Statistical analyses
The primary outcome was recovery into the normal range of CP and improvement of CP and DHI score at post-intervention. The standardized mean effect was calculated using Hedges’s g value to represent the effect sizes according to the Cochrane Handbook for Systematic Reviews of Intervention.8 Hedges’ g value is a standardized mean difference that was calculated to standardize the values measured by various measurement tools into a single unit. It is also a method to compensate for the shortcomings of the existing Cohen’s d value. In this study, the sample size was not large. Therefore, the Hedges’ g value was calculated. The Hedges’ g values were 0.2 as small, 0.5 as medium, and 0.8 as large. When calculating the total effect size, both the DHI score (with the negative effect size direction) and the other variables (with static effect direction) were included. This combination resulted in a mutually offset effect. Therefore, DHI score was converted to the reverse direction and analyzed.
The estimates were pooled using a fixed-effects model, which assumes that the effect sizes of the populations are the same, and that the differences in effect size are attributable to sampling error.11 The homogeneity of the studies was identified through Higgin’s I statistics and forest plots. The heterogeneity of the studies was tested using the forest plot and Higgins I2 statistic. Significant heterogeneity was defined as an I2 value of 25%, 50% and 75% as low, moderate, and high heterogeneity, respectively.12 When needed, subgroup and sensitivity analyses were conducted. Funnel plots and Egger’s linear regression values for each outcome were prepared and evaluated to assess potential publication bias. All analyses were performed using Cochrane Collaboration’s software (RevMan version 5.3.3; Nordic Cochrane Centre, Copenhagen, Denmark) and Comprehensive Meta-Analysis software version 3 (Borenstein M, Hedges L, Higgins J and Rothstein H; Biostat, Englewood, NJ, USA).13
Some parts violate the independence assumption when calculating the total effect size. However, since the number of documents included in the analysis was limited, there was minimal information loss by using the effect size as an analysis unit.