IDENTIFying the risk of recurrent and chronic Youth Depression (IDENTIFYD): A prospective longitudinal cohort study in help seeking young people aged 15 - 25

The primary aim of the study is to develop a multivariable prediction model, with the specified primary and secondary outcomes as the variables we aim to predict. Participants in this study will be required to fill out the following surveys/questionnaires: Baseline survey (administered through an online survey and takes participants about 2.5 hours to complete): • Demographics: age, gender, sexual orientation, indigenous status, income and education/occupation (self and parents), education, postcode, ethnicity, country of birth, living alone or with others (custom questionnaire) • Psychiatric and medical history, and family history of mental disorders (custom questionnaire) • Treatment history (custom questionnaire) • Online survey version of the Composite International Diagnostic Interview (CIDI) major depressive disorder section • Quick Inventory of Depressive Symptoms (QIDS) • SCAARED (anxiety symptoms) • Generalised Anxiety Disorder Assessment 7 Item (GAD-7) • suicidal thoughts and behaviours (STB), assessed through a custom build STB questionnaire • psychotic symptoms assessed with the Prodromal Questionnaire (PQ-16) • manic symptoms, assessed with the Mood Disorders Questionnaire (MDQ) • social and occupational functioning assessed by the Filia Social Inclusion Measure (F-SIM16) • Disordered eating behaviours assessed by the Eating Disorder Examination Questionnaire – Short 13 item (EDE-QS-13) • Substance use assessed by the Alcohol, Smoking, Substance Involvement Screening Test (ASSIST) • Childhood Trauma Questionnaire (CTQ) • Multidimensional Scale of Perceived Social Support (MSPSS) • Perceived Stress Scale (PSS) • Penn State Worry Questionnaire (PSWQ) • Pittsburgh Sleep Quality Index (PSQI) • The Brief Core Schema Scale (BCSS) • Cognitive Emotion Regulation Questionnaire (CERQ-Short) • UCLA Loneliness Scale • Multidimensional Peer-Victimisation Scale 24 (MPVS-24) • Neuropsychological Symptom Self Report (NSSR) • Eysenck Personality Questionnaire – Neuroticism Subscale (EPQ-S) • Physical Symptoms (PHQ-15) • Personality Inventory for DSM-5 Brief Form (PIDF-5-BF) • General Self Efficacy (GSE) • Anxiety Sensitivity Index (ASI) • Cognitive Behavioural Avoidance Scale (CBAS) • Pain (custom questionnaire) • Minimal Instructions Autobiographical Memory Task (MI-AMT) • Autism Quotient Scale – 10 (ASQ-10) • Brief Family Relationship Measure “Angry Conflict Within Family” Subscale The PQ-16, MDQ, F-SIM16, EDE-QS-13 and ASSIST are repeated through online surveys every 6 months for 18 months (taking 30 min to complete), while the CIDI major depressive disorder section, QIDS, SCAARED, GAD-7and custom build STB questionnaire are repeated every 2 months for 18 months (taking about 10 min to complete). Optional baseline measures: - Genetic predictors based on saliva samples (optional opt-in component: saliva samples are sent to participants home and send back by participants to research team; time to complete ~15 min) Optional time-varying measures: - Daily mobile phone survey with single item questions related to affect, stress, worry, loneliness, self-compassion, social support, rumination, self-confidence, cognition, substance use, sleep, positive events, negative events, bullying and somatic complaints (takes about 2 minutes to complete) Development of the prediction model: Using IDENTIFYD data, we will train multiple submodels using distinct data modalities (see below). Submodel predictions will be combined using ensemble techniques. This modular approach enables omission of any data modality, supporting flexible deployment based on feasibility and stakeholder preferences. Submodels, reflecting differences in data availability due to optional study components will include: - Submodel 1: clinical, demographic, psychosocial, interpersonal baseline predictors - Submodel 2: (epi)genetic predictors - Submodel 3: time-varying self-report features derived from the daily phone surveys (e.g., mood, sleep) Models will be trained and compared across a range of algorithms (e.g. ElasticNet, XGBoost, SVMs, random forests, deep learning). Embedded or wrapper-based feature selection will remove non-essential features across submodels. Submodels will be trained using nested cross-validation. Performance (e.g., AUC, sensitivity, specificity) and calibration will be assessed. Various submodel combinations will be ensembled and validated on external datasets (to be identified).