The Importance of Age in the Prediction of Mortality by a Frailty Index: A Machine Learning Approach in the Irish Longitudinal Study on Ageing

Abstract

The quantification of biological age in humans is an important scientific endeavor in the face of ageing populations. The frailty index (FI) methodology is based on the accumulation of health deficits and captures variations in health status within individuals of the same age. The aims of this study were to assess whether the addition of age to an FI improves its mortality prediction and whether the associations of the individual FI items differ in strength. We utilized data from The Irish Longitudinal Study on Ageing to conduct, by sex, machine learning analyses of the ability of a 32-item FI to predict 8-year mortality in 8174 wave 1 participants aged 50 or more years. By wave 5, 559 men and 492 women had died. In the absence of age, the FI was an acceptable predictor of mortality with AUCs of 0.7. When age was included, AUCs improved to 0.8 in men and 0.9 in women. After age, deficits related to physical function and self-rated health tended to have higher importance scores. Not all FI variables seemed equally relevant to predict mortality, and age was by far the most relevant feature. Chronological age should remain an important consideration when interpreting the prognostic significance of an FI.

Keywords

• frailty
• age distribution
• longitudinal studies
• mortality
• supervised machine learning
• sex differences