Background:
Exposure to adversity in childhood has been linked with later onset of severe mental illness. To date, quantification of exposure risk has been limited to relative ratios, derived from simple outcome-exposure analyses, using linear combinations of, at most, a few measures of exposure. In contrast, adverse environments often manifest as combinations of inter-related dimensions. A scale which better incorporates this complexity will help disentangle contributions to the risk of severe mental illness from genetic, familial and environmental sources.
Context: We have access to a large population cohort of children born in WA between 1980 and 2001 (N=467,945). Measures of their exposure to adversity covering: i) parental separation, ii) family functioning, iii) family structure, iv) family stressors, v) ecological level measures and vi) socioeconomic indicators have been determined from their birth records linked with several state health and social registers. Where relevant, the timing of exposures was recorded. Outcomes of severe mental illness were noted in June 2011 (children aged between 9.5 and 31.5 years).
Methodological implications: The large number of measurements for each child from the six dimensions above include categorical to continuous types. Correlation of measures is expected: i) within families, ii) longitudinally for a given child and iii) within geographical proximity for ecological level measures. It is likely that similar exposure at different ages will have differing impacts. Some missing and censoring of measurements exists. There is considerable censoring of the outcome.
Intent: We recognise the need for simplifying assumptions to enable the construction of our adversity scale. Initially, we propose to develop the scale based on exposure before 10 years of age. In the first of two stages, exposure profiles within each dimension will be categorised according to their associated risk of severe mental illness. This will be determined using logistic regression (ignoring outcome censoring as an approximation) and a training sample of N=200,000. The risk of severe mental illness will be modeled as a combination of many exposures to adversity within a given dimension, allowing for interactions between component measures, with the aim of summarising each dimension as a single ordinal metric. Secondly, the influence of each dimension on the risk of severe mental illness will be assessed in the presence of other dimensions. We will again use logistic regression to model the risk of severe mental illness as a function of a child’s risk profile across six dimensions. Important interactions between dimensions will be included. A final categorisation of the resultant multi-dimensional exposure profiles will produce a scale which nominates a risk of developing severe mental illness based on adversity exposure history. Assessment of the scale’s ability to predict the relative risk of developing a severe mental illness can be undertaken on a separate sample of N=200,000, by comparing the actual proportion of children noted with severe mental illness within each risk profile category, to the proportions predicted by the scale.
Request: Constructive observations on the proposed approach are sought as we strive for optimal use of this unique dataset.