Antenatal domestic violence, maternal mental health and subsequent child behaviour: a cohort study

Authors


Dr LM Howard, Health Services and Population Research Department, Institute of Psychiatry, King’s College London, Box P031, De Crespigny Park, London SE5 8AF, UK. Email louise.howard@kcl.ac.uk

Abstract

Please cite this paper as: Flach C, Leese M, Heron J, Evans J, Feder G, Sharp D, Howard L. Antenatal domestic violence, maternal mental health and subsequent child behaviour: a cohort study. BJOG 2011;118:1383–1391.

Objective  To investigate the long-term impact of antenatal domestic violence on maternal psychiatric morbidity and child behaviour.

Design  Cohort study.

Setting  Avon, UK.

Population or sample  A birth cohort of 13 617 children and mother dyads were followed to 42 months of age.

Methods  Experiences of domestic violence and depressive symptoms were gathered at 18 weeks of gestation and up to 33 months after birth, together with maternal, paternal and child characteristics.

Main outcome measures  Child behavioural problems were assessed at 42 months using the Revised Rutter Questionnaire.

Analysis  Logistic regression with the use of multiple imputation employing chained equations for missing data.

Results  Antenatal domestic violence was associated with high levels of maternal antenatal (odds ratio [OR], 4.02; 95% confidence interval [CI], 3.4–4.8) and postnatal (OR, 1.29; 95% CI, 1.02–1.63) depressive symptoms after adjustment for potential confounders. Antenatal domestic violence predicted future behavioural problems at 42 months in the child before adjustment for possible confounding and mediating factors (OR, 1.87; 95% CI, 1.45–2.40); this association was not significant after adjustment for high levels of maternal antenatal depressive symptoms, postnatal depressive symptoms or domestic violence since birth.

Conclusions  Antenatal domestic violence is associated with high levels of both maternal antenatal and postnatal depressive symptoms. It is also associated with postnatal violence, and both are associated with future behavioural problems in the child at 42 months. This is partly mediated by maternal depressive symptoms in the ante- or postnatal period.

Introduction

The British Crime Survey1 estimates that around 24% of women have suffered domestic violence. Domestic violence has significant health consequences, including physical injuries, gynaecological problems and psychiatric morbidity, namely post-traumatic stress disorder, depression, substance abuse and suicidality.2–4 It is well recognised that children who witness such violence are at greater risk of developing emotional, behavioural and educational problems,5–7 with a cumulative effect of witnessing domestic violence and being exposed to maternal depression or anxiety leading to worse child outcomes.8

When abuse occurs during pregnancy, there is an increased risk of complications, including pre-term labour, reduced birthweight, miscarriage and fetal death, in women.9,10 The recent UK Confidential Enquiry into Maternal and Child Health11 also highlighted domestic violence as a cause of perinatal maternal deaths. A review of studies measuring domestic violence during pregnancy reported the prevalence of physical abuse in most studies to be 4–8%;12 higher rates are found when sexual abuse and emotional abuse are included.

There are both biological and psychological reasons for hypothesising that domestic violence during pregnancy could be associated with long-term adverse consequences for the child. Firstly, it has been proposed that the effects of prenatal stress (which is associated with adverse outcomes for the child, including prematurity and adverse developmental outcomes13–15) are mediated by increased plasma levels of cortisol and corticotrophin-releasing hormone in the mother and fetus which alter the programming of fetal neurons13,14 and thus the child’s subsequent behaviour. There is also evidence that mothers who have suffered antenatal violence have more negative representations of themselves and their infants,16 and a weaker maternal–fetal and maternal–infant bond,17 both of which may lead to adverse child behavioural outcomes. Finally, as domestic violence is associated with common mental disorders, the impact of antenatal domestic violence on the child could be mediated by antenatal and postnatal psychiatric morbidity, as there is considerable evidence that both are associated with adverse child developmental outcomes.18–20

However, there has been little research into the examination of whether antenatal domestic violence is associated with long-term adverse behavioural outcomes in the child and no research that has investigated whether this is mediated by psychiatric morbidity. We therefore aimed to investigate whether antenatal domestic violence is associated with adverse child development and is mediated by maternal common mental disorders, in particular maternal depressive symptoms, as these are the most common symptoms in the perinatal period.21

Our hypotheses were as follows:

  • 1 Antenatal domestic violence is associated with high levels of antenatal and postnatal psychiatric depressive symptoms.
  • 2 Antenatal domestic violence is associated with adverse child behaviour outcomes at 42 months.
  • 3 Any association between antenatal domestic violence and child outcomes is mediated by high levels of depressive symptoms.

Methods

The Avon Longitudinal Study of Parents and Children (ALSPAC) cohort study22 aimed to recruit all pregnant women resident in Avon and expected to deliver their child between 1 April 1991 and 31 December 1992; 14 541 women were enrolled (around 85% of the eligible population). These 14 451 pregnancies resulted in 14 676 known fetuses, 14 062 of which were live births and 13 988 were alive at 1 year. When excluding children born as part of a multiple birth and those who did not survive beyond the first year, there were 13 617 mother and child pairs. Data were collected from pregnancy onwards using postal questionnaires.

Measures

Child behavioural problems were recorded at 42 months of age using the Revised Rutter Scale.23 The scale consists of hyperactivity, emotion, conduct problems and pro-social behaviour domains, a high score indicating more problems (low score on the pro-social subscale). There is no clinical cut-off for these measures; therefore, in line with previous research,24 we defined children with behavioural problems as being in the top 10% of scores. We ran the analysis for the continuous and dichotomised outcome, and report the results of the binary measure, as these were similar to the analysis of the continuous outcome and are easier to interpret and compare with other published findings.

Domestic violence

Antenatal domestic violence was assessed at 18 weeks of gestation. Participants were asked two questions about whether their partner had been emotionally cruel and/or had physically hurt them since the start of the pregnancy. We considered a woman to have experienced antenatal domestic violence if she responded positively to either physical or emotional cruelty since the start of the pregnancy. Similar questions were used to identify postnatal domestic violence at 2, 8, 21 and 33 months postnatally, referring to the period since the last questionnaire. Violence from the partner to the mother, partner to the child and mother to the child was reported by the mother in the same way at these times. The repeated responses were summarised into a variable indicating the number of time points at which violence had been experienced since the birth, and a separate variable for each of violence occurring to the mother, violence occurring to the child from the partner and violence occurring to the child from the mother.

Depression

Maternal depressive symptoms were assessed at the same time points as domestic violence using the Edinburgh Postnatal Depression Scale (EPDS),25 a well-established 10-item questionnaire (score 0–30) shown to be useful outside as well as during the postnatal period.26,27 A cut-off of ≥13 points is used as a definition of probable clinical depression;27,28 we have called this a high level of depressive symptoms (i.e. clinically significant, but not indicating a diagnosis of depression, as EPDS cannot generate a diagnosis). As with domestic violence, a summary measure, indicating the number of time points at which probable depression had occurred since the birth, was used in the multivariate analysis. Paternal depressive symptoms were assessed in the same way at 18 weeks of gestation and at 2, 8 and 21 months postnatally, and a summary measure of high levels of post-birth depressive symptoms was derived.

Life events

A life events scale based on that by Brugha and Cragg29 was derived, covering 13 life events, including serious illness/death in the family, marital difficulties and financial problems, measured at 8, 21 and 33 months after birth.

Maternal characteristics

Maternal education was simplified into three categories: low (UK CSE, vocational); medium (UK O-levels at 16 years); high (A-levels at 18 years and/or university degree). Women were considered to have smoked during pregnancy if they had smoked any form of tobacco in the 2 weeks prior to completion of the 18-week questionnaire. Women were questioned on their alcohol use at 18 weeks of gestation (responses categorised: <1 unit per day and one or more units per day).

Obstetric outcomes

Information was collected on birthweight and gestational age, which was used with height, weight, ethnicity of the mother and parity of the child to determine the birthweight centile of the child using the Gestation Network calculator;30,31 children in the lowest 10% were considered to be small for gestational age.

Analysis

Associations between antenatal domestic violence and maternal depressive symptoms at 18 weeks of gestation and 8 weeks postnatally were assessed using chi-square test. Multivariate analysis was conducted using logistic regression to adjust for possible confounding factors: maternal education, marital status, maternal age, homeownership status, parity, ethnicity and paternal depression. Possible confounders were decided on, prior to the analysis, based on a knowledge of the area and plausible mechanisms. Bivariate associations between behavioural problems, antenatal domestic violence and baseline characteristics of the mother and child were assessed using chi-square or t-tests. The association between antenatal domestic violence and child behavioural problems was investigated using multivariate logistic regression. The data were analysed in three stages: (1) calculation of unadjusted odds ratio (OR) of antenatal domestic violence in relation to behavioural problems; (2) adjustment for possible antenatal confounding factors—maternal age, education, homeownership status, smoking and alcohol use during pregnancy, sex of child, size of child for gestational age and antenatal depression; (3) addition of potential post-birth mediators to the model—postnatal violence (to mother, from mother to child and from father to child), paternal depression, maternal depression and recent life events. To investigate the possible mediating effects of antenatal maternal depressive symptoms, postnatal maternal depressive symptoms and postnatal domestic violence on the association between antenatal domestic violence and behavioural problems, the steps of Baron and Kenny32 were followed and the coefficients of the effect of antenatal domestic violence were compared between models with and without the inclusion of these factors. A new variable was also created for domestic violence exposure, that is no violence, antenatal violence only, antenatal and postnatal violence, postnatal violence only, to give a direct comparison of the different levels of exposure. The analysis was repeated for the three subscales of conduct, hyperactivity and emotional problems, and antenatal violence by gender interactions were explored.

The data were initially analysed using complete cases only, and patterns of missing data were investigated. Missing covariates were judged to be likely to be missing at random, meaning that the probability of missing data can be modelled by the observed data. To improve efficiency, missing covariate values were imputed using chained equations33 with the ‘ice’ command in Stata.34 This was carried out on the full dataset, including all variables used in this analysis as well as additional variables known to be related to nonresponse. As a result of the low level of complete data, 20 imputations were created using all variables. Variables were imputed using the appropriate model: linear, logistic or multinomial logistic. Behavioural problem score scales were imputed using prediction matching on the behavioural scales. This method predicts the missing value and then allocates the closest nonmissing prediction as the imputed value, that is it uses actual observed values. This was necessary because of the skewed nature of the scales, as it maintains the shape and range of the distribution. Missing values of the outcome variables were imputed together with the covariates; however, analyses only included those respondents with the outcomes reported. Using the ‘mim’ command in Stata, analyses were run on the imputed datasets and pooled to give estimates and their variances.

Results

As a result of the longitudinal nature of this cohort study, there is a high level of missing data in both the exposure and outcome variables. The amount of missing data increases at each time point—many participants miss one time point but come back at another, or miss some questions and not others at one time point; 7789 (57%) of the 13 617 participants had complete information on antenatal variables for the analysis of antenatal depressive symptoms; 5681 (42%) had information on postnatal depressive symptoms. At 42 months, 9796 children (72%) had a Rutter score recorded and 7480 (55%) mother–child dyads had full information on child behavioural outcomes and antenatal covariates; 3404 (25%) mother–child dyads had data on all 14 antenatal and postnatal variables of interest.

The comparison of characteristics between mother–child dyads with a Rutter score at 42 months and those without indicates a selection bias—fewer experience antenatal domestic violence or antenatal depression, and parents are more likely to be better educated and married. The size of the child at birth, sex of the child, alcohol use during pregnancy and partner anxiety are not associated with missing data (Table 1).

Table 1.   Comparison of baseline characteristics between mother–child pairs with or without complete information at 42 months
Variable14 variables of interest
All presentAny missingSignificance
  1. SD, standard deviation.

Antenatal domestic violence (%)
No3306 (97)7433 (92)<0.001
Yes98 (3)677 (8)
Maternal antenatal depression (%)
No3085 (91)6966 (84)<0.001
Yes319 (9)1290 (16)
Age of mother (years)
Mean (SD)29 (4.3)27 (5.1)<0.001
Ethnicity (%)
White3364 (99)8337 (97)<0.001
Nonwhite40 (1)273 (3)
Highest education (%)
Low657 (19)2978 (34)<0.001
Medium1183 (35)3002 (35)
High1564 (46)2718 (31)
Marital status at 8 weeks of gestation (%)
Never married318 (9)2113 (23)<0.001
Married2934 (86)6628 (71)
Other142 (4)624 (7)
Housing (%)
Own/mortgage2968 (87)6341 (68)<0.001
Rent436 (13)2958 (32)
Smoke during pregnancy (%)
No2996 (88)7292 (77)<0.001
Yes408 (12)2129 (23)
Alcohol use during pregnancy (%)
<1 unit/day3353 (98)9081 (98)0.034
≥1 unit/day51 (2)193 (2)
Small for gestational age (%)
No3081 (91)6097 (90)0.272
Yes323 (9)691 (10)
Sex of child (%)
Male1762 (52)5261 (52)0.808
Female1642 (48)4950 (48)
Parity (%)
01695 (50)3965 (43)<0.001
11158 (34)3238 (35)
2+551 (16)1993 (22)
Partner antenatal depression (%)
No3184 (97)5904 (95)<0.001
Yes98 (3)284 (5)

The combined results from the imputed datasets estimate that 6% of women experienced emotional cruelty, 2% reported physical cruelty and a total of 7% reported emotional and/or physical violence at 18 weeks of gestation. Rates increased to a maximum of 14% experiencing domestic violence at 33 months after the child was born. Antenatal violence was highly correlated with violence after the birth—71% of women who experienced antenatal domestic violence during pregnancy also experienced violence postnatally.

Antenatal domestic violence is associated with high levels of antenatal and postnatal depressive symptoms

Fifteen percent of the 13 617 women in the imputed datasets had an EPDS score of ≥13 (i.e. probable antenatal depression) at 18 weeks of gestation; 44% of women who had experienced antenatal domestic violence had probable depression, compared with 12% of women who had not experienced antenatal domestic violence (unadjusted OR, 5.47; 95% confidence interval [CI], 4.7–6.4); 25% of women who had experienced antenatal violence experienced probable depression at 8 weeks postnatally, compared with 10% of women who had not experienced antenatal domestic violence (unadjusted OR, 2.94; 95% CI, 2.4–3.6).

The multivariate logistic regression confirmed our first hypothesis that antenatal violence is associated with high levels of antenatal depression (imputed data with maternal EPDS at 18 weeks of gestation n = 11 660; adjusted OR, 4.02; 95% CI, 3.4–4.8; P < 0.001). The analysis was repeated for postnatal depressive symptoms at 8 weeks after the birth, including additional adjustments for maternal antenatal depressive symptoms, paternal postnatal depressive symptoms and size of the child adjusted for gestational age. A smaller but still significant association with antenatal violence was detected after adjustment (imputed data of dyads with probable maternal depression at 8 weeks n = 11 429; adjusted OR, 1.29; 95% CI, 1.02–1.63; P = 0.035).

Antenatal violence is associated with adverse childhood behavioural outcomes

Antenatal violence was more commonly reported in the mothers of children with behavioural problems at 42 months (11%), compared with 7% of mothers with children with no problems (unadjusted OR, 1.87; 95% CI, 1.45–2.40). Other variables, including age, education, homeowner status, smoking and alcohol use, were also associated with behavioural problems in children in the bivariate analysis (Table 2).

Table 2.   Bivariate analysis of mother and child characteristics for child behavioural problems at 42 months—20 imputed datasets
VariableRevised Rutter score at 42 months (n = 9697 dataset)
Bottom 90%Top 10%P
  1. SD, standard deviation.

Antenatal domestic violence (18 weeks)
Yes610<0.001
Any domestic violence since birth
Yes1934<0.001
Antenatal depression at 18 weeks
Yes1124<0.001
Any depression since birth
Yes2146<0.001
Mother’s age at birth (years)
Mean (SD)28.7 (4.7)27.7 (5.0)<0.001
Highest education
Low2631<0.001
Medium3636
High3933
Homeowner status (owners vs renters)
Home owner8069<0.001
Smoke
Yes1624<0.001
Alcohol
≥1 unit/day23<0.001
Size for gestational age
Small10140.001
Sex
Male5159<0.001
Life events at 33 months
02017<0.001
13023
22626
31517
4+917
Any paternal depression since birth
Yes816<0.001
Antenatal paternal depression
Yes49<0.001

Adjusting for potential baseline confounders (age of mother, smoking and alcohol use during pregnancy, education, homeownership status, sex and size of the baby) indicates that antenatal domestic violence is still associated with behaviour problems at 42 months (OR = 1.52, 95% CI, 1.17–1.97).

The potential mediating effect of antenatal depressive symptoms was then examined. Adjustment for probable antenatal depression was found to mediate the effect of antenatal domestic violence on child development (OR, 1.22; 95% CI, 0.93–1.60) (Table 3).

Table 3.   Logistic regression of antenatal domestic violence (ADV) on behaviour at 42 months, adjusted for antenatal and postnatal confounders
Variables included in the modelOutcome—behavioural problem at 42 months (complete case n = 3404)Outcome—behavioural problem at 42 months (imputed n = 9796 per dataset)
Odds ratio95% confidence intervalPOdds ratio95% confidence intervalP
  1. *Antenatal confounders: age of mother, education, homeownership status, smoking, alcohol use, sex of child, size of child for gestational age.

ADV—unadjusted1.720.953.120.0751.871.452.40<0.001
ADV—adjusted for antenatal confounders*1.600.872.940.1271.521.171.970.002
ADV—adjusted for antenatal confounders* + maternal antenatal depression1.070.572.020.8221.220.931.600.145
ADV—adjusted for antenatal confounders* + maternal depression since birth1.050.551.980.8921.140.871.500.347
ADV—adjusted for antenatal confounders* + postnatal domestic violence0.910.491.720.7800.990.741.330.948
ADV—adjusted for antenatal confounders* + postnatal domestic violence from mother to child1.270.682.370.4571.311.011.720.044
ADV—adjusted for antenatal confounders* + postnatal domestic violence from father to child1.290.692.420.4261.270.961.680.092
ADV—adjusted for antenatal confounders* + antenatal depression, domestic violence since birth, domestic violence to child from mother, domestic violence to child from father, paternal depression since birth, recent life events0.670.341.290.2280.800.591.100.170
ADV—adjusted for antenatal confounders* + antenatal depression, domestic violence since birth, domestic violence to child from mother, domestic violence to child from father, paternal depression since birth, recent life events and depression since birth0.580.281.180.1350.860.631.180.355

Adjustments for other potential post-birth mediators of domestic violence to the mother and maternal depressive symptoms were then investigated. We included violence in the 33 months after birth and probable antenatal depression or probable depression in the 33 months since birth individually to the model; the addition of each variable abolished the significant association between antenatal domestic violence and behavioural problems (OR, 0.86; 95% CI, 0.63–1.18).

We then investigated the impact of antenatal domestic violence using the variable ‘total domestic violence exposure’ in the regression models (i.e. the impact of no violence versus antenatal violence only versus antenatal and postnatal violence since birth versus violence since birth only), and found that antenatal violence alone was not significantly associated with child behavioural problems (Table 4).

Table 4.   Impact of total maternal exposure to antenatal and postnatal domestic violence on behavioural problems at 42 months
No violencen (average per imputed dataset)Odds ratio—imputed dataset95% confidence intervalP
7580Reference category
Antenatal violence only1831.310.752.320.340
Postnatal violence only16082.030.712.42<0.001
Antenatal and postnatal violence4252.591.963.41<0.001

We repeated the analysis using the Rutter behavioural outcome as a continuous measure; with this outcome, antenatal domestic violence was still associated with childhood behavioural problems after adjusting for antenatal factors, probable antenatal depression and probable postnatal depression, the presence of domestic violence leading to an average increase of 0.71 points on the Rutter scale (coefficient, 0.71; 95% CI, 0.22–1.21; P = 0.005). Again, the association was entirely mediated by postnatal domestic violence, with antenatal violence leading to a nonsignificant increase of 0.41 points on the Rutter behavioural scale (coefficient adjusted for postnatal violence, 0.41; 95% CI, −0.12 to 0.94; P = 0.132). The analyses were replicated on the questionnaire subscales for conduct, hyperactivity and emotional problems, and antenatal violence by gender interactions were investigated. Antenatal domestic violence was not associated with hyperactivity; the results for conduct and emotional problems were similar to those for the total problems scale for all children. There was a statistically significant impact of antenatal violence on behavioural problems, but this was removed when post-birth violence or probable depression was accounted for.

There were no significant interactions between antenatal violence and gender on total behavioural problems, hyperactivity or emotional problems. A significant interaction with gender was found in the conduct problems subscale (unadjusted interaction coefficient, 1.70; P = 0.02; adjusted interaction, 1.61; P = 0.049). The unadjusted OR of antenatal violence on conduct problems in boys is 1.47 (95% CI, 1.08–2.01) and in girls is 2.50 (95% CI, 1.79–3.49); the fully adjusted ORs are 0.77 (95% CI, 0.52–1.15) and 1.33 (95% CI, 0.89–2.00) in boys and girls, respectively.

Discussion

We found that antenatal domestic violence was associated with high levels of both antenatal depressive symptoms (at 18 weeks of gestation) and depressive symptoms at 8 weeks postpartum. Domestic violence during pregnancy was also associated with behavioural problems in the child at 42 months, which was mediated by high levels of maternal antenatal depressive symptoms and depressive symptoms after birth. The association was not found after adjustment for domestic violence that occurred after the birth of the child in its first years of life. There was a very strong association between antenatal and postnatal violence (only 29% of women reporting violence experienced it only antenatally); we therefore did not have sufficient statistical power to test whether antenatal violence and childhood behaviour outcomes were independently associated. This strong correlation between antenatal and postnatal violence clearly indicates that antenatal violence is a marker for postnatal violence, and subsequent poor child outcomes.

There is a growing body of evidence indicating that domestic violence is associated with psychiatric morbidity.3,35 The association of antenatal domestic violence with high levels of postnatal depressive symptoms has been described previously,36 but there have been few studies examining the relationship between domestic violence and antenatal depressive symptoms. There is also a large literature on the impact of witnessing domestic violence as a child and child developmental outcomes,5–7 and the impact of maternal depression on child development,19 but no study has previously investigated the relationship between antenatal violence and subsequent adverse impact on child development. Our study confirms this association and finds that it is at least partly a result of the mediating effect of maternal depressive symptoms and the ongoing violence in the family after birth. However, it is difficult to disentangle which factor is most important, as both have a profound effect on child development. In addition, genetic influences could also explain this association: depressed mothers are more likely to have antisocial partners,37 and heritabilities of 30–70% have been found for childhood behavioural problems.38,39

The strengths of this study are its longitudinal design with the integration of domestic violence variables with a range of linked demographic and outcome variables. Limitations include the use of postal self-report questionnaires; the respondent’s partner could therefore have been present at the time of questionnaire completion, which may have led to an under-reporting of violence. In addition, domestic violence measurement was confined to two questions which did not give action-based examples of violent behaviour (which increase the rates of reporting violence40,41) and did not include sexual violence or fear of their partner. Moreover, the questions were only administered at one time point in the antenatal period, so that violence that started later in pregnancy would not have been detected. The effect of this probable under-reporting of domestic violence would be the attenuation of associations with depression. Detailed data on domestic violence and the status of the perpetrator (living with, separated, etc.) over time are not available, but domestic violence can continue and even increase in severity after separation,1 so that changes in partner status (which occurred in only 6–9% of women [percentages vary depending on which wave of data collection was examined] in the period of follow-up in this study42) are unlikely to change the associations reported here.

The EPDS was used to identify depressive symptoms, and recent research has suggested that it also measures anxiety symptoms.43 EPDS scores do not establish psychiatric diagnoses, but high scores indicate probable common mental disorders, particularly depression, but also anxiety disorders. Missing data are inevitably a problem over this length of follow-up, but the use of multiple imputation techniques has enabled us to increase the efficiency of the available data, and reduces the bias that may occur in a complete case analysis.

Conclusions

This study confirms that antenatal domestic violence is an indicator of an increased risk of postnatal violence, antenatal and postnatal psychiatric morbidity and adverse effects on child development. Maternity services in some countries are now mandated to identify domestic violence and psychiatric morbidity in pregnancy,44,45 and we would urge policy makers internationally to improve the detection of both; pregnancy is a time at which a woman is in frequent contact with health professionals and therefore more likely to disclose domestic violence and psychiatric symptoms. Currently, there is a very limited evidence base on interventions addressing antenatal violence and antenatal depression, and we suggest that this should be a priority for future research.

Disclosure of interests

We have read the journal’s policy and acknowledge the following conflicts: GF and LMH are members of the UK Department of Health, Violence Against Women and Children Implementation Group.

Contribution to authorship

LMH conceived the study design and led the drafting of the manuscript. CF carried out the data analysis and contributed to the writing of the manuscript. JH and ML supervised the data analysis and interpretation of the findings, and contributed to the writing of the manuscript. JE, DS and GF participated in the design of the study and helped to draft the manuscript. All authors read and approved the final manuscript.

Details of ethics approval

Ethics approval for the study was obtained from the ALSPAC Law and Ethics Committee and the Local Research Ethics Committees. ALSPAC has a child protection and confidentiality policy. This includes details of what staff should do in situations in which a member of staff feels that a young person is in serious danger and that emergency action needs to be taken to protect him or her. In such circumstances, a Child Protection Form (without identifying details) is passed onto the Executive Director (or another Executive member if she is not available) immediately for a decision about whether to refer to the community paediatrician. If the case is referred, the community paediatrician will make a decision about whether to refer to social services within 24 hours and will inform the Executive Director immediately.

Funding

LMH is affiliated with the National Institute for Health Research South London and Maudsley/Institute of Psychiatry Biomedical Research Centre for Mental Health. GF and LMH are funded by the UK Higher Education Funding Council. The UK Medical Research Council (Grant Ref: 74882), the Wellcome Trust (Grant Ref: 076467) and the University of Bristol provide core support for ALSPAC. The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript.

Acknowledgements

We are extremely grateful to all the families who took part in this study, the midwives for their help in recruiting them and the whole ALSPAC team, including interviewers, computer and laboratory technicians, clerical workers, research scientists, volunteers, managers, receptionists and nurses.

Ancillary