JMIR Publications

ABSTRACT

Background:

Frailty is one of the most important age-related conditions in older adults. It is defined as a syndrome of physiological decline in late life, characterized by marked vulnerability to adverse health outcomes. A clear operational definition of frailty, however, has not been agreed so far. There is a wide range of studies on the detection of frailty and their association with mortality. Several of these studies have focused on the possible risk factors associated with frailty in the elderly population, but predicting who will be at increased risk of frailty is still overlooked in clinical settings. Also, no applications of machine learning have been proposed to predict frailty.

Objective:

The objective of our study was to develop predictive models for frailty conditions in older people using different machine learning methods based on a database of clinical characteristics and socioeconomic factors.

Methods:

The study was based on administrative health databases containing 58 input and 6 output variables. We first identify and define six problems as surrogates of frailty. We then resolve the extremely imbalanced nature of the data and a comparative study between the different machine learning algorithms – Artificial neural network(ANN), Genetic programming (GP), Support vector machines (SVM), Random Forest (RF) and Decision tree (DT) – was carried out. The performance of each model was evaluated using a separate unseen dataset.

Results:

For mortality, disability and fracture problems, ANN represented higher values of average performance (sensitivity = 0.81; specificity = 0.77; Accuracy = 0.78) compared to the other machine learning models. SVM also produced slightly similar results (sensitivity = 0.79; specificity = 0.77; Accuracy = 0.76). For the rest of the problems, GP yields better performance to detect the frail subjects with average sensitivity =0.76. The lowest result was obtained with DT in all problems, while RF produced an acceptable performance.

Conclusions:

Our models have good performance in predicting frailty conditions (mortality, urgent hospitalization, disability, fracture, and emergency admissions). Through further enhancement, the models could serve as a foundation for developing decision-support tools to improve early identification and prediction of frail older adults.