Vonk 2009.

Study characteristics
Methods	RCT. ’computer‐generated randomization scheme’
Participants	139 patients, mean age ± SD in (I): 45.7 ± 12.1 years; (R): 45.7 ± 12.7 years; with diagnosis of non‐specific NP. The exclusion criteria were: diagnosis of a specific disorder, physical/manual therapy treatment during the previous six months, chronic disease or undergoing surgery in the near future. 12 and 18 patients in the (R) and (I) groups, respectively, dropped out at 9 weeks. The number of drop‐outs increased to 24 and 23 in the (R) and (I) groups, respectively, at 52 weeks. The analyses were carried out according to the intention‐to‐treat principle.
Interventions	CBT treatment (I): behaviour graded activity program. The duration of the treatment was about 30 minutes and patients could received up to 18 treatments. The treatment was according to a biopsychosocial model. During the treatment, patients discussed their beliefs about pain following the pain model and defined personal aims and baseline levels of activities in order to systematically increase them throughout graded exercises. Patients learned to manage their pain and relapses period (n = 68). Reference treatment (R): conventional exercise composed of exercises and physiotherapy techniques (i.e. massage, mobilization, traction). The duration of the treatment was about 30 minutes and patients could received up to 18 treatments (n = 71).
Outcomes	The primary outcome was GPE: GPE for recovery of complaints and GPE for recovery of functioning in daily activities. At 4 weeks, the (R) group showed significantly more recovery of complaints compared to the (I) group (odds ratio 0.25, 95% CI 0.06 to 0.99). At 9 weeks, recovery of complaints was similar for both groups. Thereafter, recovery stabilized in the (R) group, whereas in the (I) group it increased until follow‐up at 26 weeks. The pattern of recovery in daily functioning was similar in both groups. No significant differences between treatments were found. Secondary outcomes were physical and psychological measurements. For the physical outcomes, no significant differences were found between the two groups at any time point of measurement. However, for the severity of the main complaint, pain severity, and impediment, both treatments showed a clinically significant improvement (> 2 points), which was maintained until 52 weeks follow‐up and was even enhanced for impediment. For the psychosocial outcomes the (I) group showed significantly higher improvements compared to (R) only for catastrophising and pain self‐efficacy at the end of the treatment period (9 weeks), and for pain self‐efficacy at 26 weeks of follow‐up. All other secondary measures were not significantly different.
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Quote: “An independent examiner using a computer‐generated randomisation schema performs randomisation.”
Allocation concealment (selection bias)	Low risk	Quote: “An independent examiner using a computer‐generated randomisation schema performs randomisation.” Comment: Randomisation was performed by an independent examiner.
Blinding of participants	High risk	Althought authors wrote that the "patients were blinded for treatment allocation", due to the nature of the intervention patients could not be considered as blinded.
Blinding of personnel/ care providers (performance bias)	High risk	Quote in Vonk 2004: “The physiotherapists are not blinded for allocation, but the physiotherapists from each treatment group are kept strictly separate and are not involved in the outcome measurement”; Quote in Vonk 2009: “Physiotherapists were not blinded but were not involved in the outcome measurement.”
Blinding of outcome assessment (detection bias) self‐reported measures	High risk	Self‐reported outcome measures were collected and participants were not blinded.
Incomplete outcome data (attrition bias) All outcomes	High risk	21% dropouts at 9 weeks; 31% dropouts at 52 weeks.
Selective reporting (reporting bias)	Low risk	All of the outcome measures described in the Methods section are reported in the Results section.
Group similarity at baseline (selection bias)	Low risk	Quote: “Baseline demographics and patient characteristics were well balanced between the two groups.”
Cointerventions (performance bias)	Low risk	Quote: “We also examined medicine use, number of side effects (e.g. headache, dizziness, etc.) and additional treatments used, all as reported by the patients...No differences between treatments were found.” Comment: Co‐interventions were similar between groups.
Compliance (performance bias)	Unclear risk	Quote: “The therapist decided the number of treatments but the patient also had the option to stop treatment. The mean number of treatments received was 6.6 (3.0) in BGA and 11.2 (4.1) in CE.” Comment: Not sufficient information to judge.
Intention‐to‐treat‐analysis	Low risk	Quote: “All analyses were carried out according to the intention‐to‐treat principle.”
Timing of outcome assessments (detection bias)	Low risk	Quote: “Outcome of intervention will be assessed at 4 and 9 weeks after randomisation; however, if the treatment is not finished at 9 weeks, the patients will receive an additional questionnaire...after finishing the treatment. Follow‐up assessments are planned at 26 and 52 weeks after randomization.”