Study participants

We used data from the smoking, nicotine and pregnancy (SNAP) trial, a placebo randomised-controlled trial of NRT for smoking cessation in pregnant women.14 The trial recruited 1050 participants aged 16 to 45 years between 12 and 24 weeks gestation who smoked at least 10 cigarettes a day before pregnancy and at least five cigarettes a day during pregnancy, with exhaled carbon monoxide (CO) readings of ≥8 ppm. Participants were randomised to receive up to an 8 week course of either 15 mg per 16 hours of NRT patches or visually-identical placebo. For safety reasons, women were instructed that if they smoked at all while using patches, the NRT should be stopped. Participants set a target quit date and were followed up 1 month after this, during hospital admission for delivery or as soon as possible afterwards and at 24 months after delivery.

Maternal baseline data and smoking behaviour measures

Baseline data included maternal date of birth, age on leaving full-time education, number of cigarettes smoked per day prior to and during pregnancy, Body Mass Index (BMI), the Heaviness of Smoking Index (HSI) - a measure of nicotine dependence which is a six-point scale derived from the time to first cigarette after waking and the number of cigarettes smoked daily.16 Other data included were previous use of NRT in the current pregnancy and partners’ smoking status.

Self-reported smoking data were obtained 1 month after the target quit date and at delivery. Self-reported smoking abstinence at 1 month after the quit date was validated by exhaled CO measurements below 8 ppm while abstinence at delivery was validated by CO measurements and saliva cotinine concentrations below 10 ng/mL. Carbon monoxide is exhaled from the breath after smoking cigarettes and was used to confirm abstinence within the preceding 24 hours. Salivary cotinine on the other hand was used to determine smoking exposure within the last 7 days.17 No data on tobacco smoke exposure were collected between the 1 month follow-up date and delivery. Self-reported data on the use of nicotine patches were collected 1 month after the target quit date and at delivery; adherence to NRT have been reported elsewhere.14

Infant outcome measures

Birth status (live or stillbirth), gestational age at birth, birth weight, singleton or multiple birth, sex and gestational age at birth were recorded at delivery and have been reported elsewhere.14 The presence of infant developmental impairment at 24 months was determined using questionnaire responses from participants or from healthcare professionals when there was no participant response; full details have been described previously.15 Participants reported on their infants’ development as assessed within the five domains of the Ages and Stages Questionnaire, Third Edition (ASQ-3): communication, gross motor, fine motor, problem solving and personal-social skills.15 Participant questionnaires (PQ2) were posted out at 24 months and if no response was received, health professional ones were dispatched to non-respondents’ family physicians. Health professional questionnaire (HPQ) items were consistent with domains in participant questionnaires and were designed to be completed with reference to medical or health visitor records. Health professionals completing these questionnaires required relatively little knowledge of the patient. However if they were unable to complete the questionnaire, they were asked to forward these to children’s’ health visitors.18 Scores from the five ASQ-3 domains and responses to ‘non-domain’ ASQ-3 items in the participant questionnaires were used with established norms to categorise infants as having ‘developmental impairment’ or ‘no developmental impairment’. Infants with only HPQ responses were considered to have no developmental impairment when the responses from all questions indicated no potential developmental issues.15 As shown in the main study, 88% of study participants returned a PQ2 or HPQ. The proportion of participants assessed using the PQ2 or HPQ were similar in the treatment arm and placebo arm. Approximately two-thirds of participants completed and returned the PQ2, while the HPQ was the source of data for children of the remaining participants.

Measures of tobacco smoke exposure

As outlined above, smoking behaviour measures were collected at randomisation, 1 month after a target quit date and at the end of pregnancy. From these data, we derived the following categories to represent different levels of tobacco smoke exposure during pregnancy: (i) Number of cigarettes smoked daily during pregnancy, at randomisation, (ii) HSI at randomisation, categorised as high (4 to 6) or low (0 to 3), (iii) Women who reported not smoking between their target quit date and 1 month, validated using exhaled CO and (iv) Women who reported not smoking between their target quit date and the end of pregnancy or shortly afterwards, validated by either or both of exhaled CO and saliva cotinine at the end of pregnancy and also by exhaled CO at 1 month after their quit date.

As all trial participants smoked at the time of randomisation, exposure category (ii) was intended to dichotomise those who smoked more heavily and less heavily in the first weeks of pregnancy prior to study randomisation. For categories (iii) and (iv), we hypothesised that those demonstrating validated abstinence at both 1 month and the end of pregnancy (category iv) would have a lower overall tobacco smoke exposure than those for whom abstinence was only validated at 1 month after their quit date. However, it should be noted that no biochemical test can validate smoking cessation for a prolonged period and some women in either group may have smoked for a time between baseline/study enrolment and 1 month or the end of pregnancy.

Analyses

Maternal and infant characteristics were analysed by study arm, for participants who had singleton live births and 2 year data on infant developmental impairment. Summary statistics were represented as number (%), mean (SD) and median (IQR) for categorical, normal continuous and non-normal continuous variables, respectively. Univariable logistic regression investigated the association between 2 year developmental outcomes and different measures of maternal tobacco smoke exposure such as baseline number of cigarettes smoked daily in pregnancy, the heaviness of smoking index during pregnancy, maternal smoking abstinence at some point during pregnancy (measured 1 month after the participants’ set quit date) and maternal smoking abstinence during the latter stages of pregnancy (measured at delivery). In multivariable analyses, we adjusted the univariate models for baseline characteristic variables and also controlled for variables that could confound the association between maternal smoking and infant developmental impairment. Potential confounders included in the multivariable analyses were: maternal age, age at which full-time education was completed, maternal BMI, partner’s smoking status, infant birth weight, gestational age at birth and study arm (NRT or placebo). Confounder selection was done using the change-in-estimate criteria,19 and any covariate which changed the effect size of the univariate exposure-outcome model by 10% was considered an important confounder and included in the fully-adjusted model. All analyses were performed using Stata V.14. Significant associations were defined at the p≤0.05 significance level.

Patient and public involvement

The original SNAP study had a patient and public involvement (PPI) representative who provided PPI insights for the Trial Steering Committee.

Maternal and infant characteristics

The baseline characteristics of the women and children included in the analyses were found to be similar in participants randomised to the NRT or placebo study arm (presented in table 1). The majority of participants included in the study (72.4%) were less than 30 years old; the median BMI (IQR) at randomisation into the study was 25.9 (22.2 to 30.7). The average age of leaving full-time education was 16.3 years. We used the ‘Heaviness of Smoking Index’ as a measure of the level of maternal nicotine dependence and 65% of women in the study had low levels of nicotine dependence prior to being recruited in the study. More than two-thirds of the study participants had partners who smoked (67.9%). Overall, 18% of women were validated as abstinent from smoking at some point during pregnancy (measured 1 month after their target quit date) and 8.9% remained abstinent at the latter stages of pregnancy until delivery. There was a statistically significant difference in smoking cessation outcomes between participants in the NRT and placebo arms at the 1 month follow-up but not at delivery.

Analyses of the associations between different measures of maternal tobacco smoke exposure and infant developmental impairment found no statistically significant relationship between maternal smoking exposure and infant developmental impairment at age 2 (table 2). The ORs for developmental impairment in infants of mothers with validated smoking cessation at some point during pregnancy (measured at the 1 month follow-up) and in infants of mothers with validation cessation at delivery, were 0.99 (95% CI 0.59 to 1.66) and 1.55 (95% CI 0.84 to 2.87), respectively. Adjustment for maternal and infant baseline characteristics and covariates did not significantly alter the association between maternal smoking and infant development. Table 2 shows the unadjusted ORs with the different measures of cigarette smoke exposure, as well as the ORs adjusted by maternal and infant baseline characteristics.

Summary of principal findings

We found no associations between maternal smoking status at different points during pregnancy and infants’ developmental impairment at 2 years. Consequently, there is no evidence to support the hypothesis that the better infant development observed within infants born to women who were randomised to NRT in the SNAP trial resulted from smoking cessation induced by nicotine patch use.

Strengths and limitations

Data for this study are from the largest randomised controlled trial of NRT in pregnancy in which participants, healthcare and research staff were masked to treatment allocation.14 Follow-up and outcome ascertainment rates were high and there were only minor differences in characteristics of participants lost to follow-up and those remaining in the study at 2 years.15 It remains the only trial of a smoking cessation intervention in pregnancy to monitor infant outcomes; as such it is the only source of data which could be used to investigate the study hypothesis.

The ASQ-3 is a well-validated screening tool which has a sensitivity of 100% and specificity of 93% for detecting severe developmental delay at 24 months20. While some study infants may have been falsely identified as developmentally impaired, there is no reason to believe that such ASQ-3 false-positives would have occurred more frequently in infants of mothers who smoked in pregnancy. While developmental impairment was assessed in some infants using the participant questionnaire, others had the health professional questionnaire as the main source of data on impairment. Although this consequently implies a non-standardised assessment of the outcome in all infants, findings from the main study had shown that the proportion of participants assessed with either the PQ2 or the HPQ were similar in the treatment and placebo arms. Also, there were similarities in the baseline characteristics and birth outcomes of participants who returned the PQ2 and those for whom the HPQs were completed.

Although we adjusted for some potentially confounding factors which are known to be associated with infant development such as maternal socioeconomic status,21 maternal education,22 low birth weight,23 maternal exposure to passive smoking24 and maternal obesity,25 we were unable to adjust for those on which we had no data, such as maternal nutrition,26 depressive illness,27 stress and anxiety28 and alcohol consumption29 which could also potentially influence infant development but which were not measured in the SNAP trial. There remains therefore, the potential for residual confounding.

We undertook secondary analyses of randomised controlled trial data and the original sample size was intended to have sufficient power only for detecting differences in validated smoking cessation rates at the end of pregnancy. Analyses presented here may not have sufficient power to demonstrate associations between smoking and infant development; however, one would expect to find one or more such associations if, the relationship between NRT use and infant development which was demonstrated within the same data, were principally explained by smoking cessation caused by NRT.

There may have been some misattribution of smoking status. Measures of smoking status used were biochemically-validated at two different time points in pregnancy. Validation with carbon monoxide at the earlier time point could only eliminate smoking within the previous 24 hours30; and at delivery, with saliva could only do so for the previous 7 days.17 Consequently, some participants may have relapsed or smoked occasionally between these times31; however we believe it is reasonable to assume that women who were validated as not smoking both at the end of pregnancy and at 1 month after randomisation would have lower overall tobacco smoke exposure than those who were not. As recruitment into the study occurred between 12 and 24 weeks gestation, the timing of smoking status measurements in relation to gestation will have differed between participants but there is no evidence that smoking is more or less harmful at any point in pregnancy. Although these were the best available data for analyses, the assumptions outlined above need to be considered when interpreting study findings.

As most trial participants continued to smoke, analyses might have been more illuminating if data on the intensity of women’s smoking during pregnancy had been available. For example, if exhaled CO measurements or numbers of cigarettes smoked daily in pregnancy had been available for non-abstinent trial participants, these data could have been used to investigate whether smoking intensity in pregnancy was associated infant development. Increasing smoking intensity, reflected by exhaled CO levels, is associated with reduced infant birth weights,32 so an association with development seems possible.

Our analyses did not incorporate postnatal environmental cigarette smoke exposure which may have an independent influence on infant development.33 We know that there is continued maturation of the connections between brain regions after birth, with intense brain white matter myelination in the first postnatal months which becomes progressively less rapid through toddlerhood until young adulthood.34 Additionally, postnatal exposure to environmental tobacco is associated with emotional and conduct behavioural problems in school aged children35 and so may have an aetiological role in infant developmental impairment. As maternal prenatal and postnatal smoking are strongly correlated, this would increase the likelihood of finding associations between smoking in pregnancy and infant development; as none were found there is no suggestion of a large impact on findings. Nonetheless, there is a need to account for environmental cigarette smoke exposure in future studies.

Comparison with literature

As this is an original analysis in a unique trial database there is very little literature with which study findings can be compared. Previous observational studies have found prenatal tobacco exposure to be associated with offspring adverse consequences including behavioural36 and hyperactivity37 38 problems, language and reading deficits,37 conduct disorders,39 newborn basic perceptual skills40 and intellectual impairment in childhood41 and later life.42 However, other studies which adjusted for potentially confounding factors such as maternal education43 44 and parental socioeconomic status45 have not. Consistent with these latter studies, our findings found no evidence that the improved developmental outcomes in infants of pregnant smokers who used NRT in the SNAP trial15 were explained by smoking cessation. As preclinical studies show that nicotine is neurotoxic and can adversely affect the developing central nervous system,11 12 it seems unlikely that these better outcomes are due to a direct protective effect of NRT on the developing foetus. Also, compliance rates in the NRT patch and placebo study arms were low in the SNAP trial (7.2% and 2.8% compliance, respectively, at 1 month). Perhaps, it is possible that nicotine used in pregnancy has no impact on infants’ developmental outcomes and the apparently protective effect of NRT on developmental outcomes observed in the SNAP trial occurred by chance.