Improving power with repeated measures: diet and serum lipids

Area of Science:

• Nutritional Epidemiology
• Biostatistics
• Cardiovascular Health

Background:

• Cross-sectional studies struggle to link diet and serum cholesterol due to measurement error and individual variability.
• Existing statistical methods can mitigate measurement error with replicate dietary and lipid measures.
• Replicate data, even from a subsample, can improve within-subject comparisons.

Purpose of the Study:

• To evaluate statistical methods for detecting diet-serum cholesterol associations.
• To assess the impact of replicate dietary intake measures on study power.
• To determine optimal data analysis strategies in nutritional epidemiology.

Main Methods:

• Utilized a random-effects model on data from 928 participants in the San Luis Valley Diabetes Study.
• Included baseline and 4-year follow-up measures of 24-hour dietary intake and fasting lipids.
• Analyzed LDL cholesterol regression on saturated fat intake, comparing various analytical approaches.

Main Results:

• The random-effects model using all observations demonstrated a significant association (beta = 0.14, P = 0.0016).
• Restricting data to the first visit yielded non-significant results (beta = 0.05, P = 0.52).
• The random-effects model with all observations and time-varying covariates showed the greatest statistical power.

Conclusions:

• Single replicate dietary observations significantly enhance the detection of diet-serum cholesterol associations.
• Analyzing all available data, including single replicates, is superior to averaging or omitting data.
• This approach maximizes statistical power in nutritional epidemiology studies.