Detection of Frailty Using Genetic Programming The Case of Older People in Piedmont, Italy

Frailty appears to be the most problematic expression of elderly people. Frail older adults have a high risk of mortality, hospitalization, disability and other adverse outcomes, resulting in burden to individuals, their families, health care services and society. Early detection and screening would help to deliver preventive interventions and reduce the burden of frailty. For this purpose, several studies have been conducted to detect frailty that demonstrates its association with mortality and other health outcomes. Most of these studies have concentrated on the possible risk factors associated with frailty in the elderly population; however, efforts to identify and predict groups of elderly people who are at increased risk of frailty is still challenging in clinical settings. In this paper, Genetic Programming (GP) is exploited to detect and define frailty based on the whole elderly population of the Piedmont, Italy, using administrative databases of clinical characteristics and socio-economic factors. Specifically, GP is designed to predict frailty according to the expected risk of mortality, urgent hospitalization, disability, fracture, and access to the emergency department. The performance of GP model is evaluated using sensitivity, specificity, and accuracy metrics by dividing each dataset into a training set and test set. We find that GP shows competitive performance in predicting frailty compared to the traditional machine learning models. The study demonstrates that the proposed model might be used to screen future frail older adults using clinical, psychological and socio-economic variables, which are commonly collected in community healthcare institutions.