Criteria for considering studies for this review

Search methods for identification of studies

Our search is limited to publications after January 1, 1996; this is the date Baselga et al first presented data about the efficacy of trastuzumab in humans (Baselga 1996).

For randomised controlled trials (RCTs)
See: Cochrane Breast Cancer Group search strategy

The Cochrane Breast Cancer Group (CBCG) Specialized Register will be searched. Details of the search strategy applied to create the register and the procedure used to code references are described in the Group's module on The Cochrane Library. The register includes both published and unpublished (including ongoing) trials. The CBCG codes "early" and "immunotherapy" will be applied to the specialised register and combined with the following keywords (imported with the references from MEDLINE) "trastuzumab" [Substance Name], and a search of all non‐indexed fields for the following text words: Trastuzumab, Herceptin or monoclonal antibod* AND HER2.

To retrieve observational studies that report adverse events, we designed separate search strategies (as outlined below) and will consider additional sources of information.

Databases
MEDLINE (host: PubMed), January 1996 to May 2006;
EMBASE (host: Embase.com), January 1996 to May 2006;
BIOSIS (host: ISI Web of Knowledge), January 1996 to May 2006;
Cochrane Central Register of Controlled Trials (CENTRAL) (The Cochrane Library Issue 2, 2006);
TOXNET, National Library of Medicine;
ASCO Annual Meetings, American Society of Clinical Oncology, 1996 to 2005.

We will also check bulletins from regulatory agencies (Loke 2006):
UK, Current Problems in Pharmacovigilance (http://www.mhra.gov.uk/home/);
Australia, the Australian Adverse Drug Reactions Bulletin (http://www.tga.gov.au/adr/aadrb.htm);
European Public Assessment Reports from the European Medicines Evaluation Agency (http://www.emea.eu.int/#);
US, MedWatch, the Food and Drug Administration Safety information and Adverse Events Reporting Program (http://www.fda.gov/medwatch/elist.htm).

We will search CENTRAL, MEDLINE, EMBASE, TOXNET (1996 to May 2006) using the medical subject headings 'Breast Neoplasms', 'Antineoplastic Agents', 'Adverse effects' and 'Toxicity', and the text words 'Trastuzumab', 'Herceptin', 'Adverse effect', 'Side effect', 'Toxic effect', 'Drug toxicity', 'Dug tolerance', 'Causality', 'Risk', 'Adverse event', 'Adverse drug reaction', 'Breast cancer', 'Breast tumour', 'Breast tumor' and 'Breast neoplasm'. We will include reports irrespective of the language in which they were reported.

MEDLINE search strategy (PubMed)
1."Breast Neoplasms"[MeSH]
2.(breast OR mammary) AND (cancer* OR tumour* OR tumor* OR neoplasm* OR metastas* OR carcinoma) Field: Title/Abstract
3.#1 OR #2
4.trastuzumab OR herceptin
5."Antineoplastic Agents/adverse effects"[MeSH] OR "Antineoplastic Agents/controindications"[MeSH] OR "Antineoplastic Agents/toxicity"[MeSH]
6."Drug Hypersensitivity"[MeSH]
7."Drug Toxicity"[MeSH]
8."Drug Tolerance"[MeSH]
9."Causality"[MeSH]
10."Risk"[MeSH]
11."Product Surveillance, Postmarketing"[MeSH]
12.#5 OR #6 OR #7 OR #8 OR #9 OR #10 OR #11
13.safe* OR adr OR adrs OR tolerabilit* OR toxicit* OR "undesirable effect*" OR "adverse reaction*" OR hypersensitivit* Field: Title/Abstract
14.toxic effect* OR complication* OR causalit* OR risk* Field: Title/Abstract
15.(side OR adverse) AND (effect* OR event* OR outcome*) Field: Title/Abstract
16.postmarketing OR "post marketing" Field: Title/Abstract
17.#13 OR #14 OR #15 OR #16
18.#12 OR #17
19.#3 AND #4 AND #18
Our search will not be limited to human in the free‐text to include PubMed's in‐process records.

EMBASE search strategy
1.'breast cancer'/exp AND [humans]/lim AND [embase]/lim AND [1996‐2006]/py
2.(breast:ti,ab OR mammary:ti,ab OR mammaries:ti,ab) AND (cancer*:ti,ab OR tumour*:ti,ab OR tumor*:ti,ab OR neoplasm*:ti,ab OR metastas*:ti,ab OR carcinoma*:ti,ab) AND [humans]/lim AND [embase]/lim AND [1974‐2006]/py
3.#1 OR #2
4.'trastuzumab'/exp AND [humans]/lim AND [embase]/lim AND [1996‐2006]/py
5.(trastuzumab:ti,ab OR herceptin:ti,ab) AND [humans]/lim AND [embase]/lim AND [1974‐2006]/py
6.#4 OR #5
7.'antineoplastic agent'/exp/dd_ae,dd_to AND [humans]/lim AND [embase]/lim AND [1974‐2006]/py
8.'drug hypersensitivity'/exp AND [humans]/lim AND [embase]/lim AND [1974‐2006]/py
9.'drug toxicity'/exp AND [humans]/lim AND [embase]/lim AND [1974‐2006]/py
10.'drug tolerance'/exp AND [humans]/lim AND [embase]/lim AND [1974‐2006]/py
11.'risk'/exp AND [humans]/lim AND [embase]/lim AND [1974‐2006]/py
12.'postmarketing surveillance'/exp AND [humans]/lim AND [embase]/lim AND [1975‐2006]/py
13.(safe*:ti,ab OR adr:ti,ab OR adrs:ti,ab OR tolerability*.:ti,ab OR toxicit*:ti,ab OR undesirable:ti,ab AND effect*:ti,ab OR adverse:ti,ab AND reaction*:ti,ab OR hypersensitivit*:ti,ab OR toxic:ti,ab AND effect*:ti,ab OR complication*:ti,ab OR causalit*:ti,ab OR risk:ti,ab OR postmarketing:ti,ab OR post:ti,ab AND marketing:ti,ab) AND [humans]/lim AND [embase]/lim AND [1974‐2006]/py
14.(side:ti,ab OR adverse:ti,ab) AND (effect*:ti,ab OR event*:ti,ab OR outcome*:ti,ab) AND [humans]/lim AND [embase]/lim AND [1974‐2006]/py
15.#7 OR #8 OR #9 OR #10 OR #11 OR #12 OR #13 OR #14
16.#3 AND #6 AND #15

CENTRAL
1.MeSH descriptor Breast Neoplasms, this term only in MeSH products
2.MeSH descriptor Adverse Drug Reaction Reporting Systems, this term only in MeSH products OR MeSH descriptor Drug Toxicity, this term only in MeSH products
3.trastuzumab OR herceptin in All Fields in all products
4.#2 AND #3
5.#1 AND #4
6.breast AND (cancer* OR tumour* OR tumor* OR neoplasm*)
7.#6 AND #3
8.adverse effect* OR side effect* OR toxic effect* OR adverse event* OR adverse drug reaction*
9.#7 AND #8
10.#9 OR #5

BIOSIS
1. Text words 'breast cancer*', 'breast neoplasm*', 'breast tumour*' and 'breast tumor*' were searched.
2. Text words 'trastuzumab' and 'herceptin' were searched and linked to the major concept 'oncology' (definition: clinical studies of the characteristics, diagnosis, and treatment of human tumors and cancers) and concept code 'toxicology‐pharmacology' (definition: studies of toxic effects of drugs).
3. Search 1 and 2 (above) were linked and limited to publication year (1996 to 2006) and humans.

A copy of the full article for each reference reporting a potentially eligible trial will be obtained.

Data collection and analysis

Study selection
The titles and abstracts of articles found in the search will be independently screened for inclusion by two reviewers (LPM and CB). For unpublished trials, available information from conference proceedings will be assessed. Disagreements will be resolved by discussion. Further information will be sought from the authors where papers contain insufficient information to make a decision about eligibility. The selection criteria described above will be applied to each trial. Reasons for exclusion will be recorded.

Data extraction methods and their reliability
Two reviewers (from LPM, AC and CB) will independently extract information using the pro forma process piloted on a random sample of papers investigating other chemotherapy agents. We will record details of study design, participants, setting, interventions, follow‐up, quality components, efficacy outcomes and side effects. The extraction form will be available from the authors upon request. Details of previous therapy given to patients (including endocrine or systemic therapy) will be recorded. A third review author will resolve any discrepancies regarding the extraction of quantitative data or the quality assessment of RCTs. When a trial is presented in abstract form, further information will be searched for (from the internet, contacting the authors, checking for the next best available resource or publication).

Where possible, any missing data will be sought from the authors. For studies with more than one publication, we will extract data from all the publications; however, the final or updated version of each trial will be considered to be the primary reference. We will contact the trastuzumab manufacturers (Genentech Inc and Roche Holding Ltd) for additional data.

Quality assessment for RCTs
We will assess separately the methodological quality of the trials based on five components: patient randomisation (both generation of the allocation sequence and allocation concealment), blinding, completeness of follow‐up and if a multicenter study (Schulz 1995; Moher 1998; Kjaergard 2001; Juni 2001; Balk 2002). We will use the following definitions in the assessment of the methodological quality.

Generation of the allocation sequence
1. Adequate: if the allocation sequence is generated by a computer or random number table. Drawing of lots, tossing of a coin, shuffling of cards, or throwing die is considered as adequate if a person who was not otherwise involved in the recruitment of participants performed the procedure.
2. Unclear: if the trial is described as randomised but the method used for the allocation sequence generation is not described.
3. Inadequate: if a system involving dates, names, or admittance numbers are used for the allocation of patients. These studies are known as quasi‐randomised and will be excluded from the present review for efficacy data. Trials with alternating allocation will be excluded also.

Allocation concealment
1. Adequate: if the allocation of patients involved a central, independent unit; on‐site locked computer; identically appearing numbered drug bottles or containers prepared by an independent pharmacist or investigator; or sealed, opaque envelopes.
2. Unclear: if the trial is described as randomised but the method used to conceal the allocation is not described.
3. Inadequate: if the allocation sequence is known to the investigators who assigned participants, the envelopes are unsealed or transparent, or if the study is quasi‐randomised.

Blinding (or masking)
1. Adequate: if the trial is described as double blind and the method of blinding involves identical placebo or active drugs. Particularly:
‐ double blind (method described and use of a placebo(s) or dummy technique meant neither the participant nor the care provider or assessor knew which treatment was given);
‐ single blind (participant or the care provider or assessor aware of the treatment given).
2. Unclear: if the trial is described as double blind or single blind but the method of blinding is not described.
3. Not performed: if the trial is open label (all parties aware of treatment).

Follow‐up
1. Adequate: if the numbers and reasons for dropouts and withdrawals in all intervention groups were described and if 90% or more of the participants randomised were included in the analysis; or if it is specified that there were no dropouts or withdrawals.
2. Unclear: if the report gives the impression that there were no dropouts or withdrawals but this was not specifically stated.
3. Inadequate: if less than 90% of participants randomised into the trial were included in the analysis or the number or reasons for dropouts and withdrawals were not described.

Multicentre
1. Adequate: if the study includes more than one site and the recruitment is well balanced between participating centres.
2. Unclear: if the report gives the impression that there was more than one site but this is not specifically stated.
3. Inadequate: if the study includes only one site or the recruitment is not balanced between participating centres.

Authors of the primary trial reports will be contacted when necessary to clarify data and to obtain missing information. Two reviewers (from LM, AC, CB) will independently assess trials according to the predefined quality criteria. Any disagreements will be resolved by consulting a third party. The characteristics and outcomes of the included trials and details of the excluded trials will be entered into our database. Sensitivity analysis or meta‐regression to assess the effect of methodological quality is planned.

Quality assessment for observational studies
We will separately assess the methodological quality of observational studies by using a component approach considering: concurrent, concomitant treatment; how allocation occurred; any attempt to balance groups by design; blinding of outcome assessment; completeness of follow‐up; identification of prognostic factors (ie cardiovascular risk factors) and case‐mix adjustment. These components are part of a list of quality items identified through a systematic review of the literature (Deeks 2003). Overall, the review team judged seven quality items to be potentially useful for some of the secondary outcomes (i.e. cardiac toxicity) of this systematic review.

We will not evaluate the quality of case series or single case reports.

Statistical Analysis

Effectiveness
We will extract the hazard ratio (HR) and associated variances for overall survival and disease‐free survival directly from the trial publications. If not reported, we will obtain these data indirectly, using the methods described by Parmar 1998, employing either other available summary statistics or data extracted from published Kaplan‐Meier curves. To allow for immature follow‐up, we will adjust the numbers at risk based upon estimated minimum and maximum follow‐up times. If these are not reported in any of the reports available, we will estimate minimum follow‐up using the estimated time taken to complete treatment. We will estimate maximum follow‐up using the last event reported in the relevant time‐to‐event curve. We will present these follow‐up estimates in the 'Characteristics of included studies' table, under 'Notes'.

We will obtain a pooled HR through the generic inverse variance approach from the log HR and the standard error of the log HR, using the fixed‐effect and random‐effect models, along with assessments of heterogeneity (Deeks 2006). The pooled HR represents the overall risk of an event on chemotherapy regimens containing trastuzumab versus those not containing trastuzumab.

We will report ratios of treatment effects for time‐to‐event outcomes so that HRs less than 1.0 favour regimens containing trastuzumab and values greater than 1.0 favour regimens that do not contain trastuzumab. The forest plots for overall survival and progression‐free survival will present the hazard ratio (HR) and 95% confidence intervals.

If all arms in a multi‐arm trial are to be included in the meta‐analysis and one treatment arm is to be included in more than one of the treatment comparisons, then we will divide the number of events and the number of participants in that arm by the number of treatment comparisons made. This method will avoid the multiple use of participants in the pooled estimate of treatment effect while retaining information from each arm of the trial. It will compromise the precision of the pooled estimate slightly.

Heterogeneity
We will include all outcomes available from the individual studies in the meta‐analysis, with heterogeneity reported by the Q (Chi‐squared) and the I‐squared statistics (see the Cochrane Handbook). Statistical significance of the Q (Chi‐squared) statistic will be judged by P < 0.10 because of the low statistical power of the test. The I‐squared statistic indicates the percent variability due to between‐study (or inter‐study) variability, as opposed to within‐study (or intra‐study) variability. An I‐squared value greater than 50% will be considered to be large (Higgins 2002). If there is no statistical evidence for heterogeneity effect sizes, we will use the fixed‐effect model. Where statistically significant heterogeneity exists, we will conduct careful clinical review of the data for the source of such heterogeneity. Based on this review, the reviewers will make a decision to either: (1) redo the analysis using the homogenous subgroup (only if a clear and compelling reason to exclude the heterogeneous data can be made); (2) abandon statistical combining of the trials in favor of a narrative review of the literature; or (3) redo the analysis using the random‐effects model (DerSimonian 1986).

It is likely that we will collect quality of life data using a variety of instruments across trials. If data are expressed using continuous data, we will present results as weighted mean differences (WMD). However, when different scales are used, we will use standardised mean differences (SMD). If it is not possible to pool the results for clinical or methodological heterogeneity, we will not statistically synthesise data but will summarise and evaluate qualitatively.

We have pre‐specified three subgroup analyses: 1. analysis by type of adjuvant chemotherapy regimen (cyclophosphamide, methotrexate, 5‐fluorouracil (CMF); anthracycline combinations, other combinations, single agents); 2. sensitivity analysis using high‐quality methodological components (high‐quality trials versus other); 3. status of nodal involvement (lymph node (LN) positive versus LN negative, four or more LNs involved versus three or fewer LNs involved).

Toxicity
We will report toxicity data as death related to treatment and as grade III or grade IV events of any toxicities: dyspnea, leukopenia, thrombocytopenia, anemia, infection, fever, chills, nausea, vomiting, nausea and vomiting, alopecia, diarrhea, fatigue, asthenia, constipation, dermatitis, skin, pain, stomatitis, hand‐foot syndrome. We will extract all adverse events related to cardiotoxicity irrespective of grade or severity. We expect that data about toxicities will be not consistently reported across the included trials and that different scales and definitions will have been used (heart failure, left ventricular ejection fraction, symptoms, etc). When adverse effects have been reported with explicit definitions, we will extract for any harm and combine with data from other trials with similar definitions. We will make a decision about which definition and scale to include in the final analysis by consensus after the data have been collected, with a preference for standardised and validated scales (for example World Health Organization toxicity criteria or New York Heart Association criteria for cardiac symptoms). A potential list of cardiac toxicities is presented in Table 1. We will consider possible disparities in adverse effects due to different schedules of trastuzumab (that is during anthracycline, before anthracycline, immediately after anthracycline, delayed after anthracycline or without anthracycline).

We will pool and summarise adverse events noted in the course of the randomised trials, and discuss their causal relationship to trastuzumab.

We will screen, summarise and discuss reports of adverse events linked with trastuzumab in other studies, such as outside RCTs. We cannot anticipate if it will be possible to pool these studies and calculate combined estimates. Any risk of bias in attempting to produce combined estimates will be seriously taken into consideration (Egger 2001).

We also cannot anticipate if it will be possible to evaluate adverse events, particularly cardiotoxicity, for distinct care phases (for example during chemotherapy treatment or follow‐up).

When we examine the possible harmful side effects of trastuzumab, we will adopt a higher type‐I error (α = 0.10) for some pre‐specified adverse events (Shadish 2002): cardiotoxicity, death related to treatment and dyspnea. For the remaining analyses we will adopt the same type‐I error (α = 0.10) with a Bonferroni correction for the number of comparisons (Pocock 1997); these will be exploratory data analyses and will not be over interpreted.