The study is based on a secondary analysis of data from six waves of the 1970 British Cohort Study (BCS70). BCS70 is following up over 17,000 children born during one week in the UK in 1970, with data collected soon after birth (first wave) [38, 39] . Since then, additional waves have taken place in at age 5 (n=12,939), 10 (n=14,350), 16 (n=11,206), 26 (n=8,654), 30 (n=10,833), 34 (n=9,316), 38 (n=8,874) and 42 (n=9,717) years [40-42]. The surveys cover health status; health behaviours; wellbeing; educational attainment; employment and occupation; financial status; social and civic participation; social support; family formation and crime. Anonymised data from ages 5, 10, 26, 30, 34 and 42 follow-up surveys were downloaded from the UK Data Service for this study [43-48].
Identifying Participants with Borderline Intellectual Functioning
While BCS70 includes measures of child cognitive functioning at ages 5 and 10 [49], these were not validated IQ tests but instead, a range of brief tests drawn from existing IQ tests were administered, or tests of assessed attainment that is likely to be related to IQ. In this study, we have followed established procedures in deriving a proxy [49-51].
At age 10, eight tests were administered: the Shortened Edinburgh Reading Test [52]; the Friendly Maths Test [49]; the Pictorial Language Comprehension Test [49]; the Spelling Dictation task [49]; and four subscales of the British Ability Scales, Word Definitions, Word Similarities, Recall of Digits and Matrices [53]. In total, 12,885 (87%) of children participating at age 10 completed at least one assessment; 11,134 (75%) children completed all assessments [49].
Cognitive test results at age 5 were available for an additional 2,568 children who, however, did not have test results reported at age 10. At age 5, the children were administered: the Copying Designs Test [54]; the English Picture Vocabulary Test [55]; the Human Figure Drawing (Draw-a-Person) Test [56]; the Complete a Profile Test [57]; and the Schonell Reading Test [58]. In total, 13,059 (99%) of all children participating in the age 5 survey completed at least one assessment and 11,254 (86%) children completed all assessments [49].
Childhood Adversities
Data collected at age 5 and age 10 follow-ups were reviewed to identify variables that indicated exposure to low socio-economic position and specific adverse childhood experiences as included in a recent systematic review [1]. We identified 25 variables (13 at age 10, 12 at age 5). Of these, 19 were based on 11 indicators of social and/or material deprivation (living in a poor area, living in damp housing, living in overcrowded housing, living in rented housing, low parental educational attainment, low social class, low household assets, low income, living in a workless household, living in a single parent family, potential maternal psychiatric morbidity). All but three of these (living in damp housing, living in overcrowded housing, low income) were collected at ages 5 and 10. The remaining six indicators related to three health-related adversities (accident requiring medical treatment, hospital out-patient attendance at ages, hospital in-patient admission) each collected at ages 5 and 10. Initial inspection of these data indicated that the indicators based on social and/or material deprivation performed well as a simple additive scale (alpha = 0.80), but inclusion of the indicators based on health-related adversities significantly reduced the scale’s internal consistency. As a result, we created a separate additive scale for health related adversities (alpha = 0.53).
Adult Psychiatric Morbidity
At ages 26 and 30 the 24-item self-completed Malaise Inventory was used to measure levels of anxiety and/or depression, with potential adult psychiatric morbidity being identified by a score of eight or more [59]. At ages 34 and 42 an abbreviated 9-item version of the Malaise Inventory was used, with a potential mental health problem being identified by a score of four or more.
Statistical analysis
In order to maximise use of participants’ data and to reduce potential bias resulting from exclusion of partial non-respondents, missing cognitive test data for partial respondents were imputed using multiple imputation routines in IBM SPSS 22. Five parallel data sets were imputed and then averaged to create the final imputed data.
Principal components analysis was used to establish the presence of a general cognitive ability factor across tests and standardised scores on the first component were extracted as a proxy indicator for IQ [49-51]. At age 10, the first extracted component accounted for 59% of the variance of initial eigenvalues with all tests loading positively on the component (loading range 0.55- 0.88). For the respondents with missing data on cognitive testing at age 5 we followed the procedures outlined above to: (1) impute partially missing cognitive test results; (2) establish the presence of a general cognitive ability factor across tests (g); (3) use standardised scores on g as an indicator of IQ at age 5. At age 5, the first extracted component accounted for 41% of the variance of initial eigenvalues with all tests loading positively on the component (loading range 0.47-0.76).
Exploratory analyses indicated significantly higher attrition rates among participants with BIF than those without BIF. We addressed the issue of bias due to attrition by imputing missing data (arising from either wave or item non-response) as previous analyses of BCS70 had indicated that imputation models were preferable to the use of sample weights [60]. We used participant sex, BIF status, child behaviour problems at ages 5 and 10 and available responses to the Malaise Inventory at ages 26, 30, 34 and 42 to impute missing adult mental health data.
In the first stage of analysis we used simple descriptive statistics to characterise the associations between: (1) BIF and exposure to childhood adversities; and (2) exposure to childhood adversities and adult mental health.
In the second stage of analysis we estimated the strength of association between BIF and adult mental health in four models. In Model 1 we reported unadjusted prevalence rate ratios (with 95% confidence intervals) for adult mental health problems among participants with BIF (participants without BIF being the reference group). In Model 2 we estimated prevalence rate ratios for adult mental health problems among participants with BIF adjusted for between group differences in participant gender as initial exploratory analysis indicated that BIF was more common among males (14.3% vs. 12.9%). In Model 3 we estimated prevalence rate ratios for adult mental health problems among participants with BIF adjusted for between group differences in participant gender and exposure to childhood adversities. For these analyses we recoded the number of adversities into population-based quintiles. In Model 4 we included BIF*childhood adversities interaction terms into the model to estimate whether BIF status moderated the association between exposure to childhood adversities and adult mental health problems. Prevalence rate ratios were estimated using Poisson regression with robust standard errors [61, 62] using fully imputed datasets and datasets in which missing data on adult mental health outcomes were not imputed. In the results section we primarily report analyses using fully imputed datasets, commenting in the text on any notable variation in the main results between analyses conducted the imputed and non-imputed data.