Appendix 1. Methods archived for future updates
Assessment of risk of bias in included studies
The Cochrane Risk of Bias (RoB) tool will be used to determine risk of bias of selected trials.
The Cochrane Risk of Bias tool assesses the risk that a study over or under estimates the true effect of the intervention. This process involves a description and judgment on the adequacy of: sequence generation; allocation sequence concealment; blinding of outcome assessment; incomplete outcome data; selective outcome reporting and other potential sources of bias. Each criteria would have been judged using 'Yes', 'No', or 'Unclear'. 'Yes' indicates a low risk of bias, 'No' a high risk of bias and 'Unclear' that there was insufficient information or uncertainty over the risk of bias. This qualitative quality assessment of bias would not be used as a threshold for inclusion of studies, but as a possible explanation for differences in results between studies (Schulz 1995).
Measures of treatment effect
For dichotomous outcomes, such as the presence or absence of a diagnosis of PTSD, we will calculate the risk difference with 95% confidence intervals.
We will calculate the standardized mean differences for continuous outcomes if:
a) means and standard deviations are available either in the original article or from the authors;
b) there is no clear evidence of a skewed distribution (i.e where there is minimum or maximum score, the mean minus this score and divided by the standard deviation should not be less than 2).
Differences in the direction of the scale between studies will be corrected by multiplying the mean of one set of trials by -1. Where measurements are comparable and on the same scale (such as using the same tool to measure depression) these will be combined to obtain weighted mean differences. Where different scales are used to measure the same clinical outcome in different ways, standardized mean differences will be used in order to combine results across scales.
Unit of analysis issues
Trials will be analysed at the level of participants. Where cross-over trials are identified (and cross-over design is thought to be appropriate), consideration will be given to whether serious carry-over may have occurred. If carry-over is not thought to be a problem, advice will be sought from a statistician about whether appropriate methods of analysis have been used. If so, the effect estimate will be included in a meta-analysis using the generic inverse variance method as described in the Cochrane Handbook.
Where cluster-randomized trials are identified (and cluster-randomisation is thought to be suitable), advice will be sought from a statistician about whether the appropriate methods of analysis have been used. Effect estimates and their standard errors from correct analyses of cluster-randomized trials will be meta-analysed using the generic inverse variance method in RevMan version 5.
Where appropriate, sensitivity analyses will be undertaken to investigate the effects of incorporating data from cross-over and cluster randomised trials in this review.
Dealing with missing data
Missing data and attrition rates will be assessed for each of the included studies, and the number of participants who are included in the final analysis will be reported as a proportion of all participants in the study. Trialists will be contacted to obtain missing data. Reasons given for missing data will be provided in the narrative summary and the extent to which the results are altered by missing data will be ascertained. Assessment will be made of the extent to which studies have conformed to an intention-to-treat analysis. The extent to which the results are altered by missing data will be determined by a sensitivity analysis for dichotomous data, as suggested by Deeks in the Cochrane Handbook, where it is firstly assumed that "all missing participants in the first group experienced the event and those in the second group did not and then assume the opposite".
Assessment of heterogeneity
Heterogeneity of results will be tested by comparing the confidence intervals of the studies (presented graphically) and by performing a chi-square test. To quantify the inconsistency in the results statistically, we will use I² (Higgins 2003). Values greater than 50% indicates substantial heterogeneity and the reasons for such will be explored. Possible explanations could be clinical or methodological diversity. Possible causes of statistical heterogeneity are expected and pre-specified as follows:
a) clinical heterogeneity due to variation in the participants, interventions and outcomes used by the studies;
b) methodological heterogeneity due to variability in trial design and quality.
If there is substantial statistical heterogeneity, a random effects meta-analysis will only be performed where studies report similar interventions, and where data are available and sufficiently clinically and methodologically homogeneous. If statistical heterogeneity is not present, a fixed-effect meta-analysis will be performed, and potential differences between subgroups will be explored according to a priori criteria set out below.
Assessment of reporting biases
To determine if this review was likely to have been affected by reporting biases and, in particular, publication bias, a funnel plot will be prepared and checked for asymmetry.
Prior to inspection for statistical heterogeneity (as described above), a fixed effect model will be used to synthesise the data. Where statistically significant heterogeneity is identified, a random effects model will be used, but only where trials appear to be clinically and methodologically homogeneous. Where data are available, sports, games and play-based interventions will be compared with usual care, pharmacological and psychosocial interventions. Different types of sports, games and play-based interventions will also be compared with one another.
Subgroup analysis and investigation of heterogeneity
Sub group analyses will be performed where data are available on:
a. Adults versus children/adolescents (up to and including 18 years of age)
c. Self-administered versus clinician-administered scales
The first two factors have been identified as being important as they affect an individual's inclination or opportunity to engage with, and benefit from, sports, games and play interventions. The inclusion of the third factor is based on potential differences in the reliability and validity of self-administered questionnaires.
In order to assess the robustness of the review conclusions to decisions taken during the review process, sensitivity analyses will be performed according to whether allocation concealment was adequate vs. inadequate.