COMBINING MENDELIAN RANDOMISATION WITH DEPRESSION TRAJECTORIES TO IDENTIFY DEVELOPMENTALLY SPECIFIC PREDICTORS OF CHANGE IN DEPRESSIVE SYMPTOMS

Prevalence of depression is increasing, especially amongst adolescents and young adults, representing a key risk period where intervention is critical. When using Mendelian randomisation (MR) to identify causal risk factors for depression, estimates are limited to average lifetime effects, rather than being specific to developmental stages.

Methods. We have combined trajectories of depressive symptoms with MR to identify developmentally specific risk factors. We used repeated measures of depressive symptoms (short Moods and Feelings Questionnaire) in the ALSPAC cohort, with 11 repeated assessments covering ages 9 to 27 years. First, we used a repeated measures multi-level model (MLM) to describe the average trajectory of depressive symptoms. Linear splines split by knot points were used to explain the non-linear pattern of growth. Second, we used latent class analysis to explore heterogeneity in depression trajectories. Third, we combined both trajectory models with genetic instruments for depression (positive control) and with modifiable risk factors for depression.

Our models included 44,611 repeated assessments of sMFQ from 6,422 unique individuals. Our best fitting MLM trajectory had three linear splines corresponding to puberty (9-14.5 years), adolescence (14.5-21 years) and early adulthood (21-27 years). Latent classes were stable low, decreasing, transient, increasing and stable high. Positive control genetic instrument for MDD predicted trajectories, most strongly membership into the increasing and stable high class. Genetic instruments for BMI and educational attainment were not associated with change in population average depressive symptoms at any of the different developmental stages nor with class membership. This could suggest no causal effects of these risk factors at these developmental stages, or low power.

We are continuing to develop our methods, test power and incorporate additional risk factors. We believe that combining outcome trajectories with MR analyses has wide ranging application to improve specificity of causal effects and recommendations for intervention development.