5.3.1
Linkage: Families in THIN
Mothers and children were linked on the basis of delivery/birth months and family identi- fier codes in THIN. Linking each birth to the corresponding child involved matching the mother’s delivery date to the child’s month of birth and family identifier. Mother-child pairs were excluded from the cohort if the child could be linked to more than one mother or if the child first registered at their GP practice after nine months of age. To enter the co- hort for this study, children must be born between 1 January 1994 and 31 December 1997 and must be registered for at least 15 years. Parents must be registered for at least one year, so that I could assess rates of early parental depression, but they could potentially leave the cohort after this time point.
Potential fathers were linked to mother-child dyads using an algorithm that has been explored in previous work on parental depression [Dav´e et al., 2010]. I selected males older than 15 years who had the same family identifier as the mother and child at the time of birth. The age difference between the mother and potential father was also restricted to 15 years in order to decrease the risk of erroneously selecting older sons (or in rare cases grandfathers) who live with the family as fathers of the child. Previous research has shown that, in THIN, 40% and 10% of households contained either no or >1 resident adult man, respectively, and these households were excluded as these were either single parent families, or families where the father had (thus far) failed to register with a GP, or registered with a different GP.
Because of my strict inclusion criteria, only traditional ’nuclear’ families consisting of a mother and father with children were included in this cohort. Families with relat-
ives living at the same address, same sex couples or single parents (though only if they were single parents during the first year after childbirth) were not eligible for my cohort. For families where multiple children were registered, I randomly chose one child to be included in the THIN family cohort.
5.3.2
Variable definitions
The primary outcome variable of this study was adolescent depression in children aged 13 - 18 years (Table 5.1). This variable was based on different combinations of codes re- flecting depression diagnoses, depression symptoms and prescribing of anti-depressants at the appropriate therapeutic dose for treatment of depression, as explored in chapter 4. Adolescents were considered to have depression if they had a diagnosis or antidepressant code in their records, or at least two records of depression symptoms within a month. Fur- thermore, for adolescents with a record of an antidepressant prescription, I excluded cases where adolescents had a diagnosis of an eating disorder or anxiety as antidepressants can also be prescribed for these indications.
Parental depression (both maternal and paternal) in the first postpartum year was identified using the same code lists for diagnoses and prescriptions (but not depression symptoms), with the addition of codes that specify postnatal depression. These code lists were created and used in previous studies [Dav´e et al., 2010] and were developed in line with methods described in chapter 2 and reviewed by a general practitioner (Prof Irwin Nazareth). For both parental and adolescent depression, I considered one prescription sufficient for a depression indication. For this study, I focussed on whether the patient was depressed and not whether they were treated for their depression. As such, I considered the decision of the GP to prescribe an antidepressant as indicative of depression (in the absence of an eating disorder or anxiety diagnosis).
I explored childhood internalizing behaviour between the ages of 5 and 13 years as a mediator. Internalizing behaviour is not measured directly in THIN. In order to estimate this latent variable, I used the following indicator variables: sleep disorders, anxiety disorders, recurrent abdominal pain or constipation, recurrent headaches or mi- graine, chronic fatigue, and depressive symptoms (Figure 5.1). For these indications, code lists were developed and reviewed by a general practitioner. For indicators relating
to physical symptoms (abdominal pain, headaches, fatigue, dysphagia) I excluded cases that were due to physical illness. Therefore, I excluded any cases where there was a dia- gnosis for an organic disease (e.g. inflammatory bowel disease in the case of abdominal pain) within a year of first presentation with symptoms.
Table 5.1: Timing of variable measurements
0 - 1 years 1 - 5 years 5 - 13 years 13 - 18 years Early parental depression Parental depression
Covariates Childhood internalizing behaviour Adolescent depression
Information on Townsend scores, a measure for social deprivation as described in chapter 2, was extracted as this might act as a confounder in the association between parental and childhood depression. Negative life events (e.g. death of a parent or sib- ling, parental divorce) were explored as confounders as these are known to be strongly correlated to internalizing behaviour and depression.
I considered parental depression not within of the first year of life as an intermedi- ate variable (or time-varying confounder) as parents with chronic or recurrent depression could impose a greater risk on their children developing depression in later life, either as an environmental influence or through direct genetic transmission. For each year of follow-up (measured by child age), I assessed whether each parent has a record of de- pression diagnosis or antidepressant prescription. The result is a cumulative depression score ranging from 0 (no episodes of depression) to 12 (parental depression in each year up to child age 12 years) for each parent.
Potential child abuse and neglect were considered as confounders, as it known that there is a strong association between maltreatment and later depression. Moreover, chil- dren of parents who were depressed in the antenatal period are four times more likely to be exposed to child abuse [Kotch et al., 1999, Pawlby et al., 2011]. To identify child maltreatment and neglect, I used a code list for suspected or potential child maltreatment and neglect developed by Jenny Woodman [Woodman et al., 2012].
Finally, I included parent behaviour during pregnancy and the first year postpartum that could influence both their risk of depression and the childs risk for later internalizing
behaviour. This included alcohol [Khadjesari et al., 2013] and illicit drug use, severe men- tal illness (e.g. schizophrenia and bipolar disorder), comorbidity, parental age and neg- ative life events (divorce or death of parent/sibling). Other morbidities were assessed by assessing the number of prescriptions from different British National Formulary (BNF) chapters that a parent received during the first year after childbirth. Prescriptions for vaccinations, anaesthesia and contraceptives were excluded from this index.