Using graph machine learning to identify functioning in patients with low back pain within the ICF framework

As a comprehensive perspective on functioning is useful when making patient assessments, the WHO has developed the International Classification for Functioning, Disability, and Health (ICF). However, its complex structure poses a problem for implementation as part of clinical practice.The aim of this study was to test a graph machine learning engine, Headai Graphmind, to recognize ICF codes from electronic health records written in Finnish. A dataset of 93 patients aged 18 to 65 years with chronic low back pain was collected. Headai Graphmind was then tested for its ability to match free text with ICF codes on a sample of 20 patients. The results were compared against the findings of a domain expert. Headai Graphmind achieved 0.95 precision, 0.83 recall, and 0.89 F1 score.The application found 112 distinct ICF codes compared to 119 codes found by the domain expert. Headai Graphmind has the capability to recognize ICF codes from the electronic health records of patients with chronic low back pain. The method could be helpful when implementing the ICF classification in clinical practice, and enable retrospective coding of medical information for further use.