Introduction

Prenatal smoking has been found to increase the risk of attention-deficit/hyperactivity disorder (ADHD), oppositional defiant disorder (ODD), conduct disorder (CD), poor cognitive functioning, antisocial problems, aggression, delinquency, substance abuse, and internalising problems [1–7]. The majority of studies on this topic have examined children during mid or late childhood. Children with an early manifestation of disruptive behaviours have been found to develop more serious long-term psychopathologies. For instance, approximately one-quarter of children with ODD later develop conduct disorder (CD), and a few of these children develop antisocial personality disorder in adulthood [8, 9]. The short- and long-term costs of these problems are grave not only for the patients and their families, but also for society at large. Therefore, it is important to establish whether prenatal smoking affects early development. At present, only seven studies have examined the effect of prenatal smoking on preschoolers [5, 10–15]. Six of these studies used various rating scales to measure the effects of prenatal smoking on symptoms related to externalising problems. The drawback of this technique is that high scores on rating scales do not necessarily translate into disorders. One study [15] included a structured diagnostic interview, but the final analyses were performed on a composite of the results from rating scales and symptom counts derived from this diagnostic interview. At present, it is unknown whether prenatal smoking merely increases the risk for elevated levels of externalising symptoms or if it also increases the risk of diagnosable disorders. Thus, the current study addressed whether prenatal smoking increases the risk for disorders in a large community of 4-year-olds using a state-of-the-art semi-structured diagnostic interview.

The neurotoxicological mechanisms linking prenatal smoking to behavioural disorders are not fully understood. However, prenatal smoking has been found to affect the foetus by stimulating nicotinic acetylcholine receptors (nAChRs) in the developing brain, thereby altering axonal and dendritic branching in several brain regions [16]. Studies have found that prenatal nicotine exposure can create developmental disturbances in the serotonergic and dopaminergic systems [17], which are implicated in the regulation of emotion and behaviour.

Several prospective studies have demonstrated a dose–response effect of prenatal smoking on externalising behaviours [18, 19]. Such results support a causal interpretation of the association. Despite these findings, the causal relationship between maternal prenatal smoking and disruptive behaviours later exhibited by the children has been challenged because numerous factors could confound the association, including socioeconomic status, the use of other substances during pregnancy, and other maternal or paternal characteristics, such as education level, age, parental personality, ethnic background, and psychiatric history [20].

It has been suggested that mothers with antisocial traits, conduct disorder or ADHD are more likely to smoke, and therefore genes, rather than smoking per se, may cause the observed associations [19]. Several quasi-experimental studies have used innovative designs in order to attempt to disentangle the effects of prenatal conditions from maternal and offspring genomes [21], e.g., sibling designs, comparing offspring exposed to prenatal nicotine exposure to siblings who had not been prenatally exposed [22–24]. The results from these studies indicate that the risk of ADHD and ODD in offspring appears to become weaker when siblings are taken into account, indicating that the effects may partly be due to genetic or other unmeasured confounds. Furthermore, a recent study by Thapar et al. [25] on offspring conceived with assisted reproduction methods involving both biological and non-biological mothers also supported an effect of genes rather than smoking. Although it is well established that ADHD is highly heritable [26], an additional prenatal effect of smoking cannot be excluded [4]. The consideration of this topic in either/or terms is overly simplistic because it is probable that both genetic and teratological effects contribute to the association between prenatal smoking and behavioural disorders [27].

We have therefore controlled for a wide range of potential confounding variables in the current study. However, we have deliberately not controlled for various parenting behaviours as recent studies have suggested [15], as it does not seem logical or probable that parenting behaviours after birth would influence whether a mother smokes during pregnancy.

Most of the literature on maternal prenatal smoking and disruptive behaviour have focused on ADHD and behavioural problems in isolation [28]. Given the high comorbidity between ADHD and ODD, including among preschoolers [29], it is feasible that the association between prenatal smoking and ODD could result from the effect on ADHD (ODD route) or vice versa (ADHD route). Thus far, the literature provides conflicting results; in support of an ODD route, Nigg and Breslau [3] found that prenatal smoking predicted ODD when ADHD was controlled, but the reversed was not true. Similarly, Wakschlag et al. [28] found that prenatal exposure was associated with ODD as well as comorbid ADHD and ODD, but not with ADHD only. However, Nomura et al. [30]. reached the opposite conclusion, supporting an ADHD route. Specifically, mothers who smoked during pregnancy were found to have a significantly increased risk of having offspring with comorbid ADHD/ODD and ADHD but no increase in the risk of having offspring with ODD only. These latter results are compatible with the finding of Mick et al. [31] that prenatal smoking predicted ADHD even when CD had been controlled for. Moreover, a prenatal effect of smoking could potentially contribute to both ADHD and disruptive disorders via similar neuroteratological mechanisms, which would indicate a direct route to either disorder that is not mediated through the other disorder. Such a direct route was supported by the finding of Arnold et al. [32] that mothers who smoked during pregnancy were more likely to have children with ADHD and CD as well as comorbid ADHD and CD. Similar results were found by Button et al. [4] who reported that smoking had an independent effect on antisocial behaviour and symptoms of ADHD. Additional support for a direct route was provided by the finding of Hutchinson et al. [12] of a dual effect for boys (but not for girls) and the results of Biederman et al. [33] among siblings of twins (but not among the twins). A direct route is also supported by Huijbregts et al. [5], who found that prenatal smoking predicted co-occuring hyperactivity–impulsivity and physical aggression, which might be indicative of behavioural problems. Taken as a whole, the data in the literature appear to be conflicting. Furthermore, as ADHD and ODD/CD are infrequent in the general population, a lack of statistical power may produce inconsistent results. The data from the studies that supported a direct effect route tended to have the most statistical power because they involved the largest number of subjects.

If smoking during pregnancy has a dramatic impact on the developing brain and increases the risk for diagnostic disorders (i.e., ADHD and ODD), the detrimental effect should be expected to manifest by the preschool years. Therefore, it is hypothesised that (1) smoking during pregnancy increases the risk for ADHD and ODD when controlling for a wide range of potential confounding variables, and (2) prenatal smoking increases the risk of ADHD or ODD or both disorders.

Methods

Participants

The present study was conducted as part of the Trondheim Early Secure Study (TESS). All children living in Trondheim, Norway who were born in 2003–2004 and their parents were invited to participate in the study. This study was performed in collaboration with the community health clinics, which routinely invite all children for a health check-up when they reach 4 years of age.

The health clinic staff missed asking 166 families to participate. Parents with a lack of proficiency in Norwegian were excluded (n = 176). Of the 3,016 eligible parents, 2,475 (82.1 %) consented to participate in the study. The children were divided into four strata according to their scores on the Strengths and Difficulties Questionnaire (SDQ) [34], resulting in the following probabilities of selection into a further study: SDQ = 0–4, p = 0.37; SDQ = 5–8, p = 0.48; SDQ = 9–11, p = 0.70; SDQ = 12–40, p = 0.89. According to this procedure, a subsample of 1,250 parents was invited to participate in the study. Of these parents, 995 (79.5 %) completed the interview. The drop-out rate after providing consent at the health clinic did not differ across the four SDQ strata (Chi-squared = 5.70, df = 3, NS) or gender (Chi-squared = 0.23, df = 1, NS). The sample, adjusted for stratification, was compared with register information from Statistics Norway on all the parents of the 4-year-olds in Trondheim in 2007 and 2008. The sample contained significantly more divorced parents (7.6 %) than the general population (2.1 %), and the educational level of the sample group was virtually identical to that of the population. The mean age of the children was 53.0 months (range 46.3–63.0, SD = 2.1), and the mean number of siblings was 1.3 (SD = 1.0); 12.8 % of the children did not have siblings.

Procedure

The study was approved by the Regional Committee for Medical and Health Research Ethics. The parents received a letter of invitation with their scheduled appointment and the SDQ before attending the public health check-up for their 4-year-old child. The SDQ is a 31-item measure of mental health problems in children aged 4–18 years. The items relate to emotional symptoms, conduct problems, hyperactivity, and peer problems. At the health clinic, the nurse informed the parents about the study and obtained their consent to participate. The SDQ was scored manually at the health clinic, and a computer-based random number generator determined whether the family should be invited to participate in a further study.

After attending the health clinic and being selected for the study, one of the parents was interviewed at home, at work or at the University Clinic with a structured child-diagnostic interview. The same parent was then invited to attend the University Clinic with the 4-year-old for observation and testing. Approximately 20 % of the interviewed parents were fathers.

Measurements

Prenatal smoking

During the structured interview, the parent was asked a series of pregnancy-related questions related to alcohol use and smoking during pregnancy. If smoking was reported, the mother was asked the frequency of cigarettes she smoked during the first, second, or third trimester.

Birth weight

During the interview, the mothers were asked to report their offspring’s birth weight and gestational age.

Psychiatric diagnoses

The parents were interviewed using the electronic version of the Preschool Age Psychiatric Assessment (PAPA) [35], from which the DSM-IV diagnoses for ODD and ADHD as well as a range of anxiety disorders (i.e., social phobia, separation anxiety, generalised anxiety, and specific phobia) and depressive disorders (i.e., major depression, dysthymia, and depression not otherwise specified) were derived. Anxiety disorders and depressive disorders were combined. The PAPA is a semi-structured, interviewer-based diagnostic interview that is used to assess children between 2 and 6 years of age. The questions are based on the preceding 3-month period to promote accurate recall. Diagnoses were computed by computer algorithms for ADHD and ODD. ODD requires that the symptoms occur “often”. “Often” was defined post hoc as occurring within the highest 10 % in the current population, as determined by frequency counts [36]. Because the primary interest was whether prenatal smoking might cause symptoms of disorders and not necessarily the resulting impairment or distress, only the symptom criteria were applied.

The interviewers (n = 7) were trained by the developers of the PAPA. They all had at least a bachelor’s degree in relevant fields (i.e., psychologists, health nurses, social workers, day care teachers) and extensive prior experience with children and families. Regular meetings were held with master coders. In addition, the interviews were observed behind one-way mirrors to ensure adherence to the interview guide and to prevent rater drift. The main interview duration was 2.25 h. Blinded raters recoded 9 % of the interviews. The multivariate inter-rater reliabilities between pairs of raters [37] were as follows: ADHD k = 0.96 and ODD k = 0.89.

Confounders

Antisocial, schizoid, dependent, and borderline personality symptoms were assessed during the DSM-IV and ICD-10 Personality Questionnaire (DIP-Q), which is a true/false self-report questionnaire that is designed to measure all 10 DSM-IV personality disorders [38]. The DIP-Q has been shown to discriminate well between different samples [39] and has good validity [40]. The reliability coefficient theta [41] for different scales varied between Θ = 0.71 and Θ = 0.92. Parental depressive symptoms and anxiety symptoms were assessed using the Beck Depression Inventory-II (BDI-II; [42]) and Beck Anxiety Inventory, respectively. Alcohol problems were assessed using the Alcohol Use Disorders Identification Test (AUDIT) [43].

The parents’ level and type of occupation were coded according to the International Classifications of Occupations [44]. If the parents were living together, the parent with the highest-rated occupation was selected.

The interviewed parents were asked whether the targeted child had siblings (yes/no), whether they were of Norwegian origin, whether the birth of the target child was easy or difficult on a scale of 1 (very easy) to 5 (very difficult), and whether alcohol abuse occurred during the pregnancy on a scale from 0 (did not drink) to 13 (drank daily, including 6 + alcohol units at least once a week). Furthermore, the parents were asked whether the pregnancy had been stressful (yes/no), whether the pregnancy had made them feel depressed (yes/no) and whether they had been trying to become pregnant (yes/no). The parents were also asked about ever having had a mental health problem that resulted in a major breakdown in important social roles (yes/no) and whether they ever (1) sought psychiatric treatment (yes/no), (2) presently had drug/alcohol problems (yes/no), (3) had ever been arrested (yes/no), (4) had ever had alcohol/drug problems or both (yes/no), (5) had ever used medication for mental health problems (yes/no), or (6) had ever been admitted to a hospital for such problems (yes/no). Finally, the parents were asked about their capacity to pay their bills on time (very capable/quite capable/poor) (Table 1).

Table 1 Demographics
Full size table

Statistical analysis

The odds of the children having psychiatric disorders, according to the smoking status of the mother, were examined using a logistic regression. The number of ADHD symptoms and the number of ODD symptoms were analysed using linear regression. Because we oversampled children with mental health problems, the data were weighted back with the inverse of the drawing probabilities to yield true population estimates and the Huber–White sandwich estimator was used to provide robust standard errors. We used propensity score analyses to adjust for potential confounders. This was done in two steps: first, a propensity score logistic regression model was constructed with the exposure (smoking) as dependent variable, and selected potential confounders as covariates. The predicted probability of exposure is the propensity score. Second, the propensity score was used as a covariate together with the exposure variable in the final analysis. We used the quintiles of the propensity score as a categorical variable in the main analysis. This typically removes 90 % of the bias [45]. In a secondary analysis, we used the propensity score as an untransformed scalar variable. We did not use propensity score matching because there is no straight forward method for matching in a weighted sample. These 24 variables were considered as potential confounders: narcissistic personality traits, histrionic personality traits, borderline personality traits, schizotypal personality traits, paranoid personality traits, avoidant personality traits, dependent personality traits, OCD personality traits, parental alcohol use, parental anxiety, parental depresion, alcohol use during pregnancy, stress during pregnancy, depression during pregnancy, planned pregnancy, parental feelings about pregnancy, mothers’ feelings in the first month after birth, parental experience of mental breakdown, parent requested medical treatment, parent ever been arrested, parent ever been indicted by police, parental ability to pay family expenses, parent received medical treatment for psychological disorder, and parental admission to a mental health institution. The selection of variables for the propensity score was done as follows: the variables mothers’ age, SES, and antisocial personality traits were included, since these are a priori known to be important candidate confounders. For the additional variables, we first selected those with p < 0.10 adjusted for the three-mentioned variables. Then, we used backwards elimination to include in the final propensity score model only those with p < 0.10. All p values are two-sided, and p < 0.05 was considered significant unless otherwise stated. Ninety-five percent confidence intervals (CI) are reported were relevant. The analyses were performed using SPSS 18 software.

Results

The estimated percentage of women in the population who smoked during pregnancy was 14.0 %. Of the 148 mothers who smoked during pregnancy, 69.6 % smoked less than 10 cigarettes per day, and 30.4 % smoked 10 or more cigarettes per day. A total of 34 children were diagnosed with ADHD, 57 children were diagnosed with ODD, and among these, 13 children had both ADHD and ODD.

The propensity score model for smoking during pregnancy included these ten potential confounders: borderline personality traits, parental alcohol use, parental anxiety, alcohol use during pregnancy, depression during pregnancy, planned pregnancy, mother’s feelings in the first month after birth, parent ever been arrested, parental ability to pay family expenses,parental admission to a mental health institution. In addition, the variables mother’s age, SES, and antisocial personality traits were included in the propensity score model. Table 2 shows a clear increase in proportion of smokers over the quintiles of the propensity score, indicating that this propensity score is a suitable summary measure for the confounders. The unadjusted OR of having a child with ADHD for mothers who smoked during pregnancy was 3.25 (CI 2.08–5.09, p < 0.001). After adjusting for quintiles of the propensity score as a categorical variable, the OR was 2.59 (CI 1.50–4.34, p < 0.001), and 2.17(CI 1.30–3.61, p = 0.003) when using the propensity score as a covariate. Further, the unadjusted OR of having a child with ODD for mothers who smoked during pregnancy was 3.12 (CI 2.30–4.24, p < 0.001) and 2.69 (CI 1.84–3.91, p = 0.02) when adjusting for quintiles of the propensity score as a categorical variable, and 2.46 (CI 1.66–3.63, p < 0.001) when using the propensity score as a covariate (Tables 2, 3).

Table 2 Propensity score quintiles and smoking during pregnancy
Full size table
Table 3 Odds ratio (OR estimate, CI and p values) psychiatric disorders, for children exposed to smoking any time during pregnancy
Full size table

To test whether the comorbidity between ADHD and ODD could explain the effect of smoking on either disorder, the children were divided into the following four groups: those with neither ADHD nor ODD, those with only ADHD (n = 21), those with only ODD (n = 44), and those with comorbid ODD and ADHD (n = 13). Smoking during pregnancy was predictive of comorbid ADHD and ODD. The unadjusted OR was 3.67 (1.82–7.40, p < 0.001). When adjusting for quintiles of the propensity score as a categorical variable, the OR was 2.55 (1.24–5.23, p < 0.001) and 2.68 (1.84 to 3.91,p < 0.001) when using the propensity score as a covariate.

The analyses were performed again using linear regression with the number of ADHD symptoms and the number of ODD symptoms as outcomes, and smoking during pregnancy as the exposure variable. Smoking during pregnancy predicted ADHD symptoms (B = 0.67, SE = 0.13, CI = 0.41–0.93, p = <0.001 (unadjusted). After adjusting for quintiles of the propensity score as a categorical variable (B = 0.50; SE = 0.15, CI = 0.20–0.80, p < 0.001) and (B = 0.40, SE = 0.14,CI = 0.12–0.68, p = 0.005) when using the propensity score as a covariate. Smoking during pregnancy predicted ODD symptoms (B = 0.36, SE = 0.07, CI = 0.22–0.49, p < 0.001 (unadjusted) and (B = 0.34, SE = 0.09, CI = 0.18–0.51,p < 0.001) when adjusted for quintiles of the propensity score as a categorical variable, and (B = 0.30, SE = 0.08, CI = 0.14–0.46, p < 0.001) when using the propensity score as a covariate (Table 4).

Table 4 Linear regression with number of ADHD or number of ODD symptoms as dependent variable, and smoking during pregnancy as covariate (B, CI, and p values)
Full size table

Discussion

The principal findings of the present study were that smoking during pregnancy predicted ADHD and ODD, and comorbid ADHD and ODD in preschoolers. The association beween smoking and ADHD/ODD could not be attributed to a wide range of potential confounders or to the comorbidity between the two disorders. Using propensity score analysis, we could include more potential confounders in the analysis than we could have done by adjustment for these confounders. The few studies that have investigated younger children have examined aspects of behavioural problems, for instance, physical aggression, hyperactivity/impulsivity [5], externalising behaviours [10, 14], or symptom counts of disruptive disorders [15]; diagnostic disorders have not previously been used as outcomes. However, the presence of a sufficient number of symptoms to qualify for a disorder, which was the outcome applied in the current study, is more likely to indicate the prospect of a seriously altered life trajectory than an elevated mean score of rating scale symptoms. For example, a typical rating scale item for ADHD is “Can’t sit still, restless or hyperactive”. The informant (in this case, the parent) must determine the definition of “restless”, the intensity of restlessness, how long the behaviour endures, and how often the behaviour occurs before characterising a child as “restless”. Parents are likely to differ in their evaluations of these symptoms. In an interviewer-based interview, the trained interviewers ask the informant for examples until they are satisfied that a child’s symptoms meet the criteria, and then they ask for the duration, frequency, and onset of the symptoms [46]. The present study applied a more stringent measure of diagnostic disorders, and yet, the effect of prenatal smoking on ADHD and ODD remained.

To date, the literature is inconclusive on whether the association between prenatal smoking and behavioural disorders is caused by genetic transmission or the teratological effect of nicotine during pregnancy. The present design cannot exclude the possibility that the observed association between smoking and disruptive disorders results from genetic factors. Studies on the effect of prenatal nicotine exposure on rodents and primates support a teratological effect of nicotine on the developing brain, especially the hippocampus and areas in the prefrontal cortex [47]. However, the possibility also exists that prenatal smoking may be a proxy for other environmental effects (e.g., low socio-economic status) [4]. The present study controlled for a wide range of such potential confounding variables, and the association remained. Recently, evidence has been provided for genetic transmission [25]; however, strong genetic effects do not necessarily conflict with environmental effects because of the possibility of gene–environment interactions [48, 49].

Low birth weight is a well-replicated outcome of prenatal smoking and has been associated with disruptive behavioural problems [3, 50], and associations between low birth weight and prenatal nicotine exposure have also been found in animal studies. Nonetheless, the current study did not find significant differences in the birth weights of nicotine-exposed children and non-exposed children. However, nicotine at doses that do not cause intrauterine growth retardation is still toxic to the developing brain [51]. Therefore, the level of smoking among mothers in the current study was likely insufficient to cause retarded growth but sufficient to cause an increased risk for disruptive disorders.

Limitations

Several limitations of the present study should be taken into account when evaluating the results. The main limitation of this study is the inability to provide a rigorous test of whether maternal smoking during pregnancy in fact causes psychiatric disorders in children. Because a randomly controlled study that assigns mothers to smoking and non-smoking groups cannot be performed, other methods must be applied. As mentioned previously, a recent study [25] investigated mothers who either smoked or did not smoke while undergoing assisted reproduction treatments as well as their biological and non-biological offspring. This study failed to identify an association between prenatal smoking and ADHD. Several recent studies [15], however, may have overcontrolled for a wide range of parenting behaviours that are unlikely to influence whether a mother smokes during pregnancy.

The current smoking measurement was too crude to allow for a dose–effect response. This study did not control for genetic confounding variables, e.g., the presence of ADHD in the mother. We were also unable to control for IQ in the parent or offspring as we did not measure this potential confounder in this study. Previous studies have shown mixed results on whether prenatal smoking affects cognitive development [13]. Information was only available for one parent; therefore, the personality traits and potential mental health problems of the other parent were unavailable. For instance, it may have been more difficult for the mother to discontinue smoking if the father smoked [52]. The heritable factors that increased the odds of the father smoking could also increase the odds of the disruptive behaviour in the child. Moreover, approximately 20 % of the parent responders were fathers who answered the questions on smoking during pregnancy on behalf of the mothers. The possibility therefore exists that these fathers were unable to remember the details of the mother’s prenatal smoking. The mothers may also have underreported their prenatal smoking because of the social stigma against smoking during pregnancy. The reporting of prenatal smoking data relied on retrospective recall 4 years after pregnancy, which is a potential weakness. However, a study by Pickett et al. [53] that compared the retrospective recall in relation to prospective reporting and biological measurements 10 years earlier found that women’s reports were accurate and reliable, particularly for second and third trimesters of pregnancy. This accuracy has not been investigated for the reports of fathers, however, and could therefore represent a source of error.

Although we used a semi-structured diagnostic interview, we applied only symptom criteria while excluding impairment criteria for ADHD and ODD. In addition, the questions were based on the preceding 3-month period rather than on the minimum 6-month period required by the DSM-IV. We did not control for passive smoking. Few studies have controlled for postnatal smoking. However, it seems unlikely that the effect of prenatal smoking would disappear after controlling for postnatal smoking. Direct nicotine exposure occurs in higher concentrations in the foetus than in the mother, and environmental exposure is weaker and less likely to affect brain development [2]. Furthermore, it is difficult to separate the effects of prenatal and postnatal exposure because they are highly correlated [13]. Nevertheless, a few studies have reported an effect of both prenatal and postnatal smoking, although prenatal smoking appears to be more influential [54]. Finally, it was the same interviewer asking for diagnoses as for prenatal smoking. However, both pieces of information are derived from the PAPA, which is a 2–3 h interview that obtains a great deal of background information, of which prenatal smoking represents a very small part. Furthermore, the PAPA includes a set of prescribed questions and probes, and thus, it is not likely that the knowledge of prenatal smoking status would have influenced the results, although this effect cannot be ruled out.

In summary, the present findings suggest that the disruptive symptoms found in preschoolers prenatally exposed to nicotine map onto psychiatric diagnoses at 4 years of age, even after numerous confounding variables are controlled for. Furthermore, prenatal smoking may be a common risk factor for both ADHD and ODD. Targeting younger women who smoke for preventive programs or offering smoking cessation therapy to pregnant women may reduce the risk of ADHD and ODD. Although a causal effect of smoking on behavioural disorders cannot be conclusively demonstrated by the current study, it is nonetheless ethical to offer interventions to at-risk women rather than waiting for further research. This is particularly the case because assisting young pregnant women to quit smoking has a wide range of other health benefits for both mothers and offspring [55].