Deprisk

Predict depression risk before it manifests

Our model, a tuned gradient-boosted-trees (XGBoost) pipeline built on longitudinal data from The Maastricht Study (2010–2020 baseline), estimates the ~4-year risk of clinically relevant depressive symptoms (PHQ-9 ≥ 10) from wearable-derived sleep and activity metrics, plus clinical and sociodemographic covariates.

⚡️

Low-latency scoring

FastAPI + XGBoost ‘hist’ for fast, production-style inference.

🛡️

Privacy-first

Only derived fields leave your browser; raw Maastricht data remain on secure servers.

📊

Transparent metrics

AUROC 0.71 · AUPRC 0.29 on a hold-out test set of 2,789 records (train = 11,153).

Analytic cohort: 6,004 participants · 880 incident cases.