Exercise interventions for persistent non-specific low back pain: does matching outcomes to treatment targets make a difference? A systematic review and meta-analysis

Key points

1. 01A treatment target is what an exercise program is actually trying to change (for example trunk stability or flexibility), and a matched outcome is one that measures exactly that.
2. 02Only 7 of 27 trials (26%) matched their primary outcome to their treatment targets, and 19 of 31 identified targets appeared in just a single trial, showing little consensus on what exercise is meant to do.
3. 03Matched trials had a moderate effect (SMD 0.54, 95% CI 0.23 to 0.85) versus a small to moderate effect in unmatched trials (SMD 0.22, 95% CI 0.01 to 0.44).
4. 04The head-to-head difference between matched and unmatched groups was not significant (P = 0.10) and heterogeneity was very high (I2 92% to 95%), so the matching signal is suggestive only.
5. 05The authors conclude exercise effectiveness may currently be underestimated if outcomes are poorly aligned to targets, but larger datasets are needed to confirm this.

How it was conducted

Design

Systematic review and meta-analysis of RCTs (PROSPERO CRD42017072023), exploratory matching analysis

Search

Eight databases to August 2019; 19,272 records screened, 27 RCTs included (5,870 participants), each powered for at least a moderate effect (>60 per arm)

Intervention

Supervised or tailored exercise for persistent NSLBP (symptoms >12 weeks) versus non-exercise comparators (usual care, placebo, brief intervention, or waitlist)

Outcomes

Standardised mean difference of exercise versus control at the primary time point, with trials classed as matched or unmatched to stated treatment targets

Analysis

Random-effects meta-analysis (RevMan 5.3) with subgroup, ratio-of-means sensitivity analyses, and Cochrane risk of bias

What they found

• Matched trials (7 trials): pooled SMD 0.54 (95% CI 0.23 to 0.85), P = 0.0006, favoring exercise versus non-exercise controls.
• Unmatched trials (20 trials): pooled SMD 0.22 (95% CI 0.01 to 0.44), P = 0.04.
• Difference between matched and unmatched groups not significant (P = 0.10; prediction interval -0.12 to 0.76); overall pooled SMD across all 27 trials was 0.31 (95% CI 0.14 to 0.47).
• Statistically significant benefit was reported in 5 of 7 matched trials (71%) versus only 4 of 20 unmatched trials (20%).
• Pain sensitivity analysis: matched mean improvement 1.79 points (95% CI 0.11 to 3.46) versus unmatched 0.65 (95% CI 0.08 to 1.08); subgroup difference not significant (P = 0.20), heterogeneity I2 98%.

Limitations

• The between-group difference was not statistically significant and the review was underpowered to detect one, so findings are exploratory.
• Heterogeneity was very high (I2 92% to 95%) given the wide range of exercise types and populations pooled.
• Almost all included trials were judged high risk of bias, mainly from lack of blinding, and only 6 of 27 met at least 5 risk of bias domains.
• Classifying trials as matched or unmatched was a subjective, non-validated process, and trials that did match but did not state it would have been miscounted as unmatched.

Why it matters

For patients

Exercise for ongoing low back pain may help more than headline trial numbers suggest, especially when the program is measured against what it is actually trying to improve.

For clinicians

Be clear about what an exercise program is targeting and track an outcome that reflects that target, rather than defaulting to a single generic pain or function score.

For readers

How well a trial's outcome matches the intervention's goal may shape whether exercise looks effective, a methodological issue worth watching in future RCT design.