Time-varying functional regression for predicting remaining lifetime distributions from longitudinal trajectories.

Area of Science:

• Biodemography
• Gerontology
• Biostatistics

Background:

• Longitudinal studies frequently examine the link between lifespan and behavioral patterns.
• Predicting age-at-death often relies on limited covariate data, overlooking historical context.

Purpose of the Study:

• To develop a novel statistical technique for predicting age-at-death distributions.
• To incorporate entire covariate histories into lifespan prediction models.
• To estimate remaining lifetime distributions for individuals based on their behavioral trajectories.

Main Methods:

• Utilized time-varying functional principal component scores to represent covariate trajectories.
• Employed a class of time-varying functional regression models.
• Applied dimension-reduction techniques by projecting onto a single index.
• Demonstrated with longitudinal daily egg-laying data from female medflies.

Main Results:

• Successfully predicted age-at-death and remaining lifetime distributions for individual subjects.
• Provided estimates for quantiles and prediction intervals of remaining lifetime.
• Showcased the method's applicability to biodemographic data with event histories.

Conclusions:

• The proposed method offers a robust approach to lifespan prediction using comprehensive behavioral data.
• This technique enhances understanding of aging and longevity by integrating historical covariate information.
• The findings have implications for biodemographic research and personalized longevity predictions.