Identifying non-specific low back pain clinical subgroups from sitting and standing repositioning posture tasks using a novel Cardiff Dempster–Shafer Theory Classifier. (December 2019)

• Record Type: Journal Article
• Authors: Sheeran, Liba; Sparkes, Valerie; Whatling, Gemma; Biggs, Paul; Holt, Cathy
• Is Part Of: Clinical biomechanics. Volume 70(2019)
• Page Start: 237
• Page End: 244
• Publication Date: 2019-12
• Subjects: Non-specific low back pain -- Dempster-Shaffer Theory Classifier -- Objective classification; Biomechanics -- Periodicals; Osteopathic medicine -- Periodicals; Biomechanics -- Periodicals; Osteopathic Medicine -- Periodicals; 612.76
• DOI: 10.1016/j.clinbiomech.2019.10.004
• Languages: English
• ISSNs: 0268-0033

Abstract

Abstract: Background: Low back pain (LBP) classification systems are used to deliver targeted treatments matched to an individual profile, however, distinguishing between different subsets of LBP remains a clinical challenge. Methods: A novel application of the Cardiff Dempster–Shafer Theory Classifier was employed to identify clinical subgroups of LBP on the basis of repositioning accuracy for subjects performing a sitting and standing posture task. 87 LBP subjects, clinically subclassified into flexion (n = 50), passive extension (n = 14), and active extension (n = 23) motor control impairment subgroups and 31 subjects with no LBP were recruited. Thoracic, lumbar and pelvic repositioning errors were quantified. The Classifier then transformed the error variables from each subject into a set of three belief values: (i) consistent with no LBP, (ii) consistent with LBP, (iii) indicating either LBP or no LBP. Findings: In discriminating LBP from no LBP the Classifier accuracy was 96.61%. From no-LBP, subsets of flexion LBP, active extension and passive extension achieved 93.83, 98.15% and 97.62% accuracy, respectively. Classification accuracies of 96.8%, 87.7% and 70.27% were found when discriminating flexion from passive extension, flexion from active extension and active from passive extension subsets, respectively. Sitting lumbar error magnitude best discriminated LBP from no LBP (92.4% accuracy) and the flexion subset from no-LBP (90.1% accuracy). Standing lumbar error best discriminated active and passive extension from no LBP (94.4% and 95.2% accuracy, respectively). Interpretation: Using repositioning accuracy, the Cardiff Dempster–Shafer Theory Classifier distinguishes between subsets of LBP and could assist decision making for targeted exercise in LBP management. Highlights: Objective classifier discriminated back pain subsets based on repositioning sense. An accuracy comparable to that of expert clinician agreement was achieved. Sitting repositioning error best discriminated flexion related back pain. Standing repositioning error best discriminated extension related back pain.