Initial classification of low back and leg pain based on objective functional testing: a pilot study of machine learning applied to diagnostics

Abstract

The five-repetition sit-to-stand (5R-STS) test was designed to capture objective functional impairment and thus provided an adjunctive dimension in patient assessment. The clinical interpretability and confounders of the 5R-STS remain poorly understood. In clinical use, it became apparent that 5R-STS performance may differ between patients with lumbar disk herniation (LDH), lumbar spinal stenosis (LSS) with or without low-grade spondylolisthesis, and chronic low back pain (CLBP). We seek to evaluate the extent of diagnostic information contained within 5R-STS testing. Patients were classified into gold standard diagnostic categories based on history, physical examination, and imaging. Crude and adjusted comparisons of 5R-STS performance were carried out among the three diagnostic categories. Subsequently, a machine learning algorithm was trained to classify patients into the three categories using only 5R-STS test time and patient age, gender, height, and weight. From two prospective studies, 262 patients were included. Significant differences in crude and adjusted test times were observed among the three diagnostic categories. At internal validation, classification accuracy was 96.2% (95% CI 87.099.5%). Classification sensitivity was 95.7%, 100%, and 100% for LDH, LSS, and CLBP, respectively. Similarly, classification specificity was 100%, 95.7%, and 100% for the three diagnostic categories. 5R-STS performance differs according to the etiology of back and leg pain, even after adjustment for demographic covariates. In combination with machine learning algorithms, OFI can be used to infer the etiology of spinal back and leg pain with accuracy comparable to other diagnostic tests used in clinical examination. These slides can be retrieved under Electronic Supplementary Material.