Most clinical trials in the field of low back pain use mean differences and 95% confidence intervals to estimate treatment effects. However, clinicians often criticize this statistical approach, arguing that it may not provide a "real estimate" at an individual level. An alternative strategy is to categorize patients as "recovered" or "non-recovered" to better identify those who may benefit more from specific treatments. This type of analysis is known as responder analysis.

This study aims to perform a secondary analysis (responder analysis) of clinical trials involving patients with low back pain, examining each study’s results individually and investigating whether a responder analysis would alter the conclusions of these trials.

We conducted a secondary analysis of fourteen randomized controlled trials performed by our research group on patients with chronic low back pain. Pain, disability, and global perceived effect outcomes were analyzed. We generated cutoff points to classify patients as “recovered” or “not recovered” for the responder analysis. We then compared the results of the responder analysis to the original findings of these trials.

We analyzed data from 2,187 patients with chronic low back pain. In total, only 23 out of 384 (6%) analyses altered the primary findings, suggesting that although the secondary analysis modified some individual outcomes, most results remained consistent with the primary analysis.

Reanalyzing the original clinical trial results using responder analysis did not lead to significant changes in the previous conclusions of any of these trials. The hypothesis that a response-based analysis could identify a higher proportion of recovered patients in the intervention groups was not confirmed.

Despite being widely used, responder analysis appears to yield similar conclusions to traditional statistical approaches in low back pain trials.