Sample

This study examined data from 6356 children from the Avon Longitudinal Study of Parents and Children (ALSPAC) cohort who participated in the psychosis-like symptoms (PLIKS) semi-structured interview (PLIKSi) ^{Reference Horwood, Salvi, Thomas, Duffy, Gunnell and Hollis21} when they were 12 years of age (data restricted to one child per nuclear family). The initial cohort consisted of 14 062 children born to residents of the former Avon Health Authority area who had an expected date of delivery between 1 April 1991 and 31 December 1992 (www.alspac.bris.ac.uk). The cohort was set up to examine genetic and environmental determinants of health (including mental health) and development. ^{Reference Golding, Pembrey and Jones22} The parents have completed regular postal questionnaires concerning their child's health and development since birth. The children, since 7.5 years of age, have attended annual assessment clinics where they participate in a range of face-to-face interviews. Due to wave non-response, sample sizes in the analyses differ according to exposures and data-sets examined (see Results and Tables). Non-attendees participating in the PLIKSi were more likely to come from lower social class families, to have parents with lower education, to be male and to be of minority ethnic status. ^{Reference Horwood, Salvi, Thomas, Duffy, Gunnell and Hollis21}

Measures

Ethical approval

Ethical approval for the study was obtained from the ALSPAC Law and Ethics Committee and the local research ethics committees.

Statistical analysis

Logistic regression was used to calculate odds ratios (ORs) and 95% confidence intervals for associations between maternal substance use and PLIKS outcomes, both before and after adjustment for potential confounders. Our primary analyses for ordered categorical data (maternal tobacco and cannabis use) was to study a linear trend effect across increasing categories of use. In order to investigate a non-linear relationship for alcohol use, a quadratic term was used in addition to a linear term, and likelihood ratios tests (LRT) used to examine evidence of a non-linear relationship; and overall effect of alcohol use (linear and quadratic) on PLIKS outcomes.

Tobacco use during pregnancy

Maternal tobacco use during pregnancy was strongly associated with any suspected or definite PLIKS in the offspring (crude OR for linear trend across four smoking categories 1.33, 95% CI 1.18–1.49), and results were consistent with a dose–response effect (Table 2). This was attenuated only partially after adjusting for confounders (adjusted OR = 1.20, 95% CI 1.05–1.37, P = 0.007). The two confounders that had the greatest effect on attenuating this estimate were paternal smoking and single status of the mother. Further adjusting for gestation, birth weight, 5-minute Apgar score, or age 8 IQ score, as possible mediators for this association, had minimal effects on these results. These estimates were similar for definite PLIKS as an outcome although results were less precise.

Table 2

Crude and adjusted ^a odds ratios (OR) and 95% CI for psychosis-like symptoms (PLIKS) by maternal substance use during pregnancy

	n	Suspected or definite PLIKS Crude OR (95% CI)	Suspected or definite PLIKS Adjusted OR (95% CI)	Definite PLIKS Crude OR (95% CI)	Definite PLIKS Adjusted OR (95% CI)
Tobacco
None	3579	1	1	1	1
1-9 cigarettes/day	295	1.15 (0.79-1.66)	0.92 (0.63-1.36)	1.53 (0.92-2.53)	1.25 (0.73-2.14)
10-19 cigarettes/day	266	1.88 (1.35-2.61)	1.47 (1.02-2.12)	2.33 (1.48-3.66)	1.65 (0.99-2.75)
≥ 20 cigarettes/day	113	2.30 (1.45-3.65)	1.84 (1.12-3.03)	2.03 (1.01-4.10)	1.54 (0.73-3.25)
Linear trend	4253	1.33 (1.18-1.49), P < 0.001	1.20 (1.05-1.37), P = 0.007	1.39 (1.18-1.63), P < 0.001	1.21 (1.01-1.47), P = 0.047
Cannabis
None	4175	1	1	1	1
< 1/week	37	0.95 (0.34-2.70)	0.58 (0.20-1.70)	0.57 (0.08-4.20)	0.34 (0.04-2.56)
≥ 1/week	41	1.62 (0.71-3.66)	1.04 (0.45-2.43)	1.63 (0.50-5.32)	1.12 (0.33-3.84)
Linear trend	4253	1.22 (0.83-1.79), P = 0.317	0.94 (0.62-1.41), P = 0.755	1.16 (0.65-2.09), P = 0.616	0.91 (0.49-1.71), P = 0.776
Alcohol
None	2522	1	1	1	1
≤ 7 units/week	1293	0.92 (0.74-1.14)	0.92 (0.74-1.15)	0.67 (0.48-0.94)	0.68 (0.48-0.96)
8-21 units/week	410	1.05 (0.77-1.48)	1.00 (0.71-1.39)	0.58 (0.33-1.04)	0.56 (0.31-1.02)
≥ 22 units/week	28	2.58 (1.09-6.11)	2.40 (0.99-5.83)	2.14 (0.64-7.17)	1.86 (0.54-6.42)
Linear (per 10 units) b	4253	0.80 (0.51-1.25)	0.75 (0.47-1.19)	0.77 (0.48-1.25)	0.73 (0.45-1.18)
Quadratic (linear 2 ) b	4253	1.21 (1.00-1.47)	1.22 (1.00-1.49)	1.04 (0.97-1.12)	1.04 (0.97-1.12)
Likelihood ratio (for overall alcohol) b	4253	χ2 = 9.8, d.f. = 2, P = 0.008	χ2 = 8.3, d.f. = 2, P = 0.016	χ2 = 1.3, d.f. = 2, P = 0.522	χ2 = 1.8, d.f. = 2, P = 0.415

a. Adjusted for other substances used, and all variables in Table 1; data-set with no missing data for confounders 4253.

b. Results for linear and quadratic terms for alcohol use are with both included in same model.

We further examined possible effects of confounding by studying the effects of paternal smoking during pregnancy and maternal smoking post-pregnancy on risk of PLIKS, to compare these with the effect of maternal smoking during pregnancy (Table 3). Paternal smoking during pregnancy was associated with any suspected or definite PLIKS in the crude analysis, but this was eliminated after adjusting for confounders and maternal smoking (adjusted OR = 1.05, 95% CI 0.95–1.17).

Table 3

Crude and adjusted odds ratios (OR) and 95% CI for any suspected or definite psychosis-like symptoms (PLIKS) in relation to parental tobacco use within and outside the pregnancy period (linear trend across four smoking categories)

	Suspected or definite PLIKS
	Exposed n	Crude OR (95% CI)	Adjusted OR (95% CI) (model 1) a	Adjusted OR (95% CI) (model 2) b	P
Maternal smoking v. non-smoking during pregnancy	634	1.32 (1.17-1.49)	1.23 (1.08-1.39)	1.22 (0.99-1.49)	0.059
Paternal smoking v. non-smoking during pregnancy	1051	1.15 (1.05-1.26)	1.10 (1.00-1.21)	1.05 (0.95-1.17)	0.336
Maternal smoking v. non-smoking post-pregnancy	759	1.22 (1.10-1.36)	1.10 (0.97-1.24)	0.95 (0.79-1.14)	0.563

a. Adjusted for other alcohol use, cannabis use, and variables in Table 1.

b. Additionally adjusted for other variables in Table 3.

Maternal smoking post-pregnancy was also associated with any suspected or definite PLIKS in the crude analysis (Table 3), but again this was eliminated after adjusting for confounders and maternal smoking during pregnancy (adjusted OR = 0.95, 95% CI 0.79–1.14). Maternal smoking during pregnancy and maternal smoking post-pregnancy were quite strongly correlated (Kendall's τ_b = 0.76). The standard error for maternal smoking during pregnancy was increased by about 60% when both were included in the same model (Table 3), but that for maternal smoking post-pregnancy was relatively unchanged, indicating that collinearity is unlikely to explain the lack of association for smoking post-pregnancy. ^{Reference Kirkwood, Sterne, Kirkwood and Sterne32} Note that only 3730 of the 4253 adolescents had additional data on maternal smoking post-pregnancy and therefore results for maternal smoking during pregnancy in Tables 2 and 3 are slightly different as they are based on different data-sets.

We examined whether the effect of maternal tobacco use differed by trimester of exposure. Smoking during any trimester was very correlated with smoking in other trimesters (Kendall's τ_b>0.80). The offspring of mothers who used tobacco only in their third trimester had a greater risk of developing any suspected or definite PLIKS than offspring whose mothers smoked only in the first trimester (OR for smoking in third trimester only compared with first trimester only 2.1, 95% CI 0.96–4.59, P = 0.063). There were insufficient numbers of women who only used tobacco in their second trimester to examine specific second trimester effects.

Cannabis use during pregnancy

Maternal cannabis use was not associated with any suspected or definite PLIKS in the crude analysis (OR for linear trend 1.22, 95% CI 0.83–1.79). The odds ratio was reduced after adjusting for confounders (Table 2), with adjustment for maternal tobacco use having the greatest impact on attenuation of this estimate (adjusted OR = 0.94, 95% CI 0.62–1.41, P = 0.755). Of the 157 women with PLIKS data who used cannabis during pregnancy, 51 (32.5%) claimed not to have smoked tobacco during their pregnancy. There were insufficient numbers of women using cannabis to examine trimester-specific effects of cannabis use.

Alcohol use during pregnancy

Although 70% of mothers drank alcohol at least once during their pregnancy, the median number of units of alcohol per week consumed was 0 (range 0 to 102). There was an association between maternal alcohol intake during pregnancy and any suspected or definite PLIKS in the crude analysis (OR per 10-unit increase in alcohol 1.24, 95% CI 1.03–1.50), and this was not substantially altered after adjustment (adjusted OR per 10 units 1.19, 95% CI 0.97–1.45). This was a non-linear effect (likelihood ratio for inclusion of quadratic term in crude model χ² = 7.5, d.f. = 1, P = 0.006). Likelihood ratio test results for the overall effect of alcohol on risk of PLIKS are presented in Table 2. Further adjusting for possible mediators of this association had minimal effects on these results.

The increase in risk of suspected or definite PLIKS was primarily present in the offspring of the 25 mothers (0.6% of the sample) who drank >21 units per week. When omitting this extreme group, as a sensitivity analysis, there was no evidence of a non-linear relationship (χ² = 0.3, d.f. = 1, P = 0.566) and no evidence of association between alcohol use and PLIKS (adjusted OR per 10 units 0.97, 95% CI 0.72–1.31) (Table 2).

We also examined trimester-specific effects of maternal alcohol use. Alcohol intake during the first and third trimesters were correlated, although insufficiently to render collinearity a problem in an analysis with both included in the same model (Pearson's coefficient 0.54). Within such a model, first trimester alcohol use (adjusted OR per 10 units 1.41, 95% CI 0.95–2.09) but not third trimester use (adjusted OR per 10 units 0.99, 95% CI 0.63–1.55) was associated with increased risk of PLIKS, although the confidence intervals overlapped substantially.

We further examined possible effects of confounding by studying the effects of maternal alcohol use 4 years post-pregnancy on risk of PLIKS in the offspring. The correlation between alcohol use during and post-pregnancy was not very strong (Pearson's coefficient 0.29). There was no evidence of any association between maternal alcohol use post-pregnancy and any suspected or definite PLIKS either before or after adjusting for alcohol use during pregnancy (LRT for both linear and quadratic terms, χ² = 4.0, d.f. = 2, P = 0.139).

Secondary analyses

There were 165 children (2.6% of those interviewed) with definite, frequent (occurring ≥monthly) PLIKS, and 233 (3.6%) with suspected or definite ‘bizarre’ PILKS. There was no consistent pattern that associations were stronger for either of these outcomes.

Missing data

Results from the multivariable multiple-imputation models were very similar to those using the main data-set, although more precisely estimated, whether we imputed confounders only or outcome measures too.

Confounding

Maternal smoking was strongly associated with markers of adverse sociodemographic characteristics in our data, similar to previous reports, ^{Reference Spencer33} and similar to psychosis. ^{Reference Wicks, Hjern, Gunnell, Lewis and Dalman34} The sociodemographic characteristics of alcohol use in pregnancy were quite different to those of smoking, also similar to previous reports. ^{Reference Day, Cottreau and Richardson35} The association between maternal smoking and PLIKS was attenuated by about 40% after adjusting for a broad range of confounders. However, conventional epidemiological methods for dealing with confounding may be particularly problematic when assessing intrauterine influences on offspring health outcomes. ^{Reference Smith13} One approach to overcome this is to compare the effects of maternal exposure during pregnancy with those of paternal exposure, as the latter is likely to reflect, to some extent, similar confounding characteristics to the former. ^{Reference Smith13} Maternal exposure during pregnancy can also be compared with maternal exposure post-pregnancy, although such a comparison is more problematic, especially for smoking, given that women who continue to smoke throughout pregnancy may differ substantially from those who abstain during this period.

Estimates of association for paternal smoking and maternal smoking post-pregnancy were smaller than for smoking during pregnancy, and both were substantially attenuated after adjusting for maternal smoking during pregnancy. This lends some support for a biological effect of tobacco exposure in utero on risk of psychotic symptoms, although as the confidence intervals for paternal smoking overlap slightly with those for maternal smoking during pregnancy we cannot conclude that these exposures are clearly different from each other.

For alcohol use, adjustment for confounding had minimal effect on results, and there was no association between alcohol use in the post-pregnancy period and PLIKS independent from effects of use during pregnancy. However, the results of the sensitivity analyses indicate there was minimal association between alcohol use and PLIKS once women drinking >21 units of alcohol per week were excluded from the analysis. This level of drinking occurred in less than 1% of women in this cohort. Such women are likely to be very different from other women in the cohort in relation, for example, to personality traits and mental health characteristics. Despite the minimal effect of adjusting for confounders, it is difficult to be confident that this pattern of association is not because of confounding.

Bias

Although this is a large cohort, with a wealth of detailed information, missing data due to attrition and wave non-response in this cohort was not insubstantial, a problem common to other large-scale longitudinal studies. ^{Reference Callaway, McIntyre, O'Callaghan, Williams, Najman and Lawlor36,Reference Plewis, Calderwoof, Hawkes and Nathan37} All estimates, however, were similar in the multiple-imputation analyses, indicating that attrition was unlikely to have substantially biased the results.

Some underreporting of substance use is likely to have occurred, and perhaps especially for cannabis use as this is illegal in the UK. Furthermore, a measure of weekly consumption of alcohol may not adequately capture peak levels of fetal exposure due to binge drinking patterns. Such misclassification tends to lead to underestimation of effects, although only if this were non-differential with respect to outcome.

Possible biological mechanisms

Animal studies indicate that fetal nicotine exposure can result in long-term structural and functional changes, ^{Reference Winzer-Serhan7} including decreased neuronal density and size in the hippocampus and cortex, altered regulation of neuronal apoptosis, ^{Reference Winzer-Serhan7,Reference Slotkin15} and increased expression of receptors for acetylcholine, which plays a critical role in brain maturation through modulation of axonogenesis and synaptogenesis. ^{Reference Slotkin15} However, difficulties exist, both conceptually and pragmatically, in the interpretation of results from animal models in relation to effects in humans.

We are not aware of animal studies to date that have examined the effects of nicotine exposure in utero on putative endophenotypes of schizophrenia. Although endophenotypes of schizophrenia that can be modelled in animals are yet to be clearly determined this could potentially become an informative area for future research.

We observed suggestive evidence that maternal smoking during the third trimester was most strongly associated with risk of PLIKS, although results from subgroup comparisons should be interpreted cautiously. This is rather inconsistent with results from studies of famine ^{Reference St Clair, Xu, Wang, Yu, Fang and Zhang38,Reference Susser, Neugebauer, Hoek, Brown, Lin and Labovitz39} and influenza, ^{Reference Brown, Begg, Gravenstein, Schaefer, Wyatt and Bresnahan40,Reference Byrne, Agerbo, Bennedsen, Eaton and Mortensen41} where early pregnancy exposure is associated with greatest risk of schizophrenia, but may reflect different sensitive periods of risks in brain development for different types of exposure. Maternal smoking, ^{Reference Andres and Day5,Reference Wang, Tager, Van Vunakis, Speizer and Hanrahan42} particularly during late pregnancy, ^{Reference Jaddoe, Verburg, de Ridder, Hofman, Mackenbach and Moll43} is thought to lead to lower birth weight. ^{Reference Winzer-Serhan7} However, adjusting for birth weight, as well as for gestation and 5-minute Apgar score had no effect on the results, and although these measures are likely to be rather crude markers of pre- and perinatal adversity, it seems unlikely that such adversity mediates or confounds the relationship between maternal smoking and offspring psychotic experiences.

Alcohol use is associated with adverse effects on placental development and function, ^{Reference Burd, Roberts, Olson and Odendaal44} and adverse perinatal outcomes, although evidence of adverse effects is not strong for low to moderate levels of alcohol consumption during pregnancy. ^{Reference Henderson, Gray and Brocklehurst45} Adverse long-term neurobehavioural and cognitive consequences of alcohol use have also been described, albeit more inconsistently than for tobacco use, in both human and animal studies. ^{Reference Huizink and Mulder4}

It is quite likely that biological effects of any exposures in utero will impact indirectly upon risk of psychotic phenomena, for example through effects on impulsivity, attention or subtle effects on cognition (that may not be picked up through global cognitive measures such as IQ), and which can potentially also affect development of other, non-psychotic psychopathology. In the EDSP study, maternal smoking was not associated with psychotic symptoms, or with any diagnosis of psychiatric disorder. ^{Reference Spauwen, Krabbendam, Lieb, Wittchen and van Os20} Future studies that aim to increase knowledge about such pathways will help our understanding of the mechanisms underlying disease pathogenesis.

Although at present the status of non-clinical psychosis-like experiences in relation to rare clinical disorders such as schizophrenia is not clear, results from two cohort studies ^{Reference Poulton, Caspi, Moffitt, Cannon, Murray and Harrington46,Reference Hanssen, Bak, Bijl, Vollebergh and van Os47} indicate that the presence of such symptoms may lead to increased risk of clinically important psychotic disorders in later life. Furthermore, as such symptoms are not uncommon in population-based studies ^{Reference Poulton, Caspi, Moffitt, Cannon, Murray and Harrington46,Reference Eaton, Romanoski, Anthony and Nestadt48–Reference Wiles, Zammit, Bebbington, Singleton, Meltzer and Lewis51} and are associated with decreased occupational and social functioning over time, ^{Reference Hanssen, Bak, Bijl, Vollebergh and van Os47,Reference Rossler, Riecher-Rossler, Angst, Murray, Gamma and Eich52} these phenomena may have a large impact on population health and quality of life outside the arena of clinical services, in the same way as depression does. This further highlights the importance of understanding the mechanisms underlying the aetiology of non-clinical PLIKS.

In our cohort, approximately 19% of adolescents attending the PLIKSi had mothers who smoked during pregnancy. If our results for the association between maternal smoking and PLIKS are non-biased and truly reflect a causal relationship, we can estimate that about 20% of adolescents in this cohort would not have developed psychotic symptoms if their mothers had not smoked (the population attributable fraction). Therefore, although the effect size of association for maternal smoking is rather modest, the frequency of this exposure means that maternal smoking may nevertheless be an important risk factor for the development of psychotic experiences in the population.

Implications

Observational studies are limited in determining causality due to potential problems of residual confounding. We observed an association between maternal, but not paternal, smoking during pregnancy and risk of psychotic symptoms in the offspring, consistent with accumulating evidence from animal models of adverse effects on brain development from in utero nicotine exposure. These findings suggest that risk factors for development of non-clinical psychotic experiences may operate during early development. Future studies of how in utero exposure to tobacco affects cerebral development and function may lead to increased understanding of the pathogenesis of psychotic phenomena.