| Measures of treatment effect |
We will analyse dichotomous outcomes by calculating the risk ratio (RR) and the corresponding 95% confidence interval (CI). For continuous data, we will calculate mean difference (MD) and corresponding 95% CI if studies used the same rating scales. As recommended by Higgins 2011, we will focus on final values unless change scores are used in some of the studies. We will combine in the same meta‐analysis studies that reported final values with studies that reported only change scores, provided the studies used the same rating scale (Higgins 2011). A potential problem associated with including change scores is that the standard deviation of changes may not be reported in the original study (Higgins 2011). We will contact trial authors and will attempt to estimate the standard deviation of changes if not reported. We will calculate the standardised mean difference (SMD) with 95% CIs if studies used different scales to measure the same outcomes |
| Multiple outcomes |
If studies provided multiple, interchangeable measures of the same construct at the same point in time, we will calculate the average SMD across outcomes and the average estimated variances (Higgins 2011) |
| Unit of analysis issues |
When possible, we will obtain mean treatment differences and standard errors for cross‐over trials, and will enter these into RevMan under the generic inverse variance outcome type (Higgins 2011). We will create a single pair‐wise comparison for each identified multi‐arm study by combining all relevant experimental groups into a single group, and by combining all relevant control groups into a single group (Higgins 2011) |
| Dealing with missing data |
We will attempt to contact trial investigators to request missing data. If missing data are provided by the trialists, we will conduct meta‐analysis according to intention‐to‐treat principles using all data and keeping participants in the treatment group to which they were originally randomly assigned, regardless of the treatment that they actually received (Higgins 2011). If missing data are not provided, we will analyse only available data, and we will not impute missing data given that symptoms of autism spectrum disorder (ASD) vary greatly. We will document missing data and attrition in the ’Risk of bias’ table, and we will discuss how missing data may affect interpretation of the results |
| Assessment of heterogeneity |
We will assess clinical heterogeneity by comparing the between‐trials distribution of participant characteristics (e.g. children vs adults), trial characteristics (e.g. cross‐over vs parallel design) and intervention characteristics (e.g. treatment type, dose). We will evaluate statistical heterogeneity using the I² statistic and the Chi² test of heterogeneity, with statistical significance set at P value < 0.10. We will consider I² values as follows.
0% to 29% might not be important 30% to 49% may represent moderate heterogeneity 50% to 74% may represent substantial heterogeneity 75% to 100% represents considerable heterogeneity (Higgins 2011)
|
| Assessment of reporting biases |
If 10 or more studies are found, we will use funnel plots to investigate the relationship between intervention effect and study size. Asymmetry of a funnel plot may indicate, among other things, publication bias or poor methodological quality (Egger 1997). We will explore possible reasons for any asymmetry found |
| Data synthesis |
We will synthesise results in a meta‐analysis using a fixed‐effect model when studies are similar enough with regard to the intervention, population and methods to assume that the same treatment effect is estimated. We will synthesise results in a meta‐analysis using a random‐effects model when statistical heterogeneity is found, or when studies differ enough with regard to the intervention, population and methods to assume that different yet related treatment effects are estimated, and when it is deemed to be clinically relevant (Higgins 2011) |
| Subgroup analysis and investigation of heterogeneity |
We will conduct the following subgroup analyses
Type of ASD (e.g. autistic disorder vs Asperger’s disorder) Participant age (e.g. adult vs child, preschool vs school‐age) Treatment type (e.g. dimercaptosuccinic acid (DMSA) vs other agents) Treatment dosage (e.g. DMSA dose of 10 mg/kg body weight administered 3 times per day vs higher doses) Length of follow‐up (e.g. ≤ 3 months vs > 3 months)
|
| Sensitivity analysis |
We will conduct sensitivity analyses to investigate the effect on overall results of excluding trials when
Allocation concealment or sequence generation was inadequate (selection bias) Blinding was not done (performance bias) Outcome data were incomplete (attrition bias)
|
