Analysis strategies for serial multivariate ultrasonographic data that are incomplete

Area of Science:

• Cardiovascular research
• Medical imaging analysis
• Biostatistics

Background:

• Carotid artery intima-media thickness (CIMT) ultrasonography is crucial for assessing atherosclerosis.
• Clinical trials often face challenges with incomplete serial CIMT data due to measurement variability.

Purpose of the Study:

• To compare analytical approaches for handling missing multivariate serial ultrasound data.
• To evaluate the efficiency of maximum likelihood methods versus traditional regression techniques.

Main Methods:

• Simulated ultrasound data from the Asymptomatic Carotid Artery Plaque Study were used.
• Conditional and unconditional maximum likelihood approaches were contrasted with unweighted and generalized least squares regression.
• The impact of deviations from the missing at random assumption was assessed.

Main Results:

• Maximum likelihood-based analyses demonstrated increased efficiency compared to ignoring missing data.
• Efficiency gains of up to 21% were observed using maximum likelihood methods.
• The relative impact of deviations from the missing at random assumption was evaluated for each approach.

Conclusions:

• Maximum likelihood methods offer a more efficient way to analyze serial ultrasound data with missing measurements.
• These advanced statistical techniques can improve the precision of atherosclerosis progression assessments in clinical studies.
• Careful consideration of missing data mechanisms is important when selecting analytical methods.