Category of presentation:
Epidemiology: evidence based papers on effective diagnostic and therapy outcome
DISCRIMINATING NON-SPECIFIC CHRONIC LOW BACK PAIN CLINICAL SUBGROUPS AND MONITORING RECOVERY USING AN OBJECTIVE CLASSIFICATION METHOD
Sheeran L.1, Whatling G.2, Holt C.2, Beynon M. J.3, van Deursen R.1, Sparkes V.1
1Cardiff University, School of Healthcare Studies, Cardiff, UK; 2Cardiff University, School of Engineering, Cardiff, UK; 3Cardiff University, Cardiff Business School, Cardiff, UK.
Heterogeneity of non-specific chronic low back pain (NSCLBP) can be deleterious to management success. Classification systems (CSs) that sub-classify NSCLBP to guide interventions often rely on clinical expertise and user familiarity. An objective classification method that alongside with clinical CSs aids classification and monitoring recovery with greater accuracy and less subjectivity is preferable.
To determine accuracy of an objective classification method based on Dempster-Shafer theory, the Cardiff Classifier (CC), discriminating healthy controls and clinical NSCLBP subgroups using sensory, spinal-pelvic repositioning error (RE) motor control (MC) parameters. To establish the most sensitive parameters discriminating NSCLBP subgroups from healthy and predicting recovery using CC.
Materials and Methods
Baseline and post-motor learning intervention (MLI) spinal-pelvic REs from 31 healthy (H) and 87 NSCLBP individuals with flexion (FP,n=50), passive extension (PEP,n=14) and active extension pattern (AEP,n=23) MC impairment subclassified by O’Sullivan’s CS were used. CC provided objective and visual indicators of NSCLBP subgroups, H and MLI effect. RE data were transformed into a set of three belief values: non-pathological function (NP), LBP characteristics and level of uncertainty. Each subject’s status was visually represented as a single point in a simplex plot (Figure 1). Subjects left of the central line have NP function; to the right have LBP characteristics.
CC accuracy to discriminate FP and H was 85.2%(Figure1a), AEP and H 96.3%(Figure1b), PEP and H 100%(Figure1c). The most distinguishing parameter for FP was sitting lumbar RE, for AEP & PEP was standing lumbar RE. Combining all NSCLBP reduced CC accuracy to discriminate from H to 68.6% (Figure1d). CC distinguished pre/post-intervention REs with 85.7% accuracy for FP and 90% AEP, with sitting lumbar RE and standing pelvic RE as the most sensitive predictors of recovery, respectively.
Figure 1. Simplex plots: a)FP+and H°, b)AEP+and H°, c)PEP+and H°, d)all NSCLBP+and H°
Using CC in combination with clinical CS may enhance its accuracy by accounting for subjectivity related to varied clinical expertise levels and user familiarity. Using CC to identify key parameters characterizing each subgroup may focus recovery monitoring.
Using RE, CC accurately discerned three clinical NSCLBP subgroups, concurring with clinical CS. CC identified parameters most accurately characterizing each subgroup and predicting post-intervention recovery.
This is the first CC application in NSCLBP. Importance of classification was illustrated with CC enhanced accuracy to discern between NSCLBP and H when patients were subclassified. CC should be further tested using clinical classification data.
NSCLBP classification is complex requiring synthesis of physical and clinical measures. CC can order complex data sets to aid clinicians’ interpretation and identify most sensitive parameters to efficiently monitor recovery, thereby enhancing classification robustness and usability.
NSCLBP, objective classification, repositioning error, Dempster-Shafer