Controlling for selective dropout in longitudinal dementia data: Application to the SveDem registry

Area of Science:

• Gerontology
• Neuroscience
• Biostatistics

Background:

• Loss to follow-up is a significant challenge in longitudinal dementia research.
• Cognitive function in dementia patients declines over time, making dropout a critical factor.
• Understanding and mitigating dropout bias is essential for accurate research findings.

Purpose of the Study:

• To characterize dropout and missing cognitive data within the Swedish dementia registry (SveDem).
• To identify key factors associated with participant dropout in dementia studies.
• To apply propensity score methods, specifically inverse probability of censoring weights (IPCWs), to adjust for dropout bias.

Main Methods:

• Utilized longitudinal cognitive data from 53,880 individuals in the SveDem national quality dementia registry.
• Estimated inverse probability of censoring weights (IPCWs) using a logistic regression model to account for dropout.
• Compared complete case analysis with IPCW-adjusted analysis for cognitive change.

Main Results:

• The mean annualized rate of change in Mini-Mental State Examination (MMSE) scores for individuals with low MMSE (0-10) was significantly different between analyses.
• Complete case analysis underestimated cognitive decline (+1.5 points/year) compared to the IPCW analysis (-0.3 points/year).
• IPCW adjustment yielded more plausible estimates of cognitive decline.

Conclusions:

• Adjusting for dropout using IPCWs provides more accurate and plausible estimates of cognitive decline in longitudinal dementia studies.
• This methodology is valuable for correcting biased dropout in dementia cohort research.
• The findings support the use of IPCWs to enhance the reliability of dementia research.