Mixed effect models of longitudinal Alzheimer's disease data: a cautionary note

Area of Science:

• Neuroscience
• Biostatistics
• Gerontology

Background:

• Longitudinal studies are crucial for understanding Alzheimer's disease (AD) progression.
• Accurately measuring the onset of cognitive decline is a significant challenge in AD research.
• Current methods using diagnosis or enrollment dates can introduce bias in non-linear decline rates.

Purpose of the Study:

• To compare the sensitivity of summary measure analyses versus mixed effects models in detecting cognitive decline associations.
• To evaluate the impact of biased time-of-decline measurements on analytical techniques.
• To determine which analytical approach is more robust to inaccuracies in measuring the start of cognitive decline.

Main Methods:

• Simulated cognitive decline trajectories using a multivariate normal random effect model.
• Tested the ability of summary measure analyses and mixed effects models to differentiate true from spurious associations.
• Assessed the impact of biased time-of-decline measurements on the detection of associations.

Main Results:

• Summary measure models demonstrated reduced likelihood of detecting spurious associations caused by biased time measurements.
• Summary measure models showed increased ability to detect true associations that might be obscured by biased time data.
• Mixed effects models were more susceptible to biases introduced by inaccurate timing of cognitive decline onset.

Conclusions:

• Summary measure analyses are more robust to inaccuracies in determining the onset of cognitive decline in Alzheimer's disease studies.
• These findings suggest summary measure approaches may provide more reliable insights into AD progression and its underlying biology.
• The study highlights the importance of analytical method selection for accurate interpretation of longitudinal AD data.