From Symptom Snapshots to Symptom Trajectories: Verified Analysis Code and Reproducibility Materials
Year: 2026
Type: software
Abstract: Verified analysis code and reproducibility materials accompanying the manuscript “From Symptom Snapshots to Symptom Trajectories: Machine Learning Prediction of Near-Term Suicidal Ideation From Intensive Longitudinal Depression Assessments.” This package reproduces the complete secondary-analysis workflow, including PHQ symptom recoding, NAMU-style feature engineering, temporal feature construction, participant-level locked holdout evaluation, XGBoost modeling, participant-cluster bootstrap comparisons, incident suicidal-ideation onset analysis, temporal forward validation, SHAP feature importance, and manuscript figures. The analyses were verified against the publicly available openESM Marian dataset (0052_marian). The verified analytic sample included 6,845 transitions from 144 participants. The Temporal XGBoost model showed the highest locked-test discrimination (PR AUC = .581; ROC AUC = .891). The underlying participant-level dataset is not redistributed in this archive. It is publicly available through the openESM Zenodo record at https://doi.org/10.5281/zenodo.17348268. The original study is Marian et al. (2023), https://doi.org/10.1007/s10862-022-10014-8. The archive contains the exact locked-test participant IDs, frozen model settings and thresholds, analysis scripts, participant-cluster bootstrap procedures, aggregate results, reproducibility checks, software-environment information, and file checksums. Participant-level predictions, row-level SHAP values, and other sensitive derived outputs are not included. This package is intended for research reproducibility and methodological evaluation. The models are not intended for stand-alone clinical suicide-risk assessment or treatment decisions.
Author: woo, sungbum
Institution: Hanyang Cyber University
Language: English
Subfield: Experimental and Cognitive Psychology
Field: Psychology
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7203974875