MS-TF-SD2: Multi-Scale Time-frequency Spectral Derivative Spike Detection for the early onset prediction of mood episodes in bipolar disorder in a remote longitudinal study using wearables

Bipolar disorder is characterized by recurrent depressive and (hypo)manic episodes that cause substantial disability, hospitalization, and elevated suicide risk yet early prodromal signals are frequently subtle and missed in standard clinical care. Our aim was to develop an algorithm that jointly handles non-stationarity, irregular sampling, and transient anomaly detection in wearable data, unveiling patterns of prodromal early warning signs that often manifest as subtle disruptions of activity, sleep, and mood preceding severe mood episodes. We developed and validated the Multi-Scale Time-Frequency Spectral Derivative Spike Detection (MS‑TF‑SD 2 ) algorithm on passively collected daily step counts (32,257 observations) and total sleep time (29,136 observations) from 127 participants followed for a mean ± SD of 383.5 ± 175 days, labeled using weekly self-reported symptom severity scales. We decomposed the raw signals into multi-scale intrinsic mode functions, estimated the instantaneous frequencies via Hilbert transform, and computed the multiscale composite anomaly scores to generate per‑window episode onset risk scores. Our algorithm anticipated depressive onsets with mean (±SD) sensitivity of 0.79 (±0.27) and median lead time 7.0 days (IQR 9.0) using day‑to‑day activity variability, and 0.80 (±0.26) with median lead 4.0 days (IQR 5.0) using day‑to‑day sleep variability. For hypomania, activity sensitivity was 0.76 (±0.33; median lead −1.0 days) and sleep sensitivity was 0.84 (±0.27; median lead −2.0 days). Our study provides a robust algorithm for extracting subtle early warning signs of episodes from passively collected wearable activity and sleep data, enabling patient-specific, just-in-time clinical decision support.

Authors

Institutions

Publication Details

Journal
Biomedical Signal Processing and Control
Published
2026-11-19
DOI
https://doi.org/10.1016/j.bspc.2026.111870
Primary Topic
Mental Health Research Topics
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

MS-TF-SD2: Multi-Scale Time-frequency Spectral Derivative Spike Detection for the early onset prediction of mood episodes in bipolar disorder in a remote longitudinal study using wearables

Arend Hintze, Ramzi Halabi, Abigail Ortiz
Biomedical Signal Processing and Control
Mental Health Research Topics
article

MS-TF-SD2: Multi-Scale Time-frequency Spectral Derivative Spike Detection for the early onset prediction of mood episodes in bipolar disorder in a remote longitudinal study using wearables

Arend Hintze, Ramzi Halabi, Abigail Ortiz
article en

Abstract

Bipolar disorder is characterized by recurrent depressive and (hypo)manic episodes that cause substantial disability, hospitalization, and elevated suicide risk yet early prodromal signals are frequently subtle and missed in standard clinical care. Our aim was to develop an algorithm that jointly handles non-stationarity, irregular sampling, and transient anomaly detection in wearable data, unveiling patterns of prodromal early warning signs that often manifest as subtle disruptions of activity, sleep, and mood preceding severe mood episodes. We developed and validated the Multi-Scale Time-Frequency Spectral Derivative Spike Detection (MS‑TF‑SD 2 ) algorithm on passively collected daily step counts (32,257 observations) and total sleep time (29,136 observations) from 127 participants followed for a mean ± SD of 383.5 ± 175 days, labeled using weekly self-reported symptom severity scales. We decomposed the raw signals into multi-scale intrinsic mode functions, estimated the instantaneous frequencies via Hilbert transform, and computed the multiscale composite anomaly scores to generate per‑window episode onset risk scores. Our algorithm anticipated depressive onsets with mean (±SD) sensitivity of 0.79 (±0.27) and median lead time 7.0 days (IQR 9.0) using day‑to‑day activity variability, and 0.80 (±0.26) with median lead 4.0 days (IQR 5.0) using day‑to‑day sleep variability. For hypomania, activity sensitivity was 0.76 (±0.33; median lead −1.0 days) and sleep sensitivity was 0.84 (±0.27; median lead −2.0 days). Our study provides a robust algorithm for extracting subtle early warning signs of episodes from passively collected wearable activity and sleep data, enabling patient-specific, just-in-time clinical decision support.

Biomedical Signal Processing and ControlVol. 131
Dalarna University (SE), Centre for Addiction and Mental Health (CA), University of Toronto (CA), The University of Texas Southwestern Medical Center (US)
Good health and well-being
Mental Health Research Topics
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon