To facilitate the use of the International Classification of Functioning, Disability and Health (ICF), linkage rules were developed. But these rules do not provide statistical methods for the linkage process. Thus, the aim of this study was to facilitate the use of the ICF by linking patient questionnaires to the ICF categories via using machine learning algorithms.

A total of 244 chronic low back pain patients (52% female; 49(±18) years) participated in interviews that assessed the activities and participation component of the ICF brief core set for low back pain and completed the Roland-Morris disability questionnaire, the Pain Disability Index, and rated their pain intensity on a visual analogue scale before and after a 6-month rehabilitative training. Random forest models with dichotomized ICF categories (no impairment versus impairment) as response variable and the predictor variables age, sex and the items of the disability questionnaires were built to capture the relationship between the predictor variables and the response variable.

Random forest models revealed values that indicated satisfactory performance measures (accuracy:>0.62, AUC:>0.70, kappa:>0.17) for the brief core set categories d240 ‘handling stress and other psychological demands’, d410 ‘changing basic body position’, d415 ‘maintaining a body position’, d430 ‘lifting and carrying objects’, d450 ‘walking’, d540 ‘dressing’, d640 ‘doing housework’, d845 ‘acquiring, keeping and terminating a job’ and d850 ‘remunerative employment’.

ICF linkage based on machine learning models may be successfully administered to the most relevant ICF brief core set categories for low back pain. This study proposed a new and innovative linking approach that could be further integrated in an electronic ICF sheet to facilitate the administration of the ICF in clinical practice and enable a universal and standardized language for the description of functioning, disability and health.