Data extraction and management  

For included trials, data will be abstracted as recommended in Chapter 7 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). We will extract data independently and will include:

• author, year of publication and journal citation (including language);

• country;

• setting;

• inclusion and exclusion criteria;

• study design, methodology;

• study population (participant characteristics, age, stage and postoperative residual disease);

• number of participants in each arm of the trial;

• total number of intervention groups;

• ovarian carcinosarcoma details (FIGO stage, histology, tumour grade);

• type of intervention (chemotherapy agents and radiotherapy, dosage and timing of administration relative to surgery);

• length of follow‐up;

• withdrawals from treatment protocol;

• number of participants who experienced delays in treatment or received all, part or none of the proposed treatment;

• risk of bias in study (see below);

• outcomes OS, PFS, QoL and adverse events:

  • for each outcome, outcome definition,

  • unit of measurement (if relevant),

  • for scales, upper and lower limits, and whether high or low score is good,

  • results, number of participants allocated to each intervention group,

  • for each outcome of interest, sample size, missing participants.

Assessment of risk of bias in included studies  

The risk of bias in included RCTs will be assessed using the Cochrane Collaboration's tool and the criteria specified in Chapter 8 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). This will include an assessment of:

• sequence generation;

• allocation concealment;

• blinding (of participants, healthcare providers and outcome assessors);

• incomplete outcome data, where we will code a satisfactory level of loss to follow‐up for each outcome as:

  • Yes, if fewer than 20% of women are lost to follow‐up and reasons for loss to follow‐up were similar in both treatment arms,

  • No, if more than 20% of women are lost to follow‐up or reasons for loss to follow‐up differed between treatment arms,

  • Unclear, if loss to follow‐up is not reported;

• selective reporting of outcomes;

• other possible sources of bias.

The risk of bias tool will be applied independently by two review authors (TS, AB) and differences will be resolved by discussion. Results will be presented in the risk of bias tables and also in both a risk of bias graph and a risk of bias summary plot. Results of the meta‐analyses will be interpreted in light of the findings with respect to risk of bias.

Measures of treatment effect  

We will use the following measures of the effect of treatment.

• For time to event data, we will use the HR.

• For dichotomous outcomes, we will use the RR.

• For continuous outcomes, we will use the mean difference between treatment arms (if trials measured outcomes on the same scale) or standardised mean differences (if trials measured outcomes on different scales).

Dealing with missing data  

We will not impute missing outcome data for the primary outcome. If data are missing or only imputed data are reported, we will contact trial authors to request data on the outcomes only among participants who were assessed.

Assessment of heterogeneity  

Heterogeneity between studies will be assessed by visual inspection of forest plots, by estimation of the percentage heterogeneity between trials which cannot be ascribed to sampling variation (Higgins 2003), by a formal statistical test of the significance of the heterogeneity (Deeks 2001) and, if possible, by subgroup analyses (see below). If there is evidence of substantial heterogeneity, the possible reasons for this will be investigated and reported.

Assessment of reporting biases  

We will not assess the potential for small study effects such as the potential for publication bias because, given the rarity of ovarian carcinosarcoma, there will be an insufficient number of included studies in which to make this assessment. We did not identify any included trials from 1950 to October 2010 so future updates in the short term are unlikely to yield many trials in this area.

Data synthesis  

If sufficient clinically similar studies are available, their results will be pooled in meta‐analyses. Adjusted summary statistics will be used, if available; otherwise unadjusted results will be used.

• For time to event data, HRs will be pooled using the generic inverse variance facility of RevMan 5.

• For dichotomous outcomes, the RR will be calculated for each study and these will then be pooled. 

• For continuous outcomes, the mean differences between the treatment arms at the end of follow‐up will be pooled if all trials measured the outcome on the same scale, otherwise standardised mean differences will be pooled. 

Random‐effects models with inverse variance weighting will be used for all meta‐analyses (DerSimonian 1986).

Subgroup analysis and investigation of heterogeneity  

Subgroup analyses will be performed, grouping the trials by:

• disease‐free interval (DFI).

Factors such as age, stage, length of follow‐up, adjusted or unadjusted analysis will be considered in the interpretation of any heterogeneity. 

Sensitivity analysis  

Sensitivity analyses will be performed excluding trials which did not report: (i) concealment of allocation, and (ii) blinding of the outcome assessor.