Association of Interparental Violence and Maternal Depression With Depression Among Adolescents at the Population and Individual Level

Key Points Question Is exposure to parental intimate partner violence (IPV) and/or maternal depression during childhood associated with depression at age 18 years at the population and individual level? Findings In this cohort study of 5029 children born in 1991 to 1992 in the Avon region of southwest England, exposure to IPV or maternal depression in childhood was associated with 24% to 68% higher risk of having severe depressive symptoms at age 18 years. The estimation of an individual developing depression in adolescence based solely on information about parental IPV or maternal depression is poor. Meaning Prevention of IPV and maternal depression can improve children’s mental health at the population level; however, screening children for maternal depression and IPV to target interventions to prevent adolescent depression will not identify many children who might benefit and may unnecessarily target many others who will not develop depression.


Paternal social class
Maternal education Maternal and paternal education was self-reported (32 weeks pregnant). Using the original variables (ALSPAC: c645a, c666a), we derived three categories (due to low numbers in certain groups): 1) CSE/Vocational/O level, 2) A level, 3) Degree.

Paternal education
Family's ethnicity Family ethnicity was reported by mother (ALSAPC: c804, 32 weeks pregnant) as white or non-white. Financial difficulties Financial difficulties were reported by mother (ALSAPC: c525, 32 weeks pregnant). Mother was asked "how difficult at the moment do you find it to afford these items: 1) food, 2) clothing, 3) heating, 4) rent or mortgage, 5) things you will". The response options included: 0 -not difficult, 1 -slightly difficult, 2 -fairly difficult, 3very difficult. The responses were summed up, producing a score between 0 and 15, with a higher score indicating greater financial difficulties. Mother smoking during pregnancy Mothers were asked (ALSAPC: b665, 18 weeks pregnant) whether they smoked cigarettes, cigars or pipe during the first months of pregnancy). We binarized this variable as 'yes' (smoked any of these) or 'no' (did not smoke any of these).

Mother drinking during pregnancy
Mothers were asked about frequency of drinking alcohol in the first three months of pregnancy (ALSAPC: b721, 18 weeks pregnant) and after birth (ALSAPC: b721, after birth). Any indicator of drinking was classified as 'yes', not drinking at all was classified as 'no'.

Mother's age at birth
Age at birth was reported after self-reported (ALSAPC: c994, after birth). Housing tenure Home ownership status was self-reported by mother (ALSPAC: a006, 1 year old). We derived a variable indicating home being: 1) mortgaged/owned, 2) not owned.

Maternal partnership status
Maternal partnership status during pregnancy was self-reported by mother (ALSPAC: a525, 1 year old) mother. Due to low numbers in certain categories, we derive a variable with four categories: 1) Never married, 2) Widowed/divorced/separated, 3) 1st marriage, 4) 2nd/3rd marriage.  (0)  The missing values were replaced with multiple imputation. The proportion of missing information in each variable is given in eTable 2. In line with the recommendations, we included all the variables from the analysis in the imputation model to preserve the relationship between the variables. 20,32 The multiple imputation works under the missing at random (MAR) assumption. 33,34 The MAR mechanism implies that systematic differences between the missing and the observed values can be explained by observed data. 33 The missing-at-random assumption is largely untestable. 35 Hence, we also enriched the imputation model and further maximised the plausibility of the MAR assumption with birthweight as an auxiliary variable. This variable was not part of the substantive model of interest, but it was associated with missingness and with depressive symptoms. Including auxiliary variables can improve the accuracy of the MI and minimise non-random variation in the imputed values. 20 In addition, the imputation model was rich due to including covariates used in the analyses, such as socioeconomic indicators or mental health in childhood, which were predictive of both missingness (see eTable 3) and depressive symptoms.
The missing data were imputed using multiple imputation by chained equations (MICE), due to the non-monotone pattern of missing values, and due to its ability to accommodate various types of variables in the imputation model, including continuous and categorical ones. This approach uses a series of univariate conditional imputation models to impute missing data. 36 Continuous variables were imputed using predictive mean matching and categorical variables using logistic regressions.
The predictive mean matching approach provides robust estimates if the normality assumption is in question, 37 this is particularly relevant to mental health outcomes as they tend to be skewed, 38