Individual Submission Summary
Share...

Direct link:

Poster #44 - Can data science illuminate antecedents to optimal outcomes for ADHD?

Sat, March 23, 12:45 to 2:00pm, Baltimore Convention Center, Floor: Level 1, Exhibit Hall B

Integrative Statement

50% of children with Attention Deficit Hyperactivity Disorder (ADHD) will be impaired as adults (Faraone, Biederman, & Mick, 2006; Lara et al., 2009). But what about the other 50%? We know that intelligence and socio-economic status (SES) are factors in outcomes, but we don’t know much about psychosocial factors (Costello & Maughan, 2015), which may be more open to influence.

It is challenging to study psychosocial factors in long-term outcomes for ADHD because the current definition is relatively recent. Large representative samples with ADHD identified in childhood, rich psychosocial data, and longitudinal adult (age 30+) life outcomes are non-existent. However, datasets do exist with all of the above except ADHD identified, and data science methods can address this gap.

We developed a method to retrospectively measure DSM-5 ADHD (American Psychiatric Association, 2013) at age 10 in the 1970 British Cohort Study (BCS70; N=11,426; Centre for Longitudinal Studies, 2015). A 16-item ADHD scale (Cronbach’s alpha =0.85) was derived by mapping DSM-5 ADHD criteria to teacher and parent rated questionnaire items about child behaviour. The mapping was validated with an expert panel. Based on DSM-5 diagnostic criteria, we inferred an ADHD symptomatology categorical subgroup (N=594; 5.2%), which aligned in size and composition with ADHD prevalence estimates from epidemiology (see Table 1; Willcutt, 2012). The scale was then used to fit an Item Response Theory (IRT) mixture model and predict a nuanced dimensional measure of latent ‘ADHDness’ for the entire age 10 sample (Figure 1).

Current work builds on these categorical and dimensional ADHD measures, and utilises data from BCS70 at age 42 (t2). State regulation theory, a theory of stress and human performance (Sanders, 1983), is the basis for hypotheses. We are using statistical models to evaluate relationships between latent variables: SES, intelligence, stressors, and protective (psychosocial) factors measured at age 10 (t1), and life satisfaction, educational attainment, SES, and health measured at t2.

298 (50.2%) of the 594 participants identified in the t1 ADHD subgroup responded at t2. Relative risk ratio comparisons between ADHD and non-ADHD respondents at t2 showed the ADHD group was more likely to have no academic qualifications (1.74), a partly skilled or unskilled job (1.99), and low life satisfaction (1.48).

What about ADHD subgroup members with positive outcomes? 191 (64.1%) of the 298 in the t1 ADHD subgroup reported life satisfaction at t2 as high, and 99 (35.9%) as low. A between groups t-test comparison showed the mean dimensional ADHD score did not differ significantly (t(288) = 0.83, p=0.41) suggesting that ADHD symptomatology alone could not be used to predict life satisfaction. Future work will use regression and generalized structural equation models (GSEM) to test additional hypotheses about t1 predictors and t2 outcomes.

Data science applied to large cohort studies offers statistical power that is often a challenge for clinical research. This is an exciting opportunity to inform educational policy for ADHD with an emphasis on positive adult outcomes.

Authors