application of machine learning techniques to physical and rehabilitative medicine

Abstract

Nowadays, digital information has increased exponentially in every field to such an extent that it generates huge amounts of electronic data, namely Big Data. In the field of Artificial Intelligence, Machine Learning can be exploited in order to transform the large amount of information to improve decision-making. We retrospectively evaluated the data collected from 2016 to 2018, using the database of approximately 4000 rehabilitation hospital discharges (SDO) of the Latium Region (Italy). Three models of machine learning algorithms were considered: Support of vector machine; Neural networks; Random forests. Applying this model, the estimate of the average error is 9.077, and specifically, considering the distinction between orthopedic and neurological patients, the average error obtained is 7.65 for orthopedic and 10.73 for neurological patients. SDO information flow can be used to represent and quantify the potential inadequacy and inefficiency of rehabilitation hospitalizations, although there are limitations such as the absence of description of pre-pathological conditions, changes in health status from the beginning to the end of hospitalization, specific short- and long-term outcomes of rehabilitation, services provided during hospitalization, as well as psycho-social variables. Furthermore, information from wearable devices capable of providing clinical parameters and movement data could be integrated into the dataset.