Apparent predictive value of single-session smartwatch biosignals for depressive symptom severity depends on validation design: analysis code (Zenodo 2026)

Analysis code for a 29-participant single-lab-session study using raw Galaxy Watch 5 IR PPG (plus ECG, SpO2 and bioimpedance exports). It runs 30-s epochs with template QC, then derives pulse rate, PRV and pulse-morphology features. Analyses: associations with PHQ-9, BDI-II and PSS-10; leave-one-subject-out prediction against mean/median/age-sex baselines (ridge, RF); permutation and Freedman–Lane tests; and two deliberately optimistic validation designs (record-wise epoch splits and test-set feature selection) with shuffled-score controls.

• Form factor: wrist (Samsung Galaxy Watch 5)

Analysis code accompanying the manuscript "Apparent predictive value of single-session smartwatch biosignals for depressive symptom severity depends on validation design: an exploratory study" (Cho and Kim; submitted).

The code processes raw infrared photoplethysmography (PPG) exports recorded with a Galaxy Watch 5 during a single laboratory session (30-s epochs, template-based quality control), derives pulse rate, pulse rate variability and pulse-morphology features, and reproduces the reported analyses: signal validity and reliability; Spearman and sex- and age-adjusted partial Spearman associations with PHQ-9, BDI-II and PSS-10; leave-one-subject-out prediction with mean, median and age-and-sex baselines, ridge regression and random forests; permutation and Freedman–Lane tests of incremental value; two optimistic validation designs (record-wise splits of epochs and test-set feature selection) with shuffled-score controls; and the manuscript figures.

Usage: run_all.py runs the full pipeline from raw exports (arguments –data and –out; optional –labels, –from and –only). See README.md for details.

Data are NOT included. Participant data are subject to informed-consent and institutional review board restrictions (KYU IRB 2023-07-016-002; 29 participants). The data are available from the corresponding author upon reasonable request, subject to approval by the Konyang University Institutional Review Board.

Environment: tested on Windows 11 with Python 3.13 (NumPy 2.4, pandas 3.0, SciPy 1.17, scikit-learn 1.8, Matplotlib 3.10). A full run takes about 11 minutes.

Verbatim from DataCite (opens in a new tab) · Licence: https://opensource.org/licenses/MIT

No results from this paper on ChatPPG leaderboards yet. Browse the leaderboards.