Association Between Maternal Smoking During Pregnancy and Severe Mental Illness in Offspring

Key Points

Question  Does exposure to maternal smoking during pregnancy increase the risk of severe mental illness in offspring?

Findings  In a population-based cohort of 1.7 million Swedish offspring, maternal smoking during pregnancy was associated with an increased risk of severe mental illness in offspring. However, sibling comparisons, which ruled out all genetic and environmental confounders that make siblings similar, revealed much weaker and statistically nonsignificant associations.

Meaning  This study suggests that much of the association between smoking during pregnancy and severe mental illness in offspring is likely explained by familial confounding rather than by causal teratogenic effects.

Importance  Several recent population-based studies have linked exposure to maternal smoking during pregnancy to increased risk of severe mental illness in offspring (eg, bipolar disorder, schizophrenia). It is not yet clear, however, whether this association results from causal teratogenic effects or from confounding influences shared by smoking and severe mental illness.

Objective  To examine the association between smoking during pregnancy and severe mental illness in offspring, adjusting for measured covariates and unmeasured confounding using family-based designs.

Design, Setting, and Participants  This study analyzed population register data through December 31, 2013, for a cohort of 1 680 219 individuals born in Sweden from January 1, 1983, to December 31, 2001. Associations between smoking during pregnancy and severe mental illness in offspring were estimated with adjustment for measured covariates. Cousins and siblings who were discordant on smoking during pregnancy and severe mental illness were then compared, which helped to account for unmeasured genetic and environmental confounding by design.

Exposures  Maternal self-reported smoking during pregnancy, obtained from antenatal visits.

Main Outcomes and Measures  Severe mental illness, with clinical diagnosis obtained from inpatient and outpatient visits and defined using International Classification of Diseases codes for bipolar disorder and schizophrenia spectrum disorders.

Results  Of the 1 680 219 offspring included in the analysis, 816 775 (48.61%) were female. At the population level, offspring exposed to moderate and high levels of smoking during pregnancy had greater severe mental illness rates than did unexposed offspring (moderate smoking during pregnancy: hazard ratio [HR], 1.25; 95% CI, 1.19-1.30; high smoking during pregnancy: HR, 1.51; 95% CI, 1.44-1.59). These associations decreased in strength with increasing statistical and methodologic controls for familial confounding. In sibling comparisons with within-family covariates, associations were substantially weaker and nonsignificant (moderate smoking during pregnancy: HR, 1.09; 95% CI, 0.94-1.26; high smoking during pregnancy: HR, 1.14; 95% CI, 0.96-1.35). The pattern of associations was consistent across subsets of severe mental illness disorders and was supported by further sensitivity analyses.

Conclusions and Relevance  This population- and family-based study failed to find support for a causal effect of smoking during pregnancy on risk of severe mental illness in offspring. Rather, these results suggest that much of the observed population-level association can be explained by measured and unmeasured factors shared by siblings.

Maternal smoking during pregnancy (SDP) is associated with a breadth of adverse offspring outcomes, including pregnancy-related, neurodevelopmental, and behavioral problems.^1-3 In particular, recent studies have provided novel evidence of associations between SDP and offspring bipolar disorder,⁴ schizophrenia,⁵ and related outcomes.⁶ These associations raise the possibility that SDP exposure has causal teratogenic effects on the risk of severe mental illness (SMI).⁷ Because nicotine and carbon monoxide cross the placenta and may directly and indirectly (eg, via hypoxia) affect fetal neurodevelopment,⁸ this hypothesis is biologically plausible. Moreover, 8% of pregnant women in the United States smoke,⁹ meaning that SDP exposure could represent a potentially modifiable and important source of risk for SMI.¹⁰ Recent editorials have made explicit causal claims regarding teratogenic effects of SDP on offspring neurodevelopment and psychopathology.^11,12

Understanding the processes underlying associations between SDP and offspring SMI is essential to evaluating a teratogenic hypothesis. Some animal research is consistent with a possible SDP effect on schizophrenia risk,⁷ whereas prior human observational studies have found inconsistent evidence,^13-18 possibly because of differences in methodologies and statistical power.⁵ One study of bipolar disorder, for example, found that the SDP association was greatly attenuated when adjusted for parental characteristics,¹⁹ suggesting that family background, rather than a teratogenic effect, may explain the association.

Tests of teratogenic SDP hypotheses that use observational data are, more broadly, subject to the serious threat of confounding from unmeasured familial factors, including other prenatal exposures, postnatal environments, and genetically transmitted risk of SMI and other psychopathology (ie, passive gene-environment correlation).^2,20,21 Indeed, there is reason to suspect that familial factors may contribute to associations between maternal SDP and offspring SMI, as the association between an individual’s own smoking and risk of psychosis is partially explained by shared familial risk.²² Numerous sibling comparison studies have demonstrated substantial familial confounding of associations between SDP and offspring outcomes, including intellectual performance, academic achievement, externalizing psychopathology, and criminality,^1,23-28 with few exceptions.^29-32 This confounding appears to result from shared genetic influences in mothers and offspring.^33,34 Thus, there is a clear need for well-powered studies that can more fully account for familial confounding in the association between SDP and offspring SMI.

The present study used population-level data and family-based comparisons of cousins and siblings to examine the association between SDP and multiple indices of offspring SMI.^35-37 Sibling comparisons, in particular, rule out all genetic and environmental influences that make siblings similar to one another, producing a strong test of a teratogenic hypothesis. To our knowledge, this study is the first such family-based examination of SDP and offspring SMI.

We analyzed data through December 31, 2013, from single-birth offspring born in Sweden from January 1, 1983, to December 31, 2001. As described previously,^24,25 these data were drawn from the following merged Swedish population registers: (1) Medical Birth Register (live births and antenatal care visits in Sweden since 1973),³⁸ (2) Multi-Generation Register (familial relationships for individuals born since 1932 and living in Sweden since 1961),³⁹ (3) National Patient Register (nationwide inpatient psychiatric hospital admissions since 1987 and outpatient specialist visits since 2001),⁴⁰ (4) National Crime Register (criminal convictions for those aged 15 years or older since 1973),⁴¹ (5) Education Register (highest level of formal education since 1990),⁴² (6) Cause of Death Register (dates and causes of death since 1961),⁴³ and (7) Migration Register (dates of emigration out of Sweden since 1915).⁴⁴

Of the 1 878 371 single births during the included period, 1 809 183 (96.3%) offspring were still living in Sweden at age 12 years and were therefore included in this study. We sequentially excluded offspring missing data on paternity (n = 9532), SDP (n = 110 101), paternal (n = 7002) or maternal (n = 2280) educational level, or paternal (n = 35) or maternal (n = 14) nationality, resulting in a complete case cohort of 1 680 219 offspring (89.5% of the target population of single births). This study was approved by the Indiana University Institutional Review Board and the Regional Ethical Review Board in Stockholm.

From 1983 onward, mothers reported their daily cigarette smoking quantity at the first antenatal visit, typically in the first trimester. We contrasted moderate SDP (1-9 cigarettes per day) and high SDP (≥10 cigarettes per day) with no reported smoking (Table 1). The antenatal SDP variable has been used extensively in research and has demonstrated expected associations with multiple maternal³⁴ and offspring^24,25 behavioral outcomes. In addition, in a validity study of antenatal visits, only 6% of women who denied smoking had cotinine levels consistent with active smoking.⁴⁵

We evaluated associations between SDP and first inpatient or outpatient SMI diagnosis in offspring, requiring offspring to be at least age 12 years to receive a diagnosis. We defined SMI diagnoses with International Classification of Diseases, Ninth Revision, and International Statistical Classification of Diseases and Related Health Problems, Tenth Revision, codes for bipolar disorder and schizophrenia spectrum disorders, which we adapted from prior Swedish register–based research (eTable 1 in the Supplement).^46,47 The probability of any SMI diagnosis by age 25 years was 1.3% (Table 1). Because comorbidity with substance use disorder (SUD) appears to explain much of the risk of violence among patients with SMI,²¹ we also identified patients with SMI and SUD, with onset defined as the first diagnosis of either disorder.

For each offspring, we included covariates for sex, parity, and maternal and (biological) paternal ages at childbirth. We additionally included covariates for maternal and paternal nationality and the highest level of completed education, as well as any registrations for criminal conviction, hospitalization for SMI, hospitalization for suicidal behavior, and hospitalization for SUD (Table 1).

Follow-up person-time at risk for SMI varied across offspring and was right-censored by the earliest occurrence of death, first emigration out of Sweden, or the end of the study. To account for this censoring, we used survival analysis to evaluate associations between SDP and first SMI diagnosis in offspring, with offspring age as the timescale.⁴⁸

First, we plotted cumulative risk (ie, 1 − Kaplan-Meier estimates) for SMI conditional on SDP exposure in (1) all offspring and (2) the subset of siblings discordant for SDP. Second, we estimated hazard ratios (HRs) using 4 Cox proportional hazards regressions. Model 1 (population) examined population associations, with offspring sex and parity covariates and robust SEs to account for clustering of offspring born to the same mother. Model 2 (adjusted) additionally adjusted for maternal and paternal ages at childbirth and the other measured parental covariates.

Model 3 (cousin) and model 4 (sibling) compared discordant cousins or siblings in the same family, independent of all influences that make family members similar to each other. The sibling comparisons, in particular, enabled strong tests of a causal teratogenic SDP hypothesis.^35-37 If an observed association persisted in model 4, the result would be consistent with the hypothesis that SDP increases the risk of offspring SMI. In contrast, to the extent that the association was attenuated, the result would fail to support this causal hypothesis and would instead suggest that the population association resulted from familial confounding.

We estimated sibling and cousin associations using fixed-effects (stratified Cox) regression.⁴⁹ The sibling models were stratified on mothers. These models allowed the baseline hazard function to differ across groups of offspring born to distinct mothers. We also included measured covariates that could vary within families. The complete case cohort included a total of 1 245 299 siblings born to 536 772 distinct mothers. Of these families, 7201 (1.3%) were discordant on SMI. Among these, 1501 families (20.8%) with 3615 offspring varied in SDP exposure (no, moderate, or high smoking).

The cousin models included all possible nonsibling pairs of offspring within a maternal-grandmother–descended extended family (ie, all pairs of offspring of sisters). These models were stratified on cousin pairs, with robust SEs to adjust for nonindependence introduced by including individual cousins in multiple cousin pairs. The cousin comparisons included 612 191 offspring in 746 063 pairs of cousins who descended from 152 887 distinct maternal grandmothers. Of these cousin pairs, 7697 (1.0%) were discordant on SMI, among whom 3140 (40.8%; descended from 1799 maternal grandmothers) were discordant on SDP (eTable 2 in the Supplement provides exposure and outcome discordant samples for the other outcomes).

Because of their reliance on within-family discordance, sibling comparisons can be subject to bias due to carryover effects from 1 sibling to another (eg, if an offspring’s SDP exposure influences siblings’ SMI risk), measurement error, or unmeasured within-family confounding.^36,50 Sibling comparisons may also have reduced power or generalizability. We used 6 alternative approaches to examine the validity of the sibling comparison associations. First, we included cousin comparisons. Although they rule out less familial confounding by design (because of differences among sisters in the maternal generation), cousin comparisons are less likely to be biased by carryover effects. Cousins are also more likely to be differentially exposed, which increases power and generalizability. Second, to further examine generalizability, we estimated population associations in families with multiple vs single included offspring (the latter are excluded from sibling comparisons because they cannot provide information). Third, to evaluate any possible effect of decreases over time in the prevalence of SDP,⁵¹ we stratified population analyses by time. Fourth, to further examine associations independent of carryover and birth-order effects, we compared first-born cousins. Fifth, to examine the possibility that low power or SDP measurement error resulted in underestimated within-family associations, we conducted logistic regressions predicting birth outcomes generally understood to be causally affected by SDP (ie, small for gestational age [birthweight >2 SDs below the gestational age–specific mean] and preterm birth [gestational age <37 weeks]).^1,2,23,51 These models additionally controlled for year of birth. Finally, to examine the potential effect of missing SDP data, we conducted analyses assuming that all missing values were either high SDP or no SDP. We analyzed data using SAS, version 9.4 (SAS Institute Inc), and Stata, version 13.1 (StataCorp).

Offspring with any SDP exposure had a greater risk of any SMI than did offspring without SDP exposure (Figure 1). In contrast, the difference in risks was greatly attenuated when we limited the Kaplan-Meier estimates to discordantly exposed siblings, which was inconsistent with a teratogenic hypothesis.

Cox proportional hazards regressions confirmed this pattern (Table 2 and Figure 2). At the population level, offspring with moderate SDP had a 25% greater rate of any SMI than did offspring with no SDP (HR, 1.25; 95% CI, 1.19-1.30), and offspring with high SDP had a 51% greater rate (HR, 1.51; 95% CI, 1.44-1.59). Adjusting for confounding via measured parental covariates began to attenuate these associations (moderate SDP: HR, 1.12; 95% CI, 1.07-1.17; high SDP: HR, 1.27, 95% CI; 1.20-1.33), whereas the cousin comparison did not substantially attenuate the associations further (moderate SDP: HR, 1.13; 95% CI, 1.04-1.24; high SDP: HR, 1.23; 95% CI, 1.11-1.36). Sibling comparison associations were weaker still and were not statistically significant (moderate SDP: HR, 1.09; 95% CI, 0.94-1.26; high SDP: HR, 1.14; 95% CI, 0.96-1.35).

Disorder-specific results are reported in Table 2 and Figure 2 (eFigure in the Supplement presents disorder-specific Kaplan-Meier estimates). For bipolar disorder, the pattern of associations was similar to that of any SMI. At the population level, moderate SDP was associated with a 29% greater rate of offspring bipolar disorder compared with no SDP, and high SDP was associated with a 54% greater rate. However, in the sibling comparison, the moderate (6% greater rate) and high (15% greater rate) SDP associations were greatly attenuated and not statistically significant. For offspring schizophrenia spectrum disorders, most of the association could be attributed to confounding from measured parental covariates in the adjusted model, and the sibling comparison associations were again not statistically significant. For comorbid SMI and SUD in offspring, the population association was stronger (eg, a greater rate by a factor of 233% for high SDP), as was the attenuation across models, such that high SDP was associated with only a 5% greater—and not statistically significant—rate in the sibling comparison. We found a similar pattern of results for any offspring SUD diagnosis (eTable 3 in the Supplement).

Most sensitivity analyses explored the association between SDP and any offspring SMI (eTable 3 and eTable 4 in the Supplement). First, as described above, cousin comparisons were consistent in suggesting that much of the association was attributable to familial confounding. Second, population associations were modestly but not substantively larger in multiple-offspring compared with single-included-offspring families, indicating that the pattern of results in Table 2 could not be explained by the exclusion of single-offspring families from the sibling comparisons. Third, the population associations were larger in later birth years. Although this difference was not consistent with a teratogenic hypothesis, it suggested that the processes underlying the associations may be changing. Fourth, first-born cousin fixed-effects comparisons showed attenuated and nonsignificant associations independent of carryover or birth-order effects. Fifth, inconsistent with the possibility that low power or SDP measurement error explained the attenuated SMI associations, we found the expected offspring-birth-outcome sibling-comparison associations between SDP and small for gestational age (moderate SDP: odds ratio [OR], 1.63; 95% CI, 1.50-1.77; high SDP: OR, 1.83; 95% CI, 1.66-2.02) and preterm birth (moderate SDP: OR, 1.13; 95% CI, 1.06-1.20; high SDP: OR, 1.22; 95% CI, 1.13-1.32). Finally, results did not meaningfully differ regardless of the coding of missing SDP values.

The need to identify environmental risk factors for SMI that are strongly associated, modifiable, and, most important, causal has generated interest in findings of associations between maternal SDP and offspring bipolar disorder and schizophrenia.^52,53 Using data from nearly 1.7 million offspring, the present study replicated these associations at the population level.^4,5 Consistent with prior population-based studies, children of mothers who smoked heavily during pregnancy had approximately 1.5 times the rate of developing bipolar disorder or schizophrenia spectrum disorders as did children of mothers who did not smoke during pregnancy.

Critically, however, adjustment for measured parental covariates and unmeasured familial confounding largely attenuated these associations. Indeed, for each SMI outcome, sibling comparisons with within-family covariates yielded small, nonsignificant differences between offspring discordant for exposure to SDP. Some^4,5—but not all¹⁹—recent studies have found associations between SDP and an increased risk of offspring SMI even after controlling for measured parental psychiatric factors and other potential confounders, as did most of our adjusted models. However, we found that associations attenuated even further in sibling comparisons. This pattern highlights the value of family-based approaches that use relatedness to rule out passive gene-environment correlation and other familial factors rather than relying on measured covariates that may not be assessed or modeled comprehensively.³⁶

Moreover, additional strengths of the present study increase confidence in the results. The sample was population based, including 89.5% of single births in Sweden from 1983 to 2001 and 14 215 offspring with SMI diagnoses. Data on SDP and SMI were collected prospectively and from differing sources (maternal self-report and clinical diagnoses, respectively). Sibling and cousin comparisons and other sensitivity analyses supported the overall findings independent of the assumptions and limitations specific to each approach (eg, decreased power and generalizability in sibling comparisons). Taken in sum, these data provide strong evidence that much of the population association between SDP and offspring SMI results from shared risk for both outcomes rather than from a true teratogenic effect. If a causal effect of SDP exposure on offspring SMI risk exists, our results suggest that it is likely to be substantially weaker than would be expected based on the population associations alone.

One question unanswered by the present results is the exact nature of the familial confounding. Prior quantitative genetic studies,^33,34 as well as in vitro fertilization–based studies of genetically unrelated mother-offspring pairs,^54,55 point to genetic confounding, with an adoption study providing evidence against postnatal environmental confounding.⁵⁶ At the same time, there is emerging evidence that associations between smoking and SMI also reflect multiple environmentally mediated pathways.^22,52 Indeed, we found that population associations between SDP and offspring SMI strengthened over time. One possible explanation for this change is that, as smoking has become less prevalent, women who smoked—and smoked during pregnancy—more recently were those with the greatest liability for substance abuse and other psychopathology.⁵⁷ As a result, the extent of confounding from familial psychopathology liability may have expanded. Research is needed to characterize the sources of confounding observed here.

Sibling comparisons have several limitations that we could not fully address. First, they are susceptible to confounding from unmeasured factors that make siblings different from one another. The within-family associations may, therefore, overestimate or underestimate the true SDP effect. However, to explain the weak and nonsignificant sibling differences found here, unmeasured within-family confounders would have to be positively associated with SDP and negatively associated with offspring SMI (or vice versa). Nevertheless, evidence from other advanced observational approaches with complementary strengths (eg, instrumental variable analyses, in vitro fertilization–based designs, negative controls)^15,20,58 is needed to strengthen conclusions from this study.⁵⁹ Second, the maternal report assessment of SDP occurred at a single visit, relied on self-report, and did not permit us to explore details of timing, fine-grained smoking frequency, or quitting smoking. Measurement error and, especially, limited phenotyping, have been raised as potential challenges for register-based approaches.⁶⁰ However, previous studies have supported the validity of the SDP measure, and we found expected within-family associations with pregnancy-related outcomes, suggesting that measurement error alone cannot account for the main findings. Third, we analyzed a sample of Swedish offspring that included SMI onsets through the first 3 decades of life. We do not know the extent to which our results will generalize to other populations or to later-onset SMI. Finally, our analyses cannot determine whether SDP is more strongly associated with offspring SMI in specific subgroups.

There is extensive and clear evidence that exposure to SDP—in addition to smoking itself—has severe adverse consequences.⁵² This study did not, however, support a causal teratogenic hypothesis specific to offspring SMI, which matches prior family-based studies of SDP and a range of other offspring cognitive, behavioral, and psychiatric outcomes.^1,2 Within-family and other advanced designs appear to hold promise for the continued evaluation of environmental risk factors for SMI.

Accepted for Publication: February 18, 2017.

Corresponding Author: Patrick D. Quinn, PhD, Department of Psychological and Brain Sciences, Indiana University, 1101 E 10th St, Bloomington, IN 47405 (quinnp@indiana.edu).

Published Online: May 3, 2017. doi:10.1001/jamapsychiatry.2017.0456

Author Contributions: Dr Rickert had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Quinn, Rickert, Weibull, Johansson, Lichtenstein, Larsson, D'Onofrio.

Acquisition, analysis, or interpretation of data: Quinn, Rickert, Lichtenstein, Almqvist, Larsson, Iliadou, D'Onofrio.

Drafting of the manuscript: Quinn.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Rickert.

Obtained funding: Quinn, Lichtenstein, Almqvist, Larsson, D'Onofrio.

Administrative, technical, or material support: Lichtenstein, Iliadou.

Supervision: Larsson, D'Onofrio.

Conflict of Interest Disclosures: Dr Lichtenstein has served as a speaker for Medice and Dr Larsson has served as a speaker for Eli Lilly and Shire and has received a research grant from Shire, all outside the submitted work. No other disclosures were reported.

Funding/Support: This project was supported by National Institute of Mental Health grant R01MH102221; National Institute on Drug Abuse grant K99DA040727; the Indiana Clinical and Translational Sciences Institute: Pediatric Project Development Team; the Swedish Research Council (2013-2280, 2011-2492, and, through the Swedish Initiative for Research on Microdata in the Social and Medical Sciences framework, 340-2013-5867); and the Swedish Research Council for Health, Working Life, and Welfare (2012-1678).

Role of the Funder/Sponsor: The funding organizations had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Additional Contributions: Kyle Gerst, MS (Indiana University, Bloomington), assisted with statistical analyses. There was no financial compensation.