|Home | About | Journals | Submit | Contact Us | Français|
We have attempted to quantify the most up-to-date estimate of the association between cigarette smoking by the mother and preterm delivery. Studies were selected for inclusion in this review if they were prospective, reported data stratified across at least two levels of maternal smoking, and defined preterm delivery on the basis of gestational age. In a meta-analysis we combined results from multiple studies that reported on preterm delivery and maternal smoking during pregnancy. Pooled odds ratios were computed for various strata of smoking intensity with the Mantel-Haenszel fixed-effects model. Twenty studies met all inclusion criteria and were included in meta-analysis. The pooled point estimate from 20 prospective studies on any maternal smoking versus no maternal smoking was 1.27 (95% confidence interval, 1.21-1.33). Subgroup analyses stratifying maternal smoking on number of cigarettes per day suggest a dose-response relationship at low to moderate levels of smoking, which was not further increased at high levels of smoking. A nonsignificant level of publication bias appears to exist in the smoking-preterm delivery literature. Cigarette smoking is a preventable risk factor that is associated with preterm delivery. Consistent results across many study populations and research designs and evidence of a dose-response relationship support its causal role in preterm delivery.
Preterm delivery is the single most important contributor to infant mortality rates in technically developed countries,1 where rates of preterm delivery have increased over the last 20 years. In part, this observed trend may be attributed to increasing numbers of multiple births, more frequent use of ultrasonography to determine gestational age, and a growing tendency to register live births at very early gestational ages.2 However, maternal smoking is a potentially preventable cause for preterm delivery. Furthermore, a dose-response effect has been observed in single studies, in which mothers who smoke greater amounts during pregnancy have progressively higher rates of preterm deliveries. This study summarizes the currently available data on the association between smoking and preterm delivery and on a dose response of the association, by reporting pooled estimates for odds ratios in association with preterm delivery and smoking.
Studies and review articles relating preterm deliveries and maternal smoking were identified with MEDLINE (1966-1997) and by searches of references cited in identified articles. MEDLINE was searched with the subject headings premature labor and smoking, as well as the text words preterm delivery and preterm birth, which resulted in the identification of 104 citations. Any articles containing original data were retrieved in their entirety, with study type ascertained when the full text of the article was reviewed. Studies were limited to the English language. Although a few articles were identified in Eastern European literature by MEDLINE, they were mostly case reports or case-control studies and did not fulfill our inclusion criteria. Eight retrospective, case-control studies were also identified through this search and were not selected for analysis but are discussed in the Results section.
Studies selected for analysis had to meet the following requirements: (1) The study was prospective (ie, smoking exposure was ascertained before perinatal outcome); (2) the study reported data on rates of preterm delivery stratified across at least two levels of maternal smoking, or the study included a summary estimate (odds ratio or relative risk) with reported confidence intervals; (3) the study defined preterm delivery according to gestational age. The existence of publication bias was assessed among the selected studies by means of a funnel plot3 of individual study odds ratios plotted against study size.
The methods of Mantel and Haenszel4 were used to estimate effect size and confidence intervals. Calculations were conducted with Microsoft Excel5 software by methods described by Fleiss.6 Individual estimates from studies were pooled as odds ratios; odds ratio estimates were derived from the data for those studies that reported relative risks. The odds ratio is comparable to the risk ratio when the baseline risk is low (ie, when the event rate in the control group is <10%, as was the case in almost all of the identified studies).7 Whenever possible, adjusted odds ratios were used to provide more accurate estimates of the true underlying relation between smoking and preterm delivery. Most studies adjusted for maternal age, race, gravidity, parity, income, and other social and demographic factors.
A total of 64 published articles were identified that reported original data on the association between maternal smoking and preterm delivery. Twenty studies in 21 reports8-28 met all the inclusion criteria and were included for analysis (see Table I). Four studies reported outcomes only in the form of mean gestational age changes for children born to smokers versus nonsmokers. Whereas these data support the association between maternal smoking and preterm delivery, in none of these reports were sufficient data presented to provide results that could be pooled with data from the other 20 studies. As a result, these 4 studies are not included in the analysis. A total of 43 studies were excluded from analysis, most because they were not prospective studies (references available from the authors on request).
The overall association between cigarette smoking and preterm delivery is reasonably well described by cohort studies. However, most did not identify the duration of exposure, with the authors choosing simply to dichotomize duration of cigarette use into any maternal smoking during pregnancy versus none. The level or intensity of smoking, however, was better characterized. Most cohorts were split into two or three levels of exposure (for example, groups in which mothers actively smoked 0-9, 10-19, or ≥20 cigarettes per day). Only 1 study8 looked at exposure to passive smoking, and so passive smoking exposure is not included in this review.
Only 2 of 20 studies showed a protective, but statistically nonsignificant, association between smoking and preterm delivery.18, 19 In the first,18 preterm delivery was defined as >32 weeks but <37 weeks of gestation; hence a large portion of the very early preterm deliveries are missing, where the observed association is likely to be strongest. For <32 weeks’ gestation, the same study found a relative risk of 1.95 (95% confidence interval, 1.30-2.93) for smoking and preterm delivery, pointing to the masking of smoking’s effect by narrowly defining preterm delivery. The data from the group with delivery at <32 weeks’ gestation were not reported in a manner that could be combined with the group with delivery at >32 weeks’ gestation; hence for this review the most conservative data (which showed a protective association) were used. The other article19 did not have a purely nonsmoking cohort for comparison; instead they used as controls women who on average smoked <1 cigarette per day (including nonsmokers). This study is similar to 2 other studies,12, 14 which, nevertheless, reported an increased odds of preterm delivery for mothers who smoked. Exposure misclassification of this sort would bring estimates toward the null (ie, decrease the association between smoking and preterm delivery).
Table I presents data on any maternal smoking during pregnancy versus no maternal smoking. When multiple data were present in the original report, efforts were made to dichotomize results into any maternal smoking versus no maternal smoking before inclusion in this analysis. The pooled estimate with 95% confidence interval for any smoking versus no smoking and preterm delivery is 1.27 (1.21-1.33). A sensitivity analysis that excluded the 2 studies with the largest and the 2 studies with the smallest point estimates resulted in no significant changes to the pooled estimate (data not shown).
Tables IIA and IIB present maternal smoking data stratified into light to moderate smoking (<20 cigarettes per day) and heavy smoking. When data were not available for 0 to 20 cigarettes per day, alternate levels (eg, 0-9, 1-14 cigarettes per day) were used. Similarly, the heavy exposure pooled estimate used ≥11, ≥15, or ≥16 cigarettes per day when other data were not available. It was thought that 20 cigarettes per day, representing 1 pack per day, was a sensible cut point for exposure. The pooled estimate with 95% confidence interval is 1.22 (1.13-1.32) for light to moderate smoking and 1.31 (1.20-1.42) for heavy smoking.
Separating studies into two levels of smoking showed a trend toward a dose-response relationship between cigarette smoking during pregnancy and the incidence of preterm delivery. This trend was less apparent when studies were divided into 3 groups on the basis of predefined levels of smoking (see Tables IIIA, IIIB, and IIIC). For light smoking, defined as 0 to 10 cigarettes per day, the pooled estimate with 95% confidence interval was 1.25 (1.12-1.38). Moderate smoking, approximately 11 to 20 cigarettes per day, resulted in a pooled estimate of 1.38 (1.23-1.55). Heavy smoking, generally more than a pack per day, had a pooled estimate of 1.31 (1.19-1.45).
Eight case-control studies on the relationship between preterm delivery and maternal smoking were identified.29-36 Unlike the prospective studies, however; they varied widely in their quality and procedures. For example, one study31 selected controls from deliveries >39 weeks of gestation, to avoid misclassification bias that might occur with incorrect gestational dates for preterm infants, but this artificially inflates the association of smoking with preterm delivery. Another study35 determined gestational age more carefully for cases than for controls, potentially introducing other biases.
Similarly, the definition of exposure to cigarette smoking varied widely from one case-control study to the next (see Table IV). Whereas some looked at smoking at a certain time point during pregnancy, others considered smoking status before pregnancy versus during pregnancy. One study37 looked at exposure to passive cigarette smoking. Only 2 studies stratified women according to the level of smoking29, 30; the others considered smoking a dichotomous variable. No case-control studies showed a protective effect from smoking on preterm delivery. Most studies reported odds ratios in the 2.0 to 3.0 range, with the overall range from 1.0 (no effect) to 4.20. Because of the great heterogeneity between studies (eg, in choice of control group and in exposure definition and the like), the data were not considered suitable for statistical pooling.
A funnel plot to assess potential publication bias in the prospective studies is presented in Fig 1, which was created with the use of data presented in Table I. Publication bias occurs most often when small studies with “negative” results are less likely to be published. A biased plot would be asymmetric around the pooled estimate, spreading out unevenly to both sides. Among prospective studies of smoking and preterm delivery, there is a right-skewed distribution. This suggests that some studies with point estimates <1.27 (the pooled estimate) may have been conducted but were not published.
A truncated dose-response trend was observed when studies were separated into 3 groups of low, medium, and heavy smoking. It is possible that heterogeneity in design and execution of studies masked a dose-response relationship that, whereas it was apparent within individual studies, was less evident in the pooled groups. The heaviest smokers had an association between smoking and preterm delivery that was essentially the same as that for moderate smokers. An alternate explanation lies in the nature of the exposure; there are fewer people who smoke large amounts (≥40 cigarettes per day) relative to the low and medium smoking groups. As a result the heavy smoking group may be increasingly undifferentiated from the medium smoking group to create a clear dose-response contrast between medium and heavy smokers.
When a random-effects model was used to generate pooled odds ratios, results were essentially unchanged (data not shown). This type of model may better incorporate heterogeneity between studies than other models, and because its results were comparable to those found by the fixed-effects methods of Mantel and Haenszel, we may infer sufficient homogeneity between studies for this report.
There are several limitations to the data. Publication bias may affect the results of this study, but we believe its contribution is minimal. The overall pooled estimate of 1.27 closely approximates the point estimate of the largest study, with an odds ratio of 1.20. This suggests that, despite the possible existence of some publication bias, the pooled estimate is likely a valid estimate of the true underlying effect.
Another potentially limiting factor is that various methods were used to estimate gestational age, ranging from extrapolating the self-reported last menstrual period to estimates that were based on a combination of last menstrual period, ultrasonography, and clinical examination. Furthermore, there was some heterogeneity of exposure criteria, with cigarette smoking ascertained at different points in pregnancy and smoking categorization differing from one study to the next. Heterogeneity also existed in the rates of preterm delivery from one study to the next, ranging from 2% to 33%. This might be attributable to geographic differences between sites, different target populations and recruitment, and time when the study was conducted (earlier studies tended to have higher rates; studies published before 1983 had among them the four highest rates of preterm delivery). Finally, there are multiple known causes of preterm labor, which may have been differentially controlled among studies. Adjusted odds ratios were used whenever available to better control for possible sources of confounding, but the adjustments differed from one study to the next.
Despite these potential limitations, this analysis provides the most up-to-date and validated summary of the effect of maternal cigarette smoking on the incidence of preterm delivery. A prior meta-analysis conducted in 1984 found pooled odds ratios of 1.32 to 1.56 when preterm delivery was variously defined as before 34, 35, 36, 37, or 38 weeks of gestation.39 The estimate for preterm delivery based on a gestational age <37 weeks, which was the cut point used for this meta-analysis, found a pooled odds ratio of 1.40 (95% confidence interval, 1.30-1.51). This estimate was based on 42,000 subjects pooled from 5 studies, whereas the current meta-analysis pooled 20 studies and >100,000 subjects to get a pooled odds ratio of 1.27 (95% confidence interval, 1.21-1.33). The current meta-analysis used adjusted odds ratios whenever possible, whereas the 1984 meta-analysis used strictly unadjusted values. This report was limited to prospective studies to avoid any potential bias that might overestimate the effect of smoking in retrospective data collection.
A benefit of meta-analysis used to summarize data is that a broad range of study sites and populations are represented in the pooled estimate. This expands the generalizability of results, providing greater credibility and wider clinical and public health significance to the finding. Furthermore, there are a large number of observations in the overall analysis, which is oftentimes prohibitively difficult in any single study.
Maternal smoking has been shown to be associated with many other problems for the newborn in addition to preterm labor. There are strong associations with low birth weight,40 spontaneous abortion,41 abruptio placentae,42 ectopic pregnancy,43 impaired respiratory function in newborns,44 and psychiatric adjustment of the child in later life,45 to name just a few.
In his presidential address before the Royal Society of Medicine in 1965, Bradford Hill noted the following 9 conditions that help strengthen causal inference for an observed association: strength of the association, consistency, specificity, temporality, dose response, plausibility, coherence, experiment, and analogy. With all the data that are now available on the association between cigarette smoking and preterm delivery, these criteria have been met and any controversy regarding maternal smoking and preterm delivery appears to have been sufficiently addressed.