Models for estimating the change-point in gas exchange data

Area of Science:

• Exercise Physiology
• Biostatistics
• Sports Science

Background:

• The gas exchange threshold indicates metabolic acidosis onset during incremental exercise tests.
• Current methods use two-line regression models to locate this threshold, relating carbon dioxide output (VCO2) to oxygen uptake (VO2).
• These models have limitations due to VO2's non-monotone relationship with time.

Purpose of the Study:

• To propose and validate more appropriate statistical models for identifying the gas exchange threshold.
• To utilize novel statistical methodologies for change-point detection in exercise physiology.
• To improve the accuracy of pinpointing the onset of metabolic acidosis during exercise.

Main Methods:

• Application of hidden Markov models to demonstrate the existence of a change-point.
• Utilizing time series models to estimate the change-point's position.
• Considering non-multivariate normal distributions and modeling time-dependent variance in VCO2.

Main Results:

• Demonstrated the existence of a change-point using hidden Markov models.
• Estimated the change-point location using time series analysis.
• Successfully modeled increasing VCO2 variance over time in some subjects.

Conclusions:

• Statistical models with change-points set on time are more appropriate than those based on VO2.
• Hidden Markov and time series models offer a robust approach to analyzing gas exchange data.
• This methodology enhances the precise detection of the gas exchange threshold and metabolic acidosis onset.