Abstract
The ageing population imposes increasing pressure on healthcare systems, requiring more extensive care for older people and early identification of frail individuals to reduce the risk of adverse health events. The aim of this work is to develop an indicator to assess the frailty level of each individual using the administrative health database of ULSS6 Euganea, an Italian healthcare local authority. Given the multidimensional nature of frailty, a multi-outcome approach has been adopted, considering six outcomes: death, emergency room visits with a red code, hip fracture, hospitalization, disability, and dementia. After selecting a subgroup of frailty determinants for each adverse event using gradient boosting approach, six classification rules were estimated through outcome-specific logistic regression models. The frailty indicator was created by combining these classification rules, weighted according to their individual predictive capacity. The indicator shows good performance across all outcomes and allows for the use of different subgroups of frailty determinants specific to each outcome, including the subject’s gender, a factor excluded in other indicators already known in the literature.