BFLCRM: A BAYESIAN FUNCTIONAL LINEAR COX REGRESSION MODEL FOR PREDICTING TIME TO CONVERSION TO ALZHEIMER'S DISEASE

Area of Science:

• Biostatistics
• Neuroimaging
• Neurodegenerative Diseases

Background:

• Alzheimer's disease (AD) poses a significant public health challenge.
• Predicting conversion from mild cognitive impairment (MCI) to AD is crucial for early intervention.
• Existing models may not fully leverage complex functional and scalar data.

Purpose of the Study:

• To develop and validate a Bayesian functional linear Cox regression model (BFLCRM).
• To identify early markers for predicting conversion from MCI to AD.
• To utilize both functional (hippocampus morphology) and scalar (MRI volumes, cognitive scores, APOE status) covariates.

Main Methods:

• Development of a Bayesian functional linear Cox regression model (BFLCRM).
• Application of the model to 346 MCI patients from the Alzheimer's Disease Neuroimaging Initiative 1 (ADNI-1) cohort.
• Utilized Markov chain Monte Carlo (MCMC) for posterior computation.
• Conducted a simulation study to assess finite sample performance.

Main Results:

• The BFLCRM accurately predicted time to AD onset.
• Functional covariates (hippocampus surface morphology) and scalar covariates (MRI volumes, ADAS-Cog, APOE status) were significant predictors.
• 161 out of 346 MCI participants progressed to AD within 48 months.

Conclusions:

• The BFLCRM is an effective tool for predicting AD conversion in MCI patients.
• The model integrates diverse data types for enhanced predictive accuracy.
• Early identification of AD risk is improved through advanced statistical modeling.