Assessing Follow-Up Duration for Reliable Cognitive Aging Effect Estimates
Overview
This study evaluated how varying lengths of follow-up in longitudinal cognitive aging research affect the accuracy and precision of effect estimates for dementia risk factors. Using data from adults aged 65+ in the U.S. Health and Retirement Study, it was found that follow-up shorter than four waves (under 6 years) yields estimates that differ meaningfully from longer follow-up, especially in those aged 65-80. Additionally, specifying current age as the timescale improved precision with shorter follow-up compared to time-since-baseline.
Background
Longitudinal studies of cognitive decline are essential but costly and time-consuming, necessitating optimization of follow-up duration to balance resource use and data quality. The choice of timescale—time-since-baseline versus current age—affects model estimates and precision, with time-since-baseline commonly preferred to separate aging from cohort effects. However, shorter follow-up restricts variance in time-since-baseline, potentially reducing estimate precision. Real-world data are needed to understand how follow-up length and timescale choice impact effect estimates in cognitive aging research.
Data Highlights
Age Group at Baseline
Follow-Up Length (waves)
Effect Estimate Difference
Variance Difference
Precision (RMSE)
65-80 years
<4 waves
Meaningful divergence from full 7-wave model
Increased variance compared to full follow-up
Lower precision
>80 years
<4 waves
Less pronounced differences due to attrition
Smaller variance differences
Moderate precision
65-80 years
Short follow-up
Current age timescale estimates differ from time-since-baseline
Current age timescale more precise
Improved precision with current age timescale
Key Findings
Models with fewer than 4 waves of follow-up (<6 years) produce effect estimates for cognitive decline that differ significantly from those using full 7-wave (12-year) follow-up in adults aged 65-80.
Differences in estimates by follow-up length are less pronounced in adults older than 80 at baseline, partly due to selective attrition.
Using current age as the timescale yields more precise estimates than time-since-baseline, especially with shorter follow-up durations.
Time-since-baseline remains preferred for separating cohort and aging effects but may suffer from reduced variance and precision with limited follow-up.
Root mean square error comparisons indicate a trade-off between bias and variance depending on follow-up length and timescale specification.
Clinical Implications
Researchers designing cognitive aging studies should consider that follow-up shorter than approximately 6 years may yield biased and less precise estimates of risk factor associations with cognitive decline, particularly in younger elderly populations. When follow-up duration is limited, specifying current age as the timescale may improve estimate precision, though it may conflate cohort and aging effects. These insights can guide study design decisions balancing follow-up length, sample size, and analytic approach to optimize data quality and resource use.
Conclusion
Reliable estimation of cognitive decline associations requires sufficient longitudinal follow-up, with at least four waves recommended for adults aged 65-80. Timescale specification influences precision, with current age offering advantages in shorter studies, highlighting the need to tailor analytic strategies to study design constraints.
References
Health and Retirement Study 2006-2018 -- Longitudinal Cognitive Aging Data
Prior simulation studies (References 1-3) -- Effects of follow-up length on power
Timescale specification literature (References 10-16) -- Modeling cognitive decline
