E1.L3

DATASET DESCRIPTION

 Disease Area

Mental Health — Major Depressive Disorder (MDD)

 Specific Research Focus

Early detection, risk stratification, and longitudinal forecasting of adolescent-onset Major Depressive Disorder using multi-domain clinical, behavioral, and digital biomarker data

 Dataset Theme / Edition

Synthetic Longitudinal Adolescent Psychiatry 

 Short Dataset Summary

SynthMDD-Adolescent-v1.0 is a large-scale synthetic longitudinal clinical dataset of 50,000 adolescents (ages 10–19) with 798,122 visit-level records across 101 features. Every MDD case is fully DSM-5 compliant (PHQ-9 ≥10, anhedonia, sleep disturbance enforced). Each patient has 8–24 irregularly spaced clinical visits simulating real-world care schedules, with modeled disease progression, intervention response lag, seasonal mood effects, abrupt deterioration events, and recovery arcs — across 5 clinical site types.

 Intended Research Applications

🏥 Early MDD screening at first clinical contact

📊 Risk stratification and PHQ-9 severity regression

⏱️ Longitudinal forecasting with LSTM / Transformer models

🔍 XAI / SHAP feature attribution for psychiatric AI

🤝 Federated learning across simulated clinical sites

🧬 TimeGAN and conditional synthetic data generation

⚖️ Survival analysis and time-to-event modeling

🏫 School-based intervention policy modeling

 Total Number of Patients

50,000 adolescent patients · 798,122 total visit-level rows · avg. 16 visits per patient

 Number of Features / Parameters

101 total features — 29 static (patient-level) + 72 dynamic (visit-level)

Spanning: validated clinical scores · psychological indicators · lifestyle & behavioral signals · school & social environment · digital & passive behavioral markers · treatment & intervention records · longitudinal monitoring · 12 outcome/target columns

 Data Modalities Included

📝 Structured clinical assessments (PHQ-9, GAD-7, BDI-II, PSQI, Cognitive score)

🧬 Genetic & family risk scores

📱 Digital behavioral signals (screen time, late-night device activity, typing variability)

⌚ Wearable-derived activity data (steps/day)

🏫 School & social environment metrics (GPA, attendance, bullying, peer quality)

💊 Treatment & intervention records (CBT status, SSRI, therapy adherence)

🕐 Longitudinal temporal sequences (irregular visit intervals, seasonal effects)

🧪 Adverse childhood experiences (ACE score, trauma flags)

 Recommended AI Tasks

Binary Classification (MDD detection)

Multi-class Ordinal Regression (PHQ-9 band, Progression Stage)

Sequence Modeling / Time-Series Forecasting (LSTM, Transformer, S4)

Survival Analysis / Time-to-Event Prediction

Generative Modeling (TimeGAN, CVAE, Diffusion)

Federated Learning (5-site simulation)

Explainability / SHAP Attribution

Causal Inference (treatment effect estimation)

 Dataset Size

| Longitudinal CSV | 798,122 rows × 101 cols · 356 MB |

| Static Features CSV | 50,000 rows × 29 cols · 6.3 MB |

| Dynamic Features CSV | 798,122 rows × 77 cols · 271 MB |

| NumPy Padded Tensor | (50,000 × 24 × 63) · 302 MB |

| Sequence Mask (.npy) | (50,000 × 24) · 1.2 MB |

| Data Dictionary | 101 features fully documented |

 Release Date

May 2026

 Why This Dataset Exists

Real adolescent psychiatric datasets are among the most access-restricted in clinical AI — blocked by HIPAA, FERPA, and child privacy regulations — leaving researchers without large, labeled, longitudinal corpora to build and validate early-warning models. This dataset was built to close that gap: a privacy-safe, DSM-5-compliant, multi-domain dataset that mirrors the complexity of real clinical data without exposing a single real patient. It ships in every format modern ML pipelines need, from raw CSV to padded tensors, so teams can go from download to model training without preprocessing friction.

 Real-World Problem Addressed

Adolescent depression is one of the most under-detected, under-treated conditions globally — with 50% of lifetime mental health disorders emerging before age 14, and average delays of 11 years between symptom onset and first treatment. Existing clinical tools rely on self-report questionnaires administered only when a patient is already presenting in distress. This dataset enables researchers to build AI models that can detect MDD risk earlier — from passive digital signals, school performance trends, and behavioral patterns — before a crisis occurs, and to do so equitably across demographic groups without requiring access to sensitive real patient records.