Comparison of stratification techniques for optimal management of patients with chronic low back pain in spine clinics

Abstract

Identifying optimal stratification techniques for subgrouping patients with low back pain (LBP) into treatment groups for the purpose of identifying optimal management and improving clinical outcomes is an important area for further research. Our study aimed to compare performance of the STarT Back Tool (SBT) and three stratification techniques involving PROMIS domain scores for use in patients presenting to a spine clinic for chronic LBP. Retrospective cohort study. Adult patients with chronic LBP seen in a spine center between 11/14/18-5/14/19 who completed patient-reported outcomes (PROs) as part of routine care, and were followed up with completed PROs 1-year later. Four stratification techniques, including SBT, and 3 PROMIS-based techniques: the NIH Task Force recommended Impact Stratification Score (ISS), symptom clusters based on latent class analysis (LCA), and SPADE symptom clusters. The four stratification techniques were compared according to criterion validity, construct validity, and prognostic utility. For criterion validity, overlap in characterization of mild, moderate, and severe subgroups were compared to SBT, which was considered the gold standard, using quadratic weighted kappa statistic. Construct validity compared techniques' ability to differentiate across disability groups defined by modified Oswestry LBP Disability Questionnaire (MDQ), median days in the past month unable to complete activities of daily living (ADLs), and worker's compensation using standardized mean differences (SMD). Prognostic utility was compared based on the techniques' ability to predict long-term improvement in outcomes, defined as improvement in global health and MDQ at 1-year. There were 2246 adult patients with chronic LBP included in our study (mean age 61.0 (SD 14.0), 55.0% female, 83.4% white). All stratification techniques resulted in roughly a third of patients grouped into mild, moderate, and severe categories, with ISS and LCA demonstrating substantial agreement with SBT, while SPADE had moderate agreement. Construct validity was met for all techniques, with large effects demonstrated between mild and severe categories for differentiating MDQ, ADLs, and worker's compensation disability groups (SMD range 0.57-2.48). All stratification techniques demonstrated ability to detect improvement by 1-year, with severe groups experiencing the greatest improvement in multivariable logistic regression models. All four stratification techniques demonstrated validity and prognostic utility for subgrouping patients with chronic LBP based on risk of long-term disability. ISS and LCA symptom clusters may be the optimal methods given the improved feasibility of including only a few relevant PROMIS domains. Future research should investigate multidisciplinary treatment approaches to target mild, moderate, and severe patients based on these techniques.