The review question and inclusion criteria were clearly stated, but potentially subjective. Two relevant databases were searched, but limiting this to articles in English means that some relevant trials could have been missed. Trials indexed in specialist databases were not sought; the search was not comprehensive. The risk of publication bias was not assessed and there was no search for unpublished trials.
Measures were taken to minimise error and bias in study selection, but it was unclear if this was the case for data extraction. It appears that study quality was assessed when assigning a level of evidence, but no details were reported and the risk of bias in the included trials was unclear. Statistical heterogeneity in the meta-analysis was not reported, but the patients in the two trials had markedly different levels of pain and disability, so pooling the data in a meta-analysis was probably not appropriate.
The authors' conclusion that exercise and manipulation appeared to be equivalent does not reflect the limited and heterogeneous evidence presented and is unlikely to be reliable.