways.
Exposure misclassification. Our study captured temporal variation in wildfire exposure, but was limited in its ability to account for spatial variation. The SoCAB includes areas that were likely not directly exposed to heavy smoke plumes or to significant concentrations of diffused smoke. Because we relied on administrative vital statistics records, we were not able to assess whether mothers resided within the air basin throughout their pregnancies, or determine how much time they spent at their primary residence. Meteorology, time–activity patterns, and variations in the built environment likely all contributed to variations in individual exposures, and therefore to exposure misclassification. Because our study did not capture variations in exposure among the exposed, we could not evaluate dose–response relationships. A relatively small number of highly exposed mothers in this region may have been affected to a greater degree than our estimates would predict.
When we conducted a sensitivity analysis that distinguished between births located in higher versus lower PM10 tracts during the wildfire event, we observed that the decrease in birth weight associated with gestational wildfire exposure was similar between these two populations. This result may be attributable to the fact that monitoring results could not adequately capture differences in ambient PM10 levels. Improved analysis with PM10 as a continuous variable or modeling of the PM exposure using satellite data or chemical transport models might reveal a relationship between wildfire-related PM exposure and birth weight. Alternatively, it is also possible that the associations with decreased birth weight were mediated not by air pollution but by some other mechanism, such as stress.
Several factors could explain our finding of stronger associations with exposure during the second and third trimesters than exposure during the first trimester. First, there is the issue of exposure misclassification. Given that the date of conception is less certain than the date of delivery, it is possible that some infants were categorized as exposed in the first trimester, when in fact their conception date occurred after the fire was over. This misclassification bias is unlikely to affect exposure assessment in the third or second trimester, but may lead to an underestimate of effects during the first trimester. Overestimation of the length of the wildfire event would also have resulted in some unexposed births being misclassified as first-trimester exposures. However, reducing the length of the wildfire event did not substantively alter our main effects.
Exposure could also have increased the risk of preterm birth or fetal loss. Because we excluded preterm births and fetal losses, excess preterm delivery or fetal loss among the first-trimester exposed could have differentially eliminated the most vulnerable from our sample. When Breton et al. (2011)
examined prenatal exposure to high PM2.5
levels from the same wildfires among eight counties in Southern California, using vital statistics records for 2003–2004, they did not estimate a significant effect on preterm birth, but they also did not assess fetal loss. The effects of wildfire exposure on birth weight could also be stronger among those exposed in the second or third trimesters for reasons that are not yet understood. Further examination of the effects of trimester-specific exposures in other studies may help to resolve this question.
Controlling for seasonal variation in time-series studies of air pollution can be a challenge (Slama et al. 2008
; Woodruff et al. 2009
). For example, both temperature and ambient (non-wildfire) air pollution exhibit seasonal patterns, and these patterns themselves vary geographically due to differences in regional characteristics. When we controlled for seasonality using quarterly indicators, 87.6% of the infants exposed in the third trimester were conceived in Q1 (January–March), and 94.8% of those exposed in the second trimester were conceived in Q2 (April–June) (, ). This raised the possibility of confounding between trimester-specific wildfire exposure and conception in the first half of the year.
To address this, we conducted a sensitivity analysis, parameterizing the seasonal component of the model as a smooth, continuous, and periodic function of time: the cosinor (Barnett and Dobson 2010
). The general form of the cosinor is sinusoidal, like many natural seasonal phenomena, and has only two degrees of freedom, amplitude and phase [see Supplemental Material, Equations S1–S4
)]. As such, it is readily interpretable, and it has been widely applied to the analysis of seasonal and circadian rhythms (Barnett and Dobson 2010
The cosinor-based analysis yielded effect estimates consistent with the pattern described by our primary model, increasing our confidence in the results. The peak-to-peak amplitude (11.6 g) in seasonal variation was similar to the difference between the minimum and maximum seasonal coefficients from the model using indicator terms (11.9 g). The phase was also consistent with the primary model’s seasons of lowest and highest average birth weight [January–March (Q1) and July–September (Q3), respectively].
Other potential confounders.
We adjusted for several individual-level covariates known to be associated with birth weight, but data on other potential confounders were not available. For example, maternal smoking is not reported on most California birth records, and its inclusion in our study may have changed our results. However, recent studies suggest that although smoking during pregnancy has a large effect on birth weight, in studies of ambient air pollution it does not significantly confound the association between ambient air pollution exposure and adverse perinatal outcomes such as infant mortality and preterm birth (Basu et al. 2003
; Darrow et al. 2006
In previous research of wildfire health effects, few studies have attempted to separate the fraction of smoke attributable to wildfire from that attributable to background air pollution (for a review, see Dennekamp and Abramson 2011
). In areas with significant sources of other pollution, such as the SoCAB, apportionment can be a challenge. Observations from the nearest monitor, which are often used to characterize background air pollution, can be missing during a wildfire episode, sometimes due to the fire itself. To obtain ecologic or individual-level estimates of smoke exposure, several methods can be employed: satellite imagery, dispersion or chemical transport modeling, and/or spatiotemporal interpolation. However, each of these has associated difficulties in implementation and interpretation, especially during a short time window with atypical meteorology, such as the strong Santa Ana winds that fanned the 2003 fires.
To the extent that variation in wildfire-attributable pollution and background pollution are independent, including background pollution in the model could improve the precision of effect estimates, but should not affect the central tendencies. On the other hand, insofar as background concentrations are correlated in space or time with wildfire smoke concentrations (e.g., to the extent that they are similarly determined by physical geography), including background pollution could induce confounding just as including seasonality can. Without access to detailed measurements of both fractions, we elected to consider a strictly temporal contrast, reserving spatiotemporal refinements of exposure for future work.
Previous studies that have attempted to isolate the contribution of wildfire-generated smoke have also compared health effects to a reference period (e.g., Delfino et al. 2009
). However, in any interrupted time-series study, there is always the possibility of an unmeasured confounder with a similar temporal profile to that of the exposure. For example, if a foodborne illness outbreak happened at the same time as the wildfires, and had a negative impact on birth weight, it could conceivably explain part or all of the observed effect. The fact that our unexposed births were drawn from both before and after the exposure window, and from other years at the same time of year, helps to reduce such threats to validity, but cannot eliminate them.