The inverse association between smoking and preeclampsia in our study appeared to be weaker than in previous reports (an approximate aOR of 0.7 in previous studies (11
) vs. 0.8 in ours) and was not found for HPE. The sources of information on outcome and exposure could reasonably explain this discrepancy. Preeclampsia was defined on the basis of ICD-9 discharge diagnosis codes. Geller et al. (22
) examined the accuracy of these codes in assigning preeclampsia in a US population and found that the positive predictive value for severe preeclampsia was 84.8% but only 45.3% for mild preeclampsia. In most instances, the consequences of nondifferential misclassification of a binary disease variable are the same as they are for a binary exposure variable (i.e., toward the null); however, it can depend on the sensitivity and specificity of the ICD-9 codes (23
). Nevertheless, the rates of preeclampsia found in this population are similar to those found in other large cohort or data-linkage studies (11
This analysis included only those ethnic subgroups in which the rate of maternal smoking was more than 1%. Even so, the rate of smoking in our population overall was low (<5%). Although underreporting of smoking during pregnancy is known to be common (26
), our population also includes a substantial number of Hispanic and foreign-born women, and both of these groups tend to have lower rates of smoking (27
Smoking information is collected on the birth certificates of women at the time of delivery. These data are routinely used to monitor trends in maternal smoking in the United States. If underreporting of smoking behavior is nondifferential, we can expect bias to be toward the null because smoking is a binary exposure (28
), which may explain the attenuation in risk ratios we observed relative to studies with detailed questionnaire-based information on prenatal smoking. Differential misclassification (in which cases are more likely to underreport smoking) will bias the effect estimate downward so that smoking looks more protective.
Alternatively, or in addition to misclassification of maternal smoking, the attenuation in the relative risk of smoking among isolated preeclamptics may be attributable in part to misclassification of preexisting chronic hypertension (i.e., HPE misclassified as isolated preeclampsia) since there appeared to be no relation between smoking and HPE. Additionally, smoking was dichotomized as ever versus never in our study because there was no dose response evident when examining the more quantitative estimate of smoking available. Therefore, smoking timing, intensity, and frequency may be influencing the magnitude of this estimate by including a large number of light or infrequent smokers in the exposed category. Misclassification of smoking intensity may be responsible for the lack of a dose response using the more quantitative smoking question available on the questionnaire. This misclassification may result from underreporting and variation in smoking behavior over the course of pregnancy.
Additionally, the data sources we utilized lacked information on certain risk factors for preeclampsia, such as prepregnancy height to calculate body mass index, which may result in residual confounding. However, even with these limitations, a clear inverse association between smoking and isolated preeclampsia was observed, qualitatively consistent with previous reports.
The biologic mechanism underlying the repeatedly observed inverse association between smoking and preeclampsia has not been established; however, recent research suggests that carbon monoxide may be the elusive mediator of this paradoxical relation (29
). Although carbon monoxide has historically been considered a toxic by-product of smoking, it also has important physiologic functions (29
), including inhibition of proinflammatory cytokines and chemokines (35
), prevention of vascular constriction (36
), inhibition of platelet aggregation and plasminogen activation (37
), and inhibition of apoptosis (38
). Additionally, carbon monoxide has been found to inhibit formation of reactive oxygen species (34
) and inhibit apoptosis in the differentiated syncytiotrophoblast layer of the placenta specifically (30
). Furthermore, fms-like tyrosine kinase-1 is thought to be critical to the pathogenesis of preeclampsia (40
), smoking is known to decrease circulating levels of fms-like tyrosine kinase-1 (40
), and carbon monoxide has been found to inhibit fms-like tyrosine kinase-1 production in mice (41
There are few studies with sufficient power to examine modification of the smoking-preeclampsia relation by maternal age given that both prenatal smoking and pregnancy at the extremes of maternal age are relatively rare events. We found substantial evidence of modification of the smoking-preeclampsia relation by maternal age such that the protection conferred by smoking was limited to women aged approximately 30 years or less. Moreover, although no overall relation between smoking and HPE was detected, there was a suggestion that, at more advanced maternal ages, smoking may actually increase risk of HPE. That being said, there is some uncertainty in ascertaining cause and effect in the HPE analysis. Women who smoke during pregnancy are very likely to have been prepregnancy smokers, which may have been a cause of their pregestational hypertension. Additionally, preeclampsia in the presence of chronic hypertension is very difficult to diagnose, and chronic hypertension may not be accurately recorded in the discharge records in the absence of related pregnancy complications. If smoking exacerbated underlying chronic hypertension, causing decompensation that is mistakenly diagnosed as preeclampsia, then the relation between smoking and preeclampsia superimposed on chronic hypertension would at least be biased toward the null and possibly begin to look as if smoking increased the risk of HPE. In short, the results with respect to HPE are best conceptualized as cross-sectional and would be improved by more reliable clinical information.
The literature has suggested that, although smoking may decrease the risk of preeclampsia overall, smokers who develop preeclampsia have a higher risk of adverse maternal and fetal outcomes (13
). However, not all previous investigations have found evidence of this association (16
), and few have appeared to adequately account statistically for the independent and joint effects of smoking and preeclampsia on adverse birth outcomes. As a consequence, a belief has arisen that preeclampsia and smoking somehow interact biologically to produce a net harmful effect on birth outcomes via some biologic interplay involving fetoplacental circulation. Our results are counter to this view.
In our study, of a total of 25,937 preeclampsia cases, 1,014 smoked; therefore, we had sufficient power to disentangle the independent effects of preeclampsia and smoking from their joint effects on pregnancy outcome. Our results were unexpected, yet consistent across indicators of fetal growth and were estimated with exceptional precision. Rather than smokers who develop preeclampsia having an excess risk of adverse birth outcomes, we found that their risks were actually lower than would have been expected based on the independent effects of smoking and preeclampsia. Specifically, preeclampsia cases who smoked were at slightly less risk of a preterm delivery than preeclampsia cases who did not smoke, although their risk of preterm delivery was still substantially elevated relative to smokers who did not develop preeclampsia and relative to nonsmokers without preeclampsia, respectively (the reference group).
The same holds true for birth weight. Although the difference in birth weight was greater for smokers who developed preeclampsia than for smokers only, or preeclampsia cases only, it was smaller than the sum of their independent effects, meaning that there was some net negative interaction between these factors. However, unlike preterm delivery, it did appear that smokers who developed preeclampsia had a slightly higher difference in mean birth weight than preeclampsia cases who did not smoke. Note that we detected no interaction between smoking and chronic hypertension on birth weight, meaning that the differences in birth weight between preeclampsia cases who smoked and preeclampsia cases who did not smoke were indistinguishable. These data directly counter the notion that smokers who develop preeclampsia have worse fetal and maternal outcomes than nonsmokers who develop preeclampsia, and they further suggest that, at least among some members of this population, some causal antagonism may be present among smoking and preeclampsia with respect to measures of fetal growth.
However, there is potentially a problem of competing risks when considering the associations between and among smoking, preeclampsia, and preterm delivery. Smoking is related to both pregnancy outcomes, albeit in opposite directions, and preeclampsia often results in a medically indicated preterm delivery. Additionally, preterm delivery may truncate the period at risk of preeclampsia. As has been suggested in other contexts (42
), the appropriate risk set to disentangle these effects is the population of women pregnant at the time of preeclampsia diagnosis. However, our data source lacks critical information on gestational age at preeclampsia diagnosis that would be required to address these issues analytically. Moreover, imputing date of diagnosis based on date of delivery is not straightforward given that watchful waiting, particularly among the less severe cases or those that occur quite early, may result in a large difference between gestational age at delivery and gestational age at diagnosis. However, we conducted a sensitivity analysis by restricting our population to births that occurred after 32 weeks’ gestation and found that the overall relative risk of smoking and preeclampsia was unchanged (relative risks
0.88–0.90). We were unable to account reliably for disease severity with the existing data, and it is possible that smokers with preeclampsia tend to have a less severe form of the disease and are therefore more likely to be treated expectantly and less likely to deliver early. Future investigations would benefit by specific information on gestational age at diagnosis and clinical severity data so that these issues can be disentangled.
In conclusion, we examined the relation among smoking, preeclampsia, and measures of fetal growth in a linked data resource containing over 650,000 singleton pregnancies resulting in a livebirth in New York City between 1995 and 2003. The large size of this data set permitted detailed examination of potential modifiers of these associations and provided evidence against the hypothesis that smokers who develop preeclampsia have worse birth outcomes than nonsmokers who develop preeclampsia. Furthermore, the protection from preeclampsia conferred by maternal smoking appears to be limited to cases without pregestational hypertension and also to younger age groups. It is possible that these associations reflect an underlying heterogeneity in preeclampsia as a disease entity, such that older preeclampsia cases, even those without pregestational hypertension, may have a higher prevalence of comorbidities (e.g., underlying vascular, renal, or autoimmune diseases), which mitigates the degree to which smoking can be causally protective. However, clinical studies would be required to provide a biologic basis for these findings.