Does a modified STarT Back Tool predict outcome with a broader group of musculoskeletal patients than back pain? A secondary analysis of cohort data

Objectives: The STarT Back Tool has good predictive performance for non-specific low back pain in primary care. We therefore aimed to investigate whether a modified STarT Back Tool predicted outcome with a broader group of musculoskeletal patients, and assessed the consequences of using existing risk-group cut-points across different pain regions.

Setting: Secondary analysis of prospective data from 2 cohorts: (1) outpatient musculoskeletal physiotherapy services (PhysioDirect trial n=1887) and (2) musculoskeletal primary-secondary care interface services (SAMBA study n=1082).

Participants: Patients with back, neck, upper limb, lower limb or multisite pain with a completed modified STarT Back Tool (baseline) and 6-month physical health outcome (Short Form 36 (SF-36)).

Outcomes: Area under the receiving operator curve (AUCs) tested discriminative abilities of the tool's baseline score for identifying poor 6-month outcome (SF-36 lower tertile Physical Component Score). Risk-group cut-points were tested using sensitivity and specificity for identifying poor outcome using (1) Youden's J statistic and (2) a clinically determined rule that specificity should not fall below 0.7 (false-positive rate <30%).

Results: In PhysioDirect and SAMBA, poor 6-month physical health was 18.5% and 28.2%, respectively. Modified STarT Back Tool score AUCs for predicting outcome in back pain were 0.72 and 0.79, neck 0.82 and 0.88, upper limb 0.79 and 0.86, lower limb 0.77 and 0.83, and multisite pain 0.83 and 0.82 in PhysioDirect and SAMBA, respectively. Differences between pain region AUCs were non-significant. Optimal cut-points to discriminate low-risk and medium-risk/high-risk groups depended on pain region and clinical services.

Conclusions: A modified STarT Back Tool similarly predicts 6-month physical health outcome across 5 musculoskeletal pain regions. However, the use of consistent risk-group cut-points was not possible and resulted in poor sensitivity (too many with long-term disability being missed) or specificity (too many with good outcome inaccurately classified as 'at risk') for some pain regions. The draft tool is now being refined and validated within a new programme of research for a broader musculoskeletal population.

Trial registration number: ISRCTN55666618; Post results.