Predicting post COVID-19 rehabilitation duration with linear kernel SVR

We propose a machine learning model capable of predicting COVID-19 post-acute rehabilitation duration by means of data taken before the start of the rehabilitative path. Data from 62 patients recovering after SARS-CoV2 infection were processed in our study. Anagraphical and anamnestic data, COVID-19 signs and symptoms, COVID-19-related therapy, haematochemical findings, clinical and functional scales and therapies prescribed during the acute phase were included as predictors of the rehabilitation hospital length of stay. A set of 10 features was retained, through sequential feature selection technique, for training the model. Via Support Vector Regression, we obtained a median cross-validation absolute error of 5.91 days (IQR = 14.85 days) in predicting the duration of the COVID-19 rehabilitation. This study aims to introduce machine learning into the COVID-19 rehabilitative path definition. With COVID-19 cases still harassing the community, the model will be tested in clinical settings to test its efficiency and improve its generalization capability.