Smoking during pregnancy has repeatedly been associated with a modestly increased risk of infant mortality.1–5 Such an association may be due to a causal effect of smoking on infant mortality or to factors associated with smoking. The hypothesis of a causal association would be supported if women who quit smoking from one pregnancy to the next reduced their offspring's risk of mortality.
We and others have previously reported that smoking cessation reduces stillbirth risk.6,7 In this population-based cohort study, we assessed the effect of smoking cessation on the risk of infant death. We also studied how the smoking-related effect on infant mortality is mediated, by accounting for possible intervening variables in the causal chain such as preterm birth and placental abruption.
In this population-based cohort study we used data from the population-based Swedish Medical Birth Register, the Causes of Death Register (held by the Swedish National Board of Health and Welfare), the Education Register, and the Register of Population and Population Changes (held by Statistics Sweden). Because the Birth Register includes information about the unique national registration numbers of mothers and infants, it is possible to link the Birth Register to other registers as well as to study successive births to each woman.
We selected 557,318 women who had their first and second consecutive singleton births occurring between years 1983 and 2002. Our main outcome of interest was infant mortality after the second pregnancy. From the cohort of second pregnancies, we excluded women who experienced a stillbirth in the second pregnancy (n = 1515) and women with missing information on duration of pregnancy in the second pregnancy (n = 602). To exclude infants with erroneous birth weights or gestational ages, we used a cut-off of birth weights below or above 5 standard deviations from the mean at each gestational week in the present study sample (n = 155). Thus, the final cohort of live births used for analyses of infant mortality after the second pregnancy consisted of 555,046 pregnancies. By applying the same exclusion criteria, for comparison, we also studied infant mortality rates after the first pregnancy. The cohort of women with live births after the first pregnancy is smaller (n = 553,797), due to a larger number of stillbirths in the first pregnancy. In the cohort of second pregnancies, we retained women who suffered stillbirth or infant death in the first pregnancy. Similarly, in the cohort of first pregnancies, we included women who later suffered stillbirth or infant death in second pregnancy.
Exposure and Outcome Information
Starting with the first antenatal visit, information is prospectively collected for all births including demographic data, reproductive history, and complications during pregnancy, delivery, and the neonatal period. Such information is forwarded to the Birth Register through copies of standardized antenatal, obstetric, and neonatal records.
Information about smoking habits is collected at the woman's first antenatal visit, at which the woman spends about 1 hour with the midwife for an interview and an examination. More than 95% of the Swedish pregnant population attend antenatal care before the 15th gestational week.8 Current smoking habits are recorded on the antenatal care records in checkboxes and categorized as nonsmoking, 1–9 cigarettes per day, or 10+ cigarettes per day. We had data on maternal smoking for 94% and 94% of the first and second pregnancies, respectively; 89% had smoking information in both pregnancies. Based on whether the women smoked or not in the first or second pregnancy, they were defined as “never–smokers,” “quitters,” “starters,” or “persistent smokers.” We had no information on smoking cessation during pregnancy.
Information on maternal age and complications during pregnancy is recorded when the woman is discharged from the hospital. Complications during pregnancy are coded according to the Swedish versions of the International Classification of Diseases, ninth and tenth revisions (ICD-9 and ICD-10, respectively). Placental abruption was defined using ICD-9 code 641C and ICD-10 code O45.
Information about date of birth, birth weight, sex, and gestational age is registered at delivery. Early second trimester ultrasound screening was introduced in Sweden in the 1980s, and since 1990, this examination is offered to all pregnant women. Ninety-five percent of pregnant women accept this offer.9 Thus, early second-trimester ultrasonographic examinations were generally used to determine gestational age; otherwise the last menstrual period was used. When comparing estimates from ultrasound and last menstrual period, the former predicts shorter gestational durations on average. Ultrasound screening reduces the rates of postterm deliveries, with less impact on preterm rates.9 To control for calendar differences in ultrasound screening, we adjusted the analyses for year of delivery. The interpregnancy interval was calculated as the number of completed months between the birth of first child and estimated conception date of the second child.
Information about mother's country of birth was derived from the Register of Population and Population Changes, and categorized into Nordic (Sweden, Denmark, Finland, Iceland, and Norway) and non-Nordic countries. Information about highest level of education of the mother was derived from the Education Register at years 1991 and 2001. Information about infant mortality, including age at death, was obtained from the Swedish Causes of Death Register. The study was approved by the research ethics committee at Karolinska Institutet, Sweden (No. 4863/2005).
We used standard survival analysis (cohort) methods to analyze the data. The main event of interest was infant death after the second pregnancy. Infant mortality rates were calculated as number of deaths over risk time in person-years, with 95% confidence intervals (CIs) assuming a Poisson distribution. We report the rates by 1000 person-years to mimic the standard measure of infant mortality rates as deaths per 1000 births. The standard measure will approximate the rate per 1000 person-years, because a birth is assumed to be followed for 1 year on average (deaths less). If the mortality rates are low, the 2 measures will be almost the same.
Liveborn infants with a gestational age of at least 22 weeks were considered as being at risk for infant death (from birth until 1 year of age). On average, each infant contributed 52 weeks of risk time (less for deaths). During the follow-up, there were 2102 infant deaths in the cohort of second pregnancies, and 2761 deaths in the cohort of first pregnancies.
To estimate the association between smoking and infant mortality rates we used Cox proportional hazards regression analysis, yielding hazard ratios (HR) with 95% confidence intervals (CIs) as measures of association. The timescale in all Cox regression analyses was age of the infant, with the underlying time unit in days, thus automatically adjusting for attained age of the infants throughout follow-up.
We categorized smoking habits in 2 successive pregnancies as nonsmoker in both pregnancies (“never-smokers”), smoker in first/nonsmoker in second (“quitters”), nonsmoker in first/smoker in second (“starters”), and smoker in both pregnancies (“persistent smokers”). The category “never smokers” was used as the reference group. Smoking was further categorized into light smoking (1–9 cigarettes per day) and heavy smoking (10+ cigarettes per day).
We adjusted the models for the following a priori-selected potential confounders: mother's age (≤24, 25–29, 30–34, and ≥35 years), mother's education (<12, ≥12 years of schooling), mother's country of birth (Nordic or non-Nordic), interpregnancy interval (<12, 12–23, 24–35, and ≥36 completed months), and year of second delivery (1983–1987, 1988–1992, 1993–1997, and 1998–2002).
We also investigated whether the effect of smoking on infant mortality was mediated by gestational age or placental abruption by adding them one by one to the model. Gestational age at birth was categorized into 5 groups (22–27, 28–31, 32–36, 37–41, and ≥42 completed weeks), where “37–41 weeks” was the reference group. Placental abruption was measured as “yes” or “no.” For comparison between models, we present the deviance and degrees of freedom for each model.
We formally tested interactions with attained age of infant, the underlying timescale (ie, nonproportional hazards), using the likelihood ratio test with P-values. There were indications of nonproportional hazards with respect to maternal smoking habits and gestational age. We therefore proceeded to model interactions by splitting infant attained age into 3 categories (<4 completed weeks [neonatal period], 4–15 weeks, and ≥16 weeks [postneonatal periods]), and estimated separate effects of maternal smoking habits, gestational age, and placental abruption for these time windows.
We excluded all women with missing information on maternal smoking or any other covariate from all Cox regression analyses. Complete information on all covariates was available for 489,961 (88%) of the 555,046 women in the second pregnancy.
We used Stata version 10 Intercooled for the statistical analyses (release 10; StataCorp., College Station, TX).
Among the 553,797 live single births in the first pregnancy, there were 2761 infant deaths after the first pregnancy. Overall infant mortality rate was 5.0 per 1000 person-years (95% CI: 4.8–5.2). Among the 555,046 live single births in the second pregnancy, there were 2102 infant deaths. Overall infant mortality rate was 3.8 per 1000 person-years (95% CI: 3.6–4.0).
Infant mortality rates in the first and second pregnancy increased by the amount of smoking in the respective pregnancy (Table 1). The highest rates of infant mortality after the first pregnancy were observed among women who smoked in the first pregnancy (“quitters” and “persistent smokers”) (6.2 per 1000 person-years, in both groups). After the second pregnancy, the infant mortality rate was the highest among the persistent smokers (5.9 per 1000 person-years), whereas rates were substantially lower among those who quit smoking between pregnancies and never-smokers (3.5 and 3.1 per 1000 person-years, respectively). Women who started to smoke between pregnancies had a similar infant mortality rate after the first pregnancy as never-smokers, but a higher infant mortality rate after the second pregnancy. In quitters, starters, and persistent smokers, there was a dose-relationship between numbers of cigarettes smoked and infant mortality rates.
Infant mortality rates are higher in first pregnancies than second pregnancies, which gave rise to some paradoxical differences in mortality rates between the first and second pregnancies. In particular, infant mortality rates in women who were nonsmokers in the first pregnancy and light smokers in the second pregnancy (4.3 in first and 4.0 in second pregnancy) should be compared with never-smokers (4.3 in first and 3.1 in second pregnancy). Quitters and starters were also more likely to be light smokers than persistent smokers. Among persistent smokers, 61% were light smokers in the first and 62% were light smokers in the second pregnancy. Among quitters, 79% were light smokers in the first pregnancy and among starters, 81% were light smokers in the second pregnancy.
As expected, mean birth weight increased from first to second birth, and smoking was associated with lower birth weights (Table 2). The largest increase in birth weight (233 g) was seen among quitters, whereas women who started to smoke had the smallest increase (80 g).
Table 3 provides a description of the cohort and crude (unadjusted) rates of infant mortality after the second pregnancy by maternal and pregnancy characteristics. Infant mortality rates were increased among infants of young mothers, mothers with low education, those born outside the Nordic countries, and those with short interpregnancy intervals. Infant mortality rates decreased in later calendar years. Rates of infant mortality dramatically increased with decreasing gestational age. For infants born in the earliest weeks (22–27 weeks), the rate of dying was 611 per 1000 person-years, compared with 2.2 among term births (37–41 weeks). Placental abruption was associated with a substantially increased infant mortality rate.
Table 4 shows maternal smoking habits in the 2 successive pregnancies and the estimated hazard ratios of infant mortality after the second pregnancy. The model adjusted for confounders shows that light-smoking women in the first pregnancy who stopped smoking in the second pregnancy (“quitters”) faced a similar risk for infant death after the second pregnancy to those who never smoked (HR = 1.0; CI = 0.8–1.2). However, heavy smokers in the first pregnancy who stopped smoking in the second pregnancy only partly reduced their risk (1.4; 1.0–2.0). For infants born to persistent smokers who smoked heavily in both pregnancies, the corresponding hazard ratio was 2.0 (1.7–2.4). For infants born to women who smoked in the second but not in the first pregnancy (“starters”), the risk of infant mortality was not increased for light smokers in the second pregnancy (1.1; 0.8–1.5), but was for heavy smokers (1.8; 1.0–2.9).
Next, we investigated whether the smoking–infant mortality association was mediated by well-known smoking-related adverse pregnancy outcomes (preterm birth and placental abruption) by adding these variables one by one to the model. The model adjusted for confounders and gestational age showed overall decreased hazard ratios among starters and persistent smokers. For example, the HR among persistent smokers who smoked heavily in both pregnancies decreased from 2.0 to 1.5 after adjustment for gestational age. In contrast, in the model adjusted for confounders and placental abruption, the hazard ratios were essentially unchanged compared with estimates from the model only adjusting for confounders. This suggests that there was no mediating effect of placental abruption on the association between smoking and infant mortality.
When we adjusted for both gestational age and placental abruption, the hazard ratios related to persistent smoking remained elevated (among the heaviest persistent smokers 1.5; 1.3–1.9) (model adjusted for confounders, gestational age, and placental abruption). The hazard ratio for women who started to smoke heavily in the second pregnancy was similarly increased (1.5; 0.9–2.4), as was the hazard ratio for heavy-smoking women in the first pregnancy who quit smoking in the second pregnancy (1.4; 1.0–2.0). Judging from the deviances of each model, we gained most by including gestational age to the model. Also, the independent effect of placental abruption was completely explained by gestational age in the final model, which included both variables.
Among 2102 infant deaths after the second pregnancy, 1210 occurred before 4 completed weeks (mortality rate 28.4 [26.8–30.0] per 1000 person-years), 516 at 4–15 weeks (4.0 [3.7–4.4]), and 376 at 16 weeks or more (1.0 [0.9–1.1]).
Table 5 shows the association between smoking habits and infant mortality after the second pregnancy by attained age of infant. Due to sparse data, we were not able to study light and heavy smoking by age, but show only results from never-smokers, quitters, starters, and persistent smokers. When we tested for nonproportional hazards over the age of the infant, we observed an interaction between maternal smoking and infant age (likelihood ratio test P < 0.01). The model adjusted for confounders shows that, compared with never-smokers, infants born to persistent smokers had a 20% increased mortality rate during the neonatal period, a more than doubled mortality rate at 4–15 weeks of age, and an almost doubled mortality rate after week 16.
We also investigated interactions between infant age and gestational age at birth, and infant age and placental abruption (likelihood ratio test P < 0.01, P = 0.95, respectively). After also including age-varying effects for gestational age, and placental abruption (model adjusted for confounders, gestational age, and placental abruption), the hazard ratios for persistent smokers were reduced at all ages; the excess rate disappeared for the neonatal period, but the mortality rate was still more than doubled for ages 4–15 weeks and increased by 80% for ages 16 weeks and older. The hazard ratio among quitters also increased over time, from 0.9 (0.7–1.2) during the first 4 weeks to 1.2 (0.8–1.8) and 1.5 (1.0–2.2) in the age windows 4–15 weeks and 16 weeks or older, respectively, suggesting possible resumption of smoking after delivery. The effect of gestational age on infant mortality in the second pregnancy varied substantially with the age of the infant. The increase in mortality rate with severity of preterm birth was especially pronounced in the neonatal period (<4 weeks), but had also a substantial impact on mortality rates at older ages. Postterm infants (≥42 weeks) faced a nearly doubled risk of dying during the neonatal period compared with term infants but no increased risk beyond the neonatal period. After accounting for gestational age and other factors, placental abruption was not associated with neonatal or postneonatal mortality.
This study shows that smoking cessation between 2 pregnancies reduces the infant mortality rate in second pregnancy. By using smoking measurements from 2 successive pregnancies in the same woman we create a “natural experiment” of changing exposure, which should reduce confounding and thus strengthen causal inference.
Mortality rates during the first year of life depend heavily on the age of the infant, with the majority of infant deaths occurring in the first month. Our findings suggest that the smoking-related increased risk of neonatal mortality is mediated by a smoking-related effect on gestational age. Smoking during pregnancy is probably causally associated with preterm birth,10,11 which is an important cause of infant death.12
During the postneonatal period, we found that persistent smoking during pregnancies was associated with a more than doubled risk for infant death during the age 4–15 weeks and an almost doubled risk for death at 16 weeks and later, associations that could be explained only to some extent by gestational age and placental abruption. These results imply that the mechanisms by which smoking influences infant death may differ for the neonatal and the postneonatal periods. (Maternal smoking after delivery may also contribute to the postneonatal risk.) Smoking during pregnancy is consistently reported to increase the risk of sudden infant death syndrome (SIDS), which typically occurs at 2–5 months of age. In addition, as the smoking-related risk of SIDS appears to be largely independent of birth weight,13,14 it is probably also independent of gestational age, given the close association between birth weight and gestational age. Maternal smoking has also been associated with infant deaths due to infections and respiratory diseases2,15 and hospitalization for respiratory diseases.16
One strength of our population-based study is our use of the Swedish Birth Register, from which we can easily identify women with 2 successive births during a 20-year period. Essentially all births in Sweden are recorded in the register, and the long-term established reporting system reduces potential selection biases as well as misclassification.
A methodologic strength is the application of survival analysis to birth data. Traditionally, when studying the risk of infant mortality, relative risks have been estimated using odds ratios for binary outcomes. However, infant mortality data are survival outcome data, which include both a binary outcome and a time-to-event component. If the association between exposure and outcome is influenced by time, then such data are most appropriately analyzed using survival analysis, where it is possible to estimate rates, that is, number of events over risk time and hazard ratios adjusted for time. In the present study, attained age of an infant was not a strong confounder of the relationship between smoking and infant death, but a strong effect modifier. Survival analysis enabled us to fit and formally test this effect modification in a single model. Survival analysis provides an elegant unified framework to model infant mortality data, which should be considered more often in perinatal epidemiology.17
A potential weakness of our study is the self-reported information on maternal smoking in early pregnancy. We had no data on smoking cessation throughout the pregnancies or in the time period between the 2 pregnancies. If some women who reported smoking at the first visit to antenatal care later stopped smoking during pregnancy, such misclassification would bias the smoking-related associations toward the null.
We did not have information on maternal smoking after delivery and during the first year of life of the infant. The paradoxical increased risk in the second pregnancy among women who were heavy smokers in the first pregnancy, but reported nonsmoking in the second pregnancy may be due to underreporting of exposure by mothers who smoked during pregnancy or resumed smoking after delivery. A similar bias may explain the increased risk in the first pregnancy among women who reported nonsmoking in the first pregnancy and heavy smoking in the second pregnancy.
Our study design uses 2 successive pregnancies from the same woman, which reduces confounding from factors that are constant over time. However, unmeasured exposures associated with changes in smoking status across pregnancies could influence the results. For example, there is a substantial recurrence risk of SIDS and other possibly smoking-related causes of infant mortality,18 although smokers with an infant death are not more likely to quit smoking in the next pregnancy compared with other smokers.19 Another potential limitation of our results is that smoking may be a proxy for other exposures such as alcohol, socioeconomic situation, or stress. To limit such biases, we have adjusted for education, maternal age, interpregnancy interval, and country of birth.
Previous research indicates that smoking cessation reduces the risk of preterm birth.20 Our finding that smoking cessation between successive pregnancies also reduces the risk for infant mortality supports the inference of a causal association between maternal smoking and infant mortality. Moreover, the smoking effect on neonatal mortality is entirely mediated by the well-documented smoking effect on gestational age. For the postneonatal period, smoking remains a risk factor after taking gestational age and placental abruption into account. Our results thus further support continued public health and clinical efforts to reduce smoking prevalence among pregnant women.
1. Kleinman J, Pierre MJ, Madans J, Land G, Schramm W. The effects of maternal smoking on fetal and infant mortality. Am J Epidemiol
2. Malloy M, Kleinman J, Land G, Schramm W. The association of maternal smoking with age and cause of infant death. Am J Epidemiol
3. VandenBerg M. Smoking during pregnancy and post-neonatal death. N Z Med J
4. Salihu H, Aliyu M, Pierre-Louis B, Alexander G. Levels of excess infant deaths attributable to maternal smoking during pregnancy in the United States. Matern Child Health J
5. Salihu H, Wilson R. Epidemiology of prenatal smoking and perinatal outcomes. Early Hum Dev
6. Wisborg K, Kesmodel U, Henriksen T, Olsen S, Secher N. Exposure to tobacco smoke in utero and the risk of stillbirth and death in the first year of life. Am J Epidemiol
7. Högberg L, Cnattingius S. The influence of maternal smoking habits on the risk of subsequent stillbirth: is there a causal relation? BJOG
8. Lindmark G, Cnattingius S. The scientific basis of antenatal care. Report from a state-of-the-art conference. Acta Obstet Gynecol Scand
9. Høgberg U, Larsson N. Early dating by ultrasound and perinatal outcome. A cohort study. Acta Obstet Gynecol Scand
10. Cnattingius S. The epidemiology of smoking during pregnancy: smoking prevalence, maternal characteristics, and pregnancy outcomes. Nicotine Tob Res
11. Kramer M, Demissie K, Yang H, Platt R, Sauvé R, Liston R. The contribution of mild and moderate preterm birth to infant mortality. Fetal and Infant Health Study Group of the Canadian Perinatal Surveillance System. JAMA
12. Andres R, Day M. Perinatal complications associated with maternal tobacco use. Semin Neonatol
13. Oyen N, Haglund B, Skjaerven R, Irgens L. Maternal smoking, birthweight and gestational age in sudden infant death syndrome (SIDS) babies and their surviving siblings. Paediatr Perinat Epidemiol
14. MacDorman M, Cnattingius S, Hoffman H, Kramer M, Haglund B. Sudden infant death syndrome and smoking in the United States and Sweden. Am J Epidemiol
15. Tuthill D, Stewart J, Coles E, Andrews J, Cartlidge P. Maternal cigarette smoking and pregnancy outcome. Paediatr Perinat Epidemiol
16. Wisborg K, Henriksen T, Obel C, Skajaa E, Østergaard J. Smoking during pregnancy and hospitalization of the child. Pediatrics
17. Joseph K. Incidence-based measures of birth, growth restriction, and death can free perinatal epidemiology from erroneous concepts of risk. J Clin Epidemiol
18. Oyen N, Skjaerven R, Irgens L. Population-based recurrence risk of sudden infant death syndrome compared with other infant and fetal deaths. Am J Epidemiol
19. Cnattingius S, Akre O, Lambe M, Ockene J, Granath F. Will an adverse pregnancy outcome influence the risk of continued smoking in the next pregnancy ? Am J Obstet Gynecol
20. Cnattingius S, Granath F, Petersson G, Harlow B. The influence of gestational age and smoking habits on the risk of subsequent preterm deliveries. N Engl J Med