LongCriSP: a test for bump hunting in longitudinal data.

Area of Science:

• Statistics
• Biostatistics
• Longitudinal Data Analysis

Background:

• Detecting local extrema is crucial for understanding complex data patterns.
• Existing methods for detecting local extrema are not always suitable for longitudinal data.
• Computational efficiency is a challenge when analyzing large longitudinal datasets.

Purpose of the Study:

• To extend the Harezlak and Heckman test for local extrema detection to the longitudinal data setting.
• To introduce a computationally efficient statistical test for longitudinal data analysis.
• To assess the behavior of body mass index (BMI) in former prisoners of war (POWs).

Main Methods:

• Utilized penalized spline regression techniques for computational efficiency.
• Developed a novel statistical test named LongCriSP.
• Employed a smoothed bootstrap method for estimating p-values.
• Applied the test to longitudinal BMI measurements of Vietnam War POWs.

Main Results:

• The LongCriSP test demonstrated generally conservative properties in simulations.
• The test achieved a power exceeding 70% at the alpha = 0.1 nominal level across various settings.
• Analysis of former POWs' BMI data indicated non-monotone population curve behavior.

Conclusions:

• The proposed LongCriSP test is a computationally efficient tool for detecting local extrema in longitudinal data.
• The findings suggest that the mean population BMI curve for former Vietnam POWs exhibits non-monotone characteristics.
• This research provides a valuable method for analyzing complex patterns in longitudinal health data.