Is Temporal Variability a Standalone Predictor in Medical Data? An Actigraphy Study in Bipolar Disorder

Area of Science:

• Biomedical Engineering
• Chronobiology
• Psychiatry

Background:

• Measures of central tendency and temporal variability are common biomarkers in medical data.
• Skewed distributions in physiological data can reduce the interpretability of standard metrics like mean and standard deviation (SD).
• Actigraphy data is increasingly used to monitor disease dynamics, but the role of temporal variability requires further investigation.

Purpose of the Study:

• To determine if temporal variability in actigraphy data contains outcome-relevant information for bipolar disorder (BD).
• To assess whether the predictive capacity of temporal variability is confounded by mean activity levels in BD patients.

Main Methods:

• Analysis of actigraphy recordings from 326 individuals with BD.
• Statistical modeling using a subset of 34 participants with both manic and remission periods.
• Power transformation (Box-Cox, Yeo-Johnson) of daily features, aggregated weekly into mean (μ7) and SD (σ7).
• Mixed-effects logistic regression models to differentiate manic from remission weeks.

Main Results:

• Power transformations decreased the correlation between mean (μ7) and SD (σ7) of activity levels.
• Temporal variability (σ7) retained significant outcome-relevant predictive information for most actigraphy features.
• For a subset of features (27%), the predictive impact of variability was partially explained by mean-variance coupling.

Conclusions:

• Temporal variability in actigraphy data provides valuable predictive information for clinical states in bipolar disorder.
• While mean activity levels can influence variability, temporal variability remains a significant independent predictor for most features.
• Further research into mean-variance coupling is warranted to fully understand actigraphy-based biomarkers in BD.