Predicting frailty in elderly people using socio-clinical databases

Life expectancy increases globally and brings out several challenges in almost all developed countries all around the world. Early screening allows to carry out policies to prevent adverse age-related events and to better distribute the associated financial burden on social and healthcare systems. One of the main issues correlated to a longer lifespan is the onset of the frailty condition that typically characterizes who is starting to experience daily life limitations due to cognitive and functional impairment. In literature there are many different definitions of frailty condition, mainly due to the large number of variables that can reduce the autonomy of the elderly. We extend the definition proposed by the Department of Health of the UK National Health Service, with which a frail subject is at risk of hospitalisation or death within a year. In particular, in this paper we propose a predictive model for frailty condition based on 26 variables from 11 socio-clinical databases. The model assigns a frailty risk index in the range 0 to 100 to each subject aged over 65 years old. The risk index is stratified in 5 expected risk classes and allows to carry out different tailored interventions. To evaluate our method, we use a four-year depth dataset of about 100 000 over 65 subjects. The obtained results are compared to a dataset of patients being treated in local health services.