Abstract
BACKGROUND: We recently reported that commonly comorbid early onset mental health outcomes were predicted with high accuracy (>90%) and statistical robustness (p = 2.4 x 10(-5)) by a three-factor model composed of adolescent externalizing (EXT) traits, early life adversity, and midbrain dopamine autoreceptor availability (n = 52). AIMS & OBJECTIVES: Here, we investigated whether this positron emission tomography (PET) based model could be reproduced in a much larger sample using functional magnetic resonance imaging (fMRI) data. METHOD: IMAGEN database members with Childhood Trauma Questionnaire (CTQ) data were selected for analyses (n = 1338). Average adolescent EXT scores were derived from the Strengths and Difficulties Questionnaire collected at ages 14 and 16. fMRI reward anticipation responses (large minus no reward, Lg-No; large minus small reward, Lg-Sm) were assessed using a monetary incentive delay (MID) task at ages 14 and 19. Regions of interest were the ventral striatum (VS), caudate, putamen, anterior cingulate cortex (ACC), ventromedial prefrontal cortex (vmPFC), and orbitofrontal cortex (OFC). DSM-IV diagnoses were made using the Development and Wellbeing Assessment at ages 14, 16, and 19. Binary logistic regression models were constructed to determine whether a combination of EXT traits, childhood trauma, and MID fMRI contrasts at either 14 or 19 years of age identified participants with a DSM-IV disorder by age 19. Additional models tested for moderation of CTQ score effects by fMRI blood-oxygen level-dependent (BOLD) signals, and for mediation of CTQ effects by EXT traits. RESULTS: With all possible fMRI contrast parameters, the three-factor models were highly significant (p < 1.0 x 10(-21)). In each of these models, EXT and CTQ scores were significant individual predictors (p < 0.001). At age 14, reward anticipation responses in the VS (Lg-No & Lg-Sm), caudate (Lg-No & Lg-Sm), putamen (Lg-Sm), and ACC (Lg-Sm) were significant predictors in their respective models (p ≤ 0.05). At age 19, reward anticipation in the VS (Lg-No) was a significant predictor within its model (p = 0.025). The models had an overall accuracy of nearly 75% and accounted for ≥ 11% (Nagelkerke R(2)) of the variance in psychiatric disorders. The relationship between CTQ scores and diagnoses was partially mediated by EXT scores (indirect path B = 0.0535, 95% CI = 0.0301-0.0835) and moderated by age-14 ACC (p = 0.0038) and putamen (p = 0.0135) BOLD signals: the lower the reward anticipation response, the greater the effect of high CTQ scores on the likelihood of a diagnosis. DISCUSSION & CONCLUSIONS: The results extend our previous findings in a large sample, increasing confidence in our novel diathesis-stress model of commonly comorbid early onset mental health problems. The results have profound implications for diagnostic classification schemes and pleiotropic views of psychiatric disorder etiology.