Measuring frailty in the elderly: an indicator based on a super-classifier

Identifying frail older adults in an ageing population has become crucial to improve the services offered by a healthcare system. This work aims to develop a composite indicator to assess the level of frailty of individuals using administrative health data. Since frailty has a complex, multidimensional nature, a multi-outcome approach was used. After an extensive literature research, some health adverse events were identified to represent frailty. These adverse events were modelled by logistic classifiers, with frailty determinants (previously selected with a gradient tree boosting) used as covariates. The sensitivity and specificity of individual classifiers are exploited to rewrite their combined likelihood. From this combined likelihood, we are able to obtain an indicator capable of measuring the frailty of the population. The indicator demonstrates strong performance across all outcomes for various years. The main innovation brought by this indicator lies in the possibility to use diverse subgroups of frailty determinants specific to each outcome, without imposing constraints on their structural form. In conclusion, this indicator has proven to be an efficient tool for quantifying frailty of elderly individuals, providing potential help to health authorities in preventing frailty-related adverse events.