Criteria for considering studies for this review
Types of studies
All randomised controlled trials (RCTs) looking at chemotherapy regimens containing capecitabine alone or in combination versus a control arm employing a similar regimen without capecitabine for the treatment of breast cancer.
It is anticipated that three or greater RCTs will be found for each section of this review. However, if fewer than three RCTs are found for any of the three sections of this review, well-designed non-randomised controlled trials will be considered.
Types of participants
Women who have been histologically diagnosed with breast adenocarcinoma.
We will include trials where:
Types of interventions
Intervention: chemotherapy regimens containing capecitabine alone or as part of combination therapy in ER-positive and ER-negative breast cancer.
Comparator: similar chemotherapy regimens not containing capecitabine in ER-positive and ER-negative breast cancer. The comparator can include: (a) the same chemotherapy regimen without capecitabine, (b) a different chemotherapy regimen without capecitabine, and (c) the same chemotherapy regimen with another drug or drugs substituting for capecitabine.
Comparisons will include:
capecitabine-containing regimen versus non-capecitabine containing regimen in ER-positive breast cancer, and
capecitabine-containing regimen versus non-capecitabine containing regimen in ER-negative breast cancer.
We will also include studies of strategies that may be:
chemotherapy given as neoadjuvant, adjuvant or palliative treatment;
inclusive of biologic agents such as trastuzumab and bevacizumab if relevant and provided identical biologics are included in capecitabine-containing and non-capecitabine containing arms;
chemotherapy given as first or subsequent line of treatment in the context of metastatic disease.
Types of outcome measures
Overall response rate (ORR)
Progression-free survival (PFS)
Clinical benefit rate (CBR)
Specific information on adverse events will be collected from studies in each of the neoadjuvant, adjuvant and palliative chemotherapy groups. The total number of grade III and IV adverse events along with the total number of participants at risk in each trial will be summed to calculate a single odds ratio. For the following specific toxicities of interest, we will calculate the total number of toxic events:
The following outcome definitions will apply.
RFS is defined as the time from randomisation to the date of diagnosis of invasive breast cancer recurrence or death if the patient died before the recurrence of cancer.
If time to progression or time to treatment failure are recorded as endpoints rather than PFS, these can be used in place of PFS.
DFS is defined as the time from randomisation to the time of identification of recurrent or metastatic cancer or death from any cause.
OS is defined as the time from randomisation to death.
BCSS is defined as the time from randomisation to death due to breast cancer.
Response rate is defined using Response Evaluation Criteria In Solid Tumours (RECIST) (Eisenhauer 2009): the overall response is defined as the sum of complete response and partial response, representing the best response of each patient.
Search methods for identification of studies
We will search the following databases.
(a) The Cochrane Breast Cancer Group (CBCG) Specialised Register. Details of the search strategies used by the CBCG for the identification of studies and the procedure used to code references are outlined in the Group's module (http://www.mrw.interscience.wiley.com/cochrane/clabout/articles/BREASTCA/frame.html). Trials with the key words 'breast neoplasm; breast cancer; breast carcinoma; breast adenocarcinoma; breast tumour/tumor; capecitabine and xeloda' will be extracted and considered for inclusion in the review.
(b) MEDLINE (via OvidSP). See Appendix 1.
(c) EMBASE (via EMBASE.com). See Appendix 2.
(d) CENTRAL. See Appendix 3.
(e) The WHO International Clinical Trials Registry Platform (ICTRP) search portal (http://apps.who.int/trialsearch/Default.aspx) for all prospectively registered and ongoing trials. See Appendix 4.
(f) Clinicaltrials.gov (http://clinicaltrials.gov/). See Appendix 5.
Searching other resources
(a) Bibliographic searching
We aim to identify further studies from reference lists of identified relevant trials or reviews. A copy of the full article for each reference reporting a potentially eligible trial will be obtained. Where this is not possible, attempts will be made to contact authors to provide additional information.
(b) Gray literature searching
We will search the conference proceedings of the following conferences from 1996 through to the present for relevant abstracts:
American Society of Clinical Oncology Annual Scientific Meeting;
San Antonio Breast Cancer Symposium;
American Society of Clinical Oncology Breast Cancer Symposium;
European Society of Medical Oncology Annual Scientific Meeting;
European Breast Cancer Conference.
Data collection and analysis
Selection of studies
The selection criteria will be applied to each reference identified independently by two authors (AW and PL).
With regard to the selection of studies:
neither review author will be blinded to the study title, authors or publication details;
any disagreements regarding the selection of a study will be resolved by a third author (AR or MB);
neither AW nor PL are content experts, although both are knowledgeable in the field;
AR is a content expert, MB is a expert in statistics;
all relevant studies are to be included and if required, studies will be translated;
significant excluded studies will be recorded in the 'Characteristics of excluded studies' table (references will not be included in this table if they obviously do not fulfil the inclusion criteria).
Data extraction and management
Data will be extracted from each publication or abstract independently by two authors (AW and PL).
With regard to the extraction of data, data will be extracted using standard electronic extraction forms (see Appendix 6). Individual data extraction forms will be required for each of the three treatment types studied: neoadjuvant, adjuvant and palliative treatment. The study, participant and treatment information is common to all three forms while information on outcomes, data and analysis will vary between the forms. Aggregate data will be extracted in the first instance, however if there are insufficient data then the trial co-ordinators will be contacted for more information. Individual patient data will be sought if necessary.
Any disagreements regarding the extraction of quantitative data will be resolved by a third author (AR or MB). For studies with more than one publication, data from each publication will be collated utilising a single data collection form or multiple data collection forms depending on the type of publication.
Assessment of risk of bias in included studies
The risk of bias will be assessed using the Cochrane Collaboration's risk of bias assessment tool in the Cochrane Handbook for Systematic Reviews of Interventions, Chapter 8.5 (Higgins 2011). Bias will be independently assessed by two authors (AW and PL) with any disagreements to be resolved by a third author (AR or MB). The areas of bias which will be assessed include:
The important sources of bias in our review are likely to be selection bias, attrition bias and reporting bias. While all sources of bias will be reported on, these three areas of bias will be prioritised. The assessment of bias will be reported in this review using a 'risk of bias summary' figure. We will present multiple (stratified) analyses to incorporate the findings of our risk of bias assessment with a stratified forest plot.
Measures of treatment effect
The primary outcome for neoadjuvant trials is the pathologic complete response rate (pCR). In most trials this is measured using the Modified Regression Scale (von Minckwitz 2008) where the response is graded as follows.
Grade 5 - no microscopic evidence of residual tumour cells in the breast or axillary nodes.
Grade 4 - no microscopic evidence of residual tumour cells in the breast, but involved axillary nodes.
Grade 3 - residual non-invasive tumour cells in the breast.
Grade 2 - residual focal invasive tumours cells in the breast of ≤ 5 mm.
Grade 0-1 - all remaining scenarios including the presence of new invasive tumour.
Grade 4 and 5 are considered to represent pCR. This is a small ordinal scale where the event of pCR will be considered as a dichotomous outcome with Grade 4 and 5 representing pCR and all other grades representing no pCR. This outcome will be presented as a risk ratio (RR) with a 95% confidence interval (CI). The outcome will be reported for randomised and assessable patients. We will report the ratio of treatment effect for response so that a RR of less than 1.0 favours the non-capecitabine containing regimen and a RR of greater than 1.0 favours the capecitabine-containing regimen. Care will be required in the context of neoadjuvant pCRs as these are typically significantly higher in ER-negative cancers relative to ER-positive cancers. As such, in the ER-positive disease an absolute difference of 5% or greater for capecitabine-containing regimens compared with non-capecitabine containing regimens will be a relevant clinical difference.
The secondary outcomes for neoadjuvant trials are disease-free survival (DFS), recurrence-free survival (RFS) and overall survival (OS). These will be analysed as time-to-event outcomes and expressed as hazard ratios (HRs). We will use the HR provided in each study or estimate the HR indirectly using methods described by Tierney et al and Parmar et al (Parmar 1998; Tierney 2007). For meta-analytic pooling, we will use the generic-inverse variance method as described in the Cochrane Handbook for Systematic Reviews for Interventions, Chapters 7.7.6 and 9.4.9 (Higgins 2011).
The primary outcome for adjuvant trials is overall survival (OS), which will be analysed as a time-to-event outcome and expressed as a HR. We will use the HR provided in each study or estimate the HR indirectly using methods described by Tierney et al and Parmar et al (Parmar 1998; Tierney 2007). For meta-analytic pooling, we will use the generic-inverse variance method as described in the Cochrane Handbook of Systematic Reviews for Interventions, Chapters 7.7.6 and 9.4.9.
The secondary outcomes for adjuvant trials are RFS, DFS and breast cancer specific survival. These will be analysed as time-to-event outcomes and expressed as HRs. We will use the HR provided in each trial publication or estimate the HR indirectly using methods described by Tierney et al and Parmar et al (Parmar 1998; Tierney 2007). For meta-analytic pooling, we will use the generic-inverse variance method as described in the Cochrane Handbook of Systematic Reviews for Interventions, Chapters 7.7.6 and 9.4.9.
In the adjuvant setting, an absolute improvement of 5% or greater in the RFS for capecitabine-containing regimens compared with non-capecitabine containing regimens will be considered as a relevant clinical difference for ER-positive disease.
The primary outcome for palliative intent trials is OS, which will be analysed as a time-to-event outcome and expressed as a HR. We will use the HR provided in each study or estimate the HR indirectly using methods described by Tierney et al and Parmar et al (Parmar 1998; Tierney 2007). For meta-analytic pooling, we will use the generic-inverse variance method as described in the Cochrane Handbook of Systematic Reviews for Interventions, Chapters 7.7.6 and 9.4.9.
The secondary outcomes for palliative intent trials are overall response rate (ORR), progression-free survival (PFS) and clinical benefit rate (CBR). ORR is considered a small ordinal scale which will be expressed as a dichotomous outcome with complete response (CR) and partial response (PR) representing overall response and stable disease (SD) and progressive disease (PD) representing no response. This outcome will be presented as a RR with 95% CI and will be reported for randomised and assessable patients. We will report the ratio of treatment effect for response so that a RR of less than 1.0 favours the non-capecitabine containing regimens and a RR of greater than 1.0 favours the capecitabine containing regimens. CBR will also be considered a small ordinal scale and expressed as a dichotomous outcome with CR, PR and SD representing clinical benefit and PD representing no benefit. The outcome will be presented as a RR with 95% CI as per the ORR above. PFS will be analysed as a time-to-event outcome as above.
Care will also need to be taken in the context of metastatic disease when comparing both response rates and overall survivals as ER-negative disease may natively carry higher response rates in general whereas ER-positive disease is known to carry a longer median survival than ER-negative disease regardless of the treatment parameters. In this setting, an absolute 10% or greater improvement in ORR in ER-positive disease for capecitabine-containing regimens compared with non-capecitabine containing regimens will be a relevant clinical difference.
All grade III and IV adverse events, along with the total number of participants at risk, will be recorded from each trial. Where possible, data on adverse events will be collected for the treated population rather than the intention-to-treat population. A pooled odds ratio (OR) with 95% CI will be calculated for each toxicity where it is reported in two or more studies. The total number of the following specific adverse events will be recorded in this review: cytopaenias; hand-foot syndrome; mucositis; diarrhoea; and ischaemic cardiac disease.
Unit of analysis issues
Cross-over trials will not be included in this systematic review.
There will not be issues of multiple events per participant or cluster-randomised trials in this review.
Included studies may include those with multiple intervention groups. Specialist statistical advice will be sought regarding the manner in which the multiple intervention groups will be dealt with. Each study utilising multiple groups will need to be considered independently with multiple groups handled either by combining intervention groups or dividing the control group to enable pair-wise comparisons and to ensure that no unit of analysis issues arise.
Dealing with missing data
In this systematic review, missing data will be of significance as it is anticipated that many of the studies meeting eligibility criteria for inclusion may not report the study outcomes based on the hormone receptor status of the participants. This will mean that analysis of the study for inclusion in the systematic review is not possible.
The original investigators will be contacted by written correspondence to gain information regarding the hormone receptor status of participants in order to enable analysis for review. Studies for which this information is unable to be obtained will be included in the review, and the impact of the missing data for these studies will be discussed in the 'Discussion' section of the review.
With regard to studies in which other data are missing, for example participants who are lost to follow up or data for study objectives which are not reported:
the original investigators will be contacted by written correspondence where possible;
analysis will be by intention to treat, with a sensitivity analysis to consider the impact of the missing results;
missing data will not be imputed;
if individual patient data are obtained from the original investigators, we will be able to use multiple imputation methods if less than 20% of the patient data are missing;
the impact of missing data will be discussed in the 'Discussion' section of the review with regard to the assessment of risk of bias.
Assessment of heterogeneity
Clinical heterogeneity will be assessed using:
A random-effects model will likely be used to address heterogeneity, depending on evidence of statistical heterogeneity and the exploration of sources of heterogeneity. Statistician advice will be sought for the analysis of heterogeneity.
Assessment of reporting biases
Funnel plots will be used to detect publication bias in this review. Separate funnel plots will be used for each of the areas examined, including neoadjuvant, adjuvant and palliative-intent treatment.
Funnel plot asymmetry may be caused by:
There are instances in which publication bias may not lead to asymmetry in the funnel plot. Furthermore, visual inspection of the funnel plot for asymmetry alone is subjective and may lead to failure to detect publication bias. As such, funnel plot asymmetry is limited with respect to determination of publication bias. Contour-enhanced funnel plots will be used to assist in the differentiation of publication bias from other causes of funnel plot asymmetry.
Testing for funnel plot asymmetry in this review may be limited by the number of studies included for each of the primary outcome measures. It is likely that there will be fewer than 10 studies for each of the three arms for this review. However, if there are a sufficient number of studies, funnel plot asymmetry may be tested using the methods listed in the Cochrane Handbook for Systematic Reviews of Interventions (Chapter 10.4.3) together with the assistance of our statistician.
If evidence of small-study effects are found as part of the assessment of reporting bias, a sensitivity analysis will be performed in which the fixed-effect and random-effects model estimates will be compared.
Other possible sources of publication bias include duplicate or multiple publication bias; location bias; citation bias; language bias; and outcome reporting bias, all of which may affect this review. We will endeavour to detect multiple publications of the same study, although we appreciate the difficulties in doing this. We will be searching numerous electronic databases, including trial registries, to minimise location biases and we have not limited our inclusion criteria by language. Where possible, we will be contacting corresponding authors to obtain information regarding study outcomes to reduce the risk of outcome reporting bias.
Dichotomous outcomes will be pooled using the Mantel-Haenszel fixed-effect model method.
Time-to-event outcomes will be pooled using the generic inverse-variance method, allowing a mixture of log-rank and Cox model estimates to be obtained from studies.
The RevMan software will be used to perform the analysis.
A summary of findings table using GRADEprofiler will be used to summarise the findings of the review.
Subgroup analysis and investigation of heterogeneity
Subgroups to be examined in this review include:
HER2 overexpression in each of neoadjuvant, adjuvant and palliative-intent treatment groups;
capecitabine as first or subsequent line of therapy in palliative intent treatment.
A subgroup analysis will be performed to assess for heterogeneity for each of the subgroups identified above.
If there are adequate data, sensitivity analyses will be performed to assess the robustness of results. It is difficult to predetermine the likely sensitivity analyses that will be performed. Likely sensitivity analyses to be included are exclusion of studies with high risk of bias and unpublished studies.