ABSTRACT
This study investigates the relation between specific cognitive abilities and mortality using data from over 1,800 pairs of parents in the Hawaii Family Study of Cognition (HFSC; DeFries et al., 1979). Three methods of survival analysis are used: (1) Cox regression, (2) Conditional Inference Survival Trees, and (3) Random Survival Forests. The results are consistent with other studies showing that cognitive abilities contribute to the prediction of mortality above and beyond gender and education, with approximately 5% increase in prediction accuracy based on Harrell’s concordance index (Harrell et al., 1982). Specific cognitive abilities are very close in their prediction strength, and provide little to the prediction of mortality above and beyond general intelligence. Sex-specific effects may exist; survival trees suggest that numerical ability may be more important in predicting women’s mortality and that visual-spatial ability may be more important in predicting men’s mortality. Due to the high percentage of censoring (85.1%), conclusions that can be drawn from the data are limited, and the study requires additional follow up. We also demonstrate that statistical learning techniques provide new opportunities for studying survival problems, but they are still in the development stage and interpretations should be made with caution—some explicit recommendations for their practical use are discussed. We also relate these findings to other recent findings in the field of cognitive epidemiology.
