Protocol and registration
This systematic review was pre-registered in the Prospero Database (CRD42020158182) and followed a guideline for systematic review on musculoskeletal disorders.[10]
Information sources and search
One author (HK) systematically searched the following databases from inception to Dec 2020: Web of science, CENTRAL, MEDLINE, EMBASE, PsycINFO, EMCARE, CINAHL, AMED and PEDro databases. Although CENTRAL and AMED were not listed in the PROSPERO, a more accurate search was performed. The search strategies are presented in Appendix 2 (See Appendix 2). Moreover, cross-referencing was done to the primary developer of the CFT, Peter O’Sullivan, and relevant literatures cited in the studies were searched manually. Language limitation was removed although initially this was limited to English only during the first registration with the PROSPERO.
Eligibility criteria
The PICOS framework was used in developing the eligibility criteria. Enrolled participants included 1) patients with chronic LBP; 2) patients from primary, secondary, or tertiary care; and 3) patients without radiating pain to lower extremities. Studies with the following participants were not included in this systematic review: 1) participants with LBP caused by serious pathologies, including infections, neoplasms, metastases, fractures, osteoporosis, rheumatoid arthritis, and radiculopathies; 2) participants with LBP during or immediately following pregnancy; and 3) participants with postoperative back pain. Eligible interventions include the CFT and the classification-based CFT as both managements were conceptually comparable, which was confirmed by the primary developer, Peter O’Sullivan. Eligible comparisons included any types of interventions other than the CFT or the classification-based CFT. Eligible outcomes included pain intensity, back-specific disability/function status, quality of life, and psychological status. Eligible study design was only the published RCT although it included observational studies in PROSPERO.
Study selection and data collection process
Screening and full-text inspection were performed by two authors (TM and YK) independently, where publication source, authors, and publication year were not blinded. The screening was based on the information in the title and abstract. Any disagreements on eligibility were resolved through a consensus between the two authors.
Risk of bias in individual studies
The quality of the argument was assessed using a 10-point PEDro score, which was changed from the Cochrance (RoB) registered in the initial draft of the PROSPERO. This is because the PEDro score was specifically developed to assess the methodological quality of physical therapy trials [11, 12], and established scores are presented. Inclusion criteria for the meta-analysis was high-quality study, which was defined as the study with a PEDro score of 6 or higher [13, 14].
The quality of the evidence
The overall quality of evidence for each meta-analysis was identified using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) system [15] as recommended by the Updated Method Guideline for Systematic Reviews in the Cochrane Back and Neck Group [16]. The GRADE system consists of five items: 1) risk of bias, 2) inconsistency, 3) indirectness, 4) imprecision, and 5) publication bias. Each criterion is rated as high, moderate, low, or very low, and the lowest quality is chosen as the overall quality of evidence. For evidence from RCTs, we started with a rating of `high`. The quality of evidence on specific outcomes was reduced by one or two levels depending on the performance of specific comparative studies on these five factors. Regarding the risk of bias, the quality of the evidence has been reduced by one point if more than 25% and two points if more than 50% of the participants are from studies conducted in low-quality methods (e.g., PEDro score < 6). For inconsistency, the quality of evidence has been reduced by one point when the heterogeneity or variability in results was large, as indicated by an I2 > 50% and reduced by two points when an I2 > 75%. Moreover, another point is reduced if differences among populations, interventions, or outcomes are present. For indirectness, whether the issues addressed in this systematic review differed from the available evidence on populations, interventions, comparisons and outcomes has been assessed. Another point is deducted if there is indirectness in only one area and two points if there is indirectness in two or more areas. As regards imprecision, one point is deducted if the total number of participants was less than 400. In addition, if there is no significant difference in outcomes, the grade is reduced. Regarding publication bias, a funnel plot was created to compare at least ten studies, and from the funnel plot, the quality of the evidence will be reduced by one point if publication bias was suggested.
The two authors (TM and YK) independently assessed GRADE scores in each meta-analysis, and any disagreements were resolved by another author (HK).
Data items and summary measures
Data were extracted based on the PICOS framework by the two authors (TM and YK) with consensus. Extracted data of the participants included participant source and setting, age, gender, and duration of symptoms. Extracted data of the intervention included description of interventions, duration and number of sessions, therapist training level on the CFT, and delivery type such as individual or group. Extracted data of comparisons included type of intervention and duration and/or number of sessions. Extracted data of outcomes included means and SDs of pain, disability/function status, quality of life, and psychological factors during short, intermediate- and/or long-term follow-ups. In this study, the short term was defined as post-treatment. The intermediate term was defined as ≥ 3 months and < 12 months, and the long term was defined as ≥ 12 months [17]. In the presence of more eligible follow-up points, the follow-up point closest to 6 months was chosen for the intermediate term and the follow-up point closest to 12 months was chosen for the long term. The meta-analysis was attempted using change values from the baseline to each follow-up point first [18]. If there is no change values reported in the study, a corresponding author was contacted by email and was asked to provide the data. If the value of the change is not available, the values at each follow-up point was used for meta-analysis [18]. A reminder email was sent a week after the first contact, and the second reminder was sent two weeks after the first reminder, and if after two weeks, there is still no reply, then no response was considered. Data on serious adverse reports were also extracted, including serious accidents and deaths during the intervention. Extracted data of the study design included country of data collection and source of research grant.
Synthesis of results and statistical analysis
When multiple datasets of similar outcomes are present, meta-analysis was performed using Review Manager 5 (The Nordic Cochrane Centre, København Ø, Denmark). The mean difference (MD) or the standardised mean difference (SMD) with 95% confidence intervals (95% CI) was calculated using the random-effect model. The MD was calculated when similar psychometric properties were assessed with similar methods, and the SMD was used when similar psychometric properties were assessed with different methods. The I² statistic was assessed for heterogeneity among trials, whose interpretations were as follows: 0 − 40% = may be insignificant, 30 − 60% = moderate heterogeneity, 50 − 90% = substantial heterogeneity, and 75 − 100% = considerablel heterogeneity [18]. The sensitivity analysis was undertaken when considerable heterogeneity existed and thus the sensitivity analysis was possible.