Development and validation of a 2-step shared parameter model for dementia imputation in the Cardiovascular Health...

Area of Science:

• Epidemiology
• Gerontology
• Biostatistics

Background:

• Large-scale dementia ascertainment for research presents significant challenges.
• The Cardiovascular Health Study (CHS) provides a valuable dataset for exploring dementia.
• Developing efficient imputation methods is crucial for advancing dementia research.

Purpose of the Study:

• To demonstrate a novel method for imputing dementia status and onset time using existing data.
• To validate the accuracy and performance of the imputation method.
• To apply the method to a large cohort for enhanced dementia research capabilities.

Main Methods:

• Utilized a linear mixed effects model to estimate individual cognitive trajectories.
• Employed an accelerated failure time model incorporating cognitive estimates to impute dementia onset.
• Calibrated and validated the imputation model using a sub-study with confirmed dementia classifications.

Main Results:

• The imputation model demonstrated high specificity (98.5%) and accuracy (91.3%) in the validation sample.
• Sensitivity was modest (43.8%), with imputed onset times within +/-1.5 years of classified onset.
• The method successfully classified an additional 16.0% of participants without prior dementia classifications as having dementia.

Conclusions:

• The shared parameter approach is feasible in cohorts with cognitive data and a validation subset.
• The method offers high overall accuracy and specificity for dementia ascertainment.
• This approach provides an alternative to administrative data linkage for large-scale dementia research.