Cosinor analysis for temperature time series data of long duration

Area of Science:

• Chronobiology
• Time Series Analysis
• Biostatistics

Background:

• Cosinor models are widely used for analyzing biological rhythms in time series data.
• Long time series present challenges for cosinor model accuracy due to inherent noise and parameter drift.
• The reliability of cosinor model outputs, particularly amplitude, is sensitive to the length of the analyzed data.

Purpose of the Study:

• To investigate the impact of time series length on the performance and accuracy of cosinor models.
• To demonstrate the sensitivity of cosinor model parameters, specifically amplitude and model-fit, to data length.
• To explore methods for improving cosinor analysis in long time series, including serial-sectioning and data folding.

Main Methods:

• Application of cosinor models to long time series data, including body temperature from human and animal subjects.
• Analysis of data length effects on amplitude and model-fit.
• Evaluation of serial-sectioning and single-cycle folding techniques for noise and drift reduction.

Main Results:

• Cosinor model performance deteriorates with increasing time series length due to noise and parameter drift.
• Amplitude and model-fit are significantly sensitive to data length, impacting inter-study comparisons.
• Serial-sectioning and cycle folding improve model-fit but can alter amplitude values.

Conclusions:

• Amplitude comparisons between studies using cosinor analysis must account for differences in data length.
• Serial-sectioning and cycle folding are valuable techniques for enhancing cosinor model-fit and tracking parameter changes in long time series.
• While autoregressive residuals minimally affect central parameter values, noise and drift necessitate careful data handling in cosinor analysis.