Semiparametric regression for periodic longitudinal hormone data from multiple menstrual cycles

Area of Science:

• Statistics
• Biostatistics
• Longitudinal Data Analysis

Background:

• Longitudinal data analysis requires specialized models for periodic patterns.
• Existing methods may not adequately capture complex within-subject correlations in cyclical data.
• Comparing time-dependent profiles between groups is crucial in many scientific fields.

Purpose of the Study:

• To develop a semiparametric regression model for periodic longitudinal data.
• To accurately model covariate effects, time-dependent trends, and within-subject correlations.
• To introduce a statistical test for comparing time profiles between two groups.

Main Methods:

• Utilized parametric fixed effects for covariate modeling.
• Employed periodic nonparametric smooth functions for time effects.
• Modeled within-subject correlation using random effects and a periodic variance function.
• Estimated parameters via maximum penalized likelihood and restricted maximum likelihood.
• Developed a scaled chi-squared test for comparing nonparametric time functions.

Main Results:

• The proposed estimator for the time function is a periodic cubic smoothing spline.
• All model parameters are efficiently estimated using a linear mixed model.
• A novel scaled chi-squared test was developed for comparing group time profiles.
• The model and test demonstrated effective application in analyzing hormone data.

Conclusions:

• The developed semiparametric model provides a flexible framework for analyzing periodic longitudinal data.
• The methodology allows for robust estimation of covariate effects and time trends.
• The proposed test enables reliable comparison of time profiles between groups, enhancing scientific discovery.