Machine learning approaches to predict rehabilitation success based on clinical and patient-reported outcome measures

A common way to treat hip, knee or foot injuries is by conducting a corresponding physician-guided rehab over several weeks or even months. While health professionals are often able to estimate the treatment success beforehand to a certain extent based on their experience, it is scientifically still not clear to what extent relevant factors and circumstances explain or predict rehab outcomes. To this end, we apply modern machine learning techniques to a real-life dataset consisting of data from more than a thousand rehab patients (N = 1,047) and build models that are able to predict the rehab success for a patient upon treatment start. By utilizing clinical and patient-reported outcome measures (PROMs) from questionnaires, we compute patient-related clinical measurements (CROMs) for different targets like the range of motion of a knee, and subsequently use those indicators to learn prediction models. While we at first apply regression algorithms to estimate the rehab success in terms of percental admission and discharge value differences, we finally also utilize classification models to make predictions based on a three-classed grading scheme. Extensive evaluations for different treatment groups and targets show promising results with F-scores exceeding 65% that are able to substantially outperform baselines (by up to 40%) and thus show that machine learning can indeed be applied for better medical controlling and optimized treatment paths in rehab praxis. Future developments should include further relevant critical success criteria in the rehabilitation routine to further optimize the prognosis models for clinical practice.

Abstract licence: CC BY-NC-ND

1. 13.

   Validity and test-retest reliability of manual goniometers for measuring passive hip range of motion in femoroacetabular impingement patients. · 2010

39 more references, whose wording the paper’s licence keeps private.

A person counts once a day per paper. We keep no cookie and no address: a reader is a hash that changes every day. Any month under five reads shows no number.