Joint multi-family history and multi-polygenic score prediction of major depressive disorder

Major depressive disorder (MDD) is a complex psychiatric disorder influenced by both genetic and environmental factors. Family history of MDD and other psychiatric disorders are strong predictors of the disorder, while polygenic risk scores (PRSs) derived from MDD and its genetically correlated traits are weaker but stable predictors. Additionally, childhood trauma (CT) is a well-established risk factor for MDD. This study aimed to jointly model the predictive effect of multi-family history (mFH), multi-PRS (mPRS) and childhood trauma on MDD. The phenotypic and genetic data came from the NIHR BioResource Genetic Links to Anxiety and Depression (GLAD) study and other NIHR BioResource cohorts. The diagnosis of MDD was based on DSM-V algorithms, utilizing data from online questionnaire. Family history was defined as first-degree relatives with MDD or other 21 psychiatric disorders. MegaPRS was used to calculate PRSs based on recent large genome-wide association studies. Reported childhood trauma was identified when participants responded affirmatively to at least one item from a 5-item childhood trauma screener questionnaire. We applied elastic net regression to fit models in repeated 10-fold cross validation using 80% of the samples as a training set, and tested the best-fit model on the independent 20% testing sample. Our study included 9,347 individuals with MDD and 6,415 controls who provided detailed family history and childhood trauma information, and for whom genome-wide association data were available. Single family history of MDD, PRS and CT explained 15.70%, 9.02% and 13.05% of the variance for MDD, respectively. In the training set, mFH, mFH+mPRS, mFH+mPRS+CT, and mFH+mPRS+CT+demographic factors explained 19.69%, 26.33%, 33.46%, and 38.13% of the variance for MDD, respectively. Similarly, in the independent testing set, mFH, mFH+mPRS, mFH+mPRS+CT, and mFH+mPRS+CT+demographic factors explained 21.70%, 27.61%, 33.83%, and 37.74% of the variance for MDD, with corresponding area under the receiver operating characteristic curve (AUC) of 0.73, 0.77, 0.80, and 0.81, respectively. Multi-family history improved the prediction for MDD. Multi-PRS prediction yielded additional predictive effects that was additive to multi-family history. Collectively, demographic factors, childhood trauma, multi-family history and multi-PRSs explained 38% of the variance for MDD, with an AUC over 0.80.