Tipos de estudios

Ensayos controlados con asignación aleatoria (ECA). Se consideraron los estudios informados como texto completo, los publicados como resumen solamente y los datos no publicados.

Tipos de participantes

Pacientes con cáncer de mama metastásico al momento del diagnóstico inicial: TNM (tumor, ganglios linfáticos, metástasis) estadio IV (Sobin 2002). Lo anterior incluye cuando el cáncer de mama se ha propagado más allá de la mama, la pared torácica y los ganglios regionales. No se aplicaron restricciones con respecto a la edad ni el tipo histológico. Si un estudio incluyera un subgrupo de participantes elegibles, se incluiría en la revisión siempre que se pudieran extraer los resultados relevantes.

Tipos de intervenciones

• Grupo de intervención: cirugía de mama del tumor primario, además de tratamiento sistémico (quimioterapia, tratamiento endocrino, inmunoterapia o agentes biológicos).

La cirugía de mama incluyó tumorectomía (resección tumoral con margen de seguridad) o mastectomía (extracción de todo el tejido de la mama). El tratamiento de la axila podría incluir biopsia del ganglio centinela con axila clínicamente negativa y disección axilar (niveles I o II solamente, o niveles I, II y III vaciamiento axilar) con axila clínicamente comprometida o ganglio linfático centinela positivo.

El tratamiento sistémico incluyó quimioterapia, tratamiento endocrino, inmunoterapia o agentes biológicos, así como otros tratamientos novedosos. En ambos brazos se permitió cualquier secuencia de los tratamientos. La cirugía se podría haber realizado antes o después del tratamiento sistémico.

Se aceptó la radioterapia realizada de forma sistemática como parte del tratamiento locorregional del cáncer de mama. La radioterapia se podría haber realizado en la mama o la pared torácica, y en la fosa supraclavicular y la axila ipsilateral, según las guías institucionales. Lo anterior podría haber incluido un refuerzo mediante terapia intersticial de electrones o haz externo y técnicas nuevas. También se permitió la radioterapia para el alivio del dolor o paliativa, en la enfermedad local o a distancia.

• Grupo de comparación: tratamiento sistémico sin cirugía de mama.

En el grupo control no se realizó cirugía de mama. Se permitió la resección del tumor primario solo para tratamiento paliativo como la hemorragia, la ulceración, la infección, el dolor o la progresión local.

De utilizarse cointervenciones como la quimioterapia, se debían aplicar por igual a cada grupo de estudio.

Tipos de medida de resultado

Se incluyeron los estudios que midieron al menos uno de los resultados enumerados a continuación.

Búsquedas electrónicas

We searched the following databases on the 22 February 2016:

• The Cochrane Breast Cancer Group (CBCG) Specialised Register. Details of search strategies used by the CBCG for the identification of studies and the procedures to code references are outlined in the CBCG's module (onlinelibrary.wiley.com/o/cochrane/clabout/articles/BREASTCA/frame.html). Trials with the keywords 'advanced breast cancer', 'metastatic breast cancer', 'breast surgery', 'breast‐conserving surgery', 'mastectomy', 'lumpectomy', 'segmentectomy', were extracted and considered for inclusion in the review.

• The Cochrane Central Register of Controlled Trials (CENTRAL, Issue 1, 2016). See Appendix 1.

• MEDLINE (via PubMed). See Appendix 2.

• Embase. See Appendix 3 (via Embase.com for first search) and Appendix 4 (via OvidSP for top‐up search).

• The World Health Organization (WHO) International Clinical Trials Registry Platform (ICTRP) search portal (apps.who.int/trialsearch/Default.aspx) for all prospectively registered and ongoing trials. See Appendix 5.

• ClinicalTrials.gov (clinicaltrials.gov/). See Appendix 6.

Búsqueda de otros recursos

We also searched for additional studies in the reference lists of identified trials and reviews. We obtained a full copy of each article reporting a potentially eligible trial.

We contacted the authors of the primary studies to seek unpublished data or additional information about the outcomes of interest.

We conducted an additional search in the American Society of Clinical Oncology (ASCO) conference proceedings in July 2016.

Selección de los estudios

Two review authors (GT and BSM) independently screened the titles and abstracts to identify potentially relevant RCTs. We then obtained full‐text articles of all potentially relevant citations. Any disagreements regarding RCT selection were resolved by consulting a third review author (RR). There were no restrictions regarding language or reporting status. We summarised the excluded studies in the Characteristics of excluded studies table. The entire study selection process was reported according to PRISMA guidelines (Moher 2009).

Extracción y manejo de los datos

Two review authors (GT and BSM) independently conducted full data extraction, consulting a third review author (RR) to help resolve any disagreements. We contacted the authors from primary studies to seek unpublished data or additional information about the outcomes of interest (Badwe 2015; Soran 2016). We developed and piloted the data extraction forms, which included the following information from individual studies:

• publication details;

• study design, study setting, inclusion/exclusion criteria;

• participant population (e.g. age, type of surgical procedure, type of tumour);

• details of intervention;

• outcome measures;

• withdrawals, length and method of follow‐up, and the number of participants followed up.

When available, we pooled quantitative data and carried out the analyses using Review Manager 5 software (RevMan 2012).

In the case of studies with more than one publication, we extracted data from all publications and used them to fill in the same data extraction form, since the unit of analysis was the study rather than the publication. The primary reference was the first publication.

Evaluación del riesgo de sesgo de los estudios incluidos

We assessed risk of bias using Cochrane's 'Risk of bias' tool as described in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011).

Two review authors (GT and BSM) rated the studies on all seven domains of the 'Risk of bias' tool, consulting a third review author (MRT or RR) to help resolve disagreements. For each domain, we classified the study as having a low, unclear, or high risk of bias. We evaluated the following seven domains:

• random sequence generation;

• allocation concealment;

• blinding of participants and personnel;

• blinding of outcome assessors;

• incomplete outcome data;

• selective reporting;

• other bias.

We took into account the risk of bias in the Authors' conclusions.

Medidas del efecto del tratamiento

We reported time‐to‐event outcomes such as overall survival, PFS, and breast cancer‐specific survival as hazard ratios (HR) with 95% confidence interval (CI). If necessary, we indirectly estimated HRs using the methods described by Parmar and colleagues (Parmar 1998). We performed intention‐to‐treat analyses.

We intended to report continuous outcomes such as quality of life as mean differences (MD) with 95% CIs. We planned to use standardised mean differences (SMD) if continuous outcomes were reported in different scales.

We reported dichotomous outcomes such as toxicity as risk ratios (RR).

In cases where it was not possible to pool data through meta‐analysis, we presented outcome data narratively.

In an ASCO consensus meeting, the members considered that relevant improvements in median overall survival of at least 20% are necessary to define a clinically meaningful improvement in outcome. They considered HR < 0.8, corresponding to an improvement in median overall survival within a range of 2.5 to 6 months, as the minimum incremental improvement over standard therapy that would define a clinically meaningful outcome (Ellis 2014). We used this definition in the review.

Cuestiones relativas a la unidad de análisis

The unit of analysis was the individual participant, rather than surgical unit, hospital, or centre.

Manejo de los datos faltantes

Where data were missing or unsuitable for analyses (e.g. intention‐to‐treat data were not available), we contacted the study authors to request further information. If data could not be obtained, we did not include the study in the meta‐analysis and discussed these results in the review.

Evaluación de la heterogeneidad

First, we inspected heterogeneity graphically using forest plots displaying effects of individual studies with 95% CIs.
If appropriate, we assessed heterogeneity between studies using the Chi² statistic (considering a P value < 0.10 as significant). We also used the I² statistic as an approximate guide to interpret the magnitude of heterogeneity: an I² value between 30% and 60% was indicative of moderate heterogeneity, while values greater than 50% were considered substantial heterogeneity (Higgins 2011).

We investigated the following factors as potential causes of heterogeneity in the included studies, using the framework described below.

1. Clinical heterogeneity: related to study location and setting, full characteristics of participants, comorbidity, and treatments that women may be receiving at trial entry. We considered how outcomes were measured, the definition of outcomes, and how they were recorded.

2. Methodological heterogeneity: related to randomisation process and overall methodological quality of primary studies.

Evaluación de los sesgos de notificación

We planned to create a funnel plot to explore the risk of reporting bias in the case of a meta‐analysis including at least 10 trials. However, none of the meta‐analyses achieved this number of studies. We also planned to perform exploratory analyses to investigate possible reasons for visual asymmetry of funnel plot (chance, publication bias, and true heterogeneity).

Síntesis de los datos

We synthesised data using Review Manager 5 software (RevMan 2012). To estimate the effect size, we planned to use the MD or SMD for continuous outcomes. We used the HR for time‐to‐event outcomes and RR for dichotomous outcomes. We used the inverse variance method to estimate the combined effect size for the outcomes. We used the random‐effects model by default, as we expected clinical or methodological heterogeneity, or both, among the included studies. Where the data were too diverse for combining effect sizes in a meaningful or valid manner, we presented the results of individual studies in table and graphical format and used a narrative approach to summarise data.

Análisis de subgrupos e investigación de la heterogeneidad

As we expected there to be a small number of published studies, we anticipated that subgroup analyses would not be feasible. Whenever possible, we considered the following subgroup analyses:

• age > 55 years or < 55 years;

• oestrogen receptor (ER) status (Lang 2013; Neuman 2010; Perez‐Fidalgo 2011; Petrelli 2012; Samiee 2012);

• HER2 status (Neuman 2010; Samiee 2012);

• only bone metastases (Rapiti 2006; Rhu 2015);

• radiotherapy at primary site or not (Bourgier 2009; Le Scodan 2009).

We based the choice of these planned subgroups on published evidence that the intervention effect could be modified by these characteristics.

Análisis de sensibilidad

We intended to conduct the following sensitivity analyses:

• by excluding trials at high risk of bias in more than four of the seven domains;

• by removing studies with eligibility criteria that differed markedly from most of the included studies;

• by excluding studies in which it was necessary to re‐estimate HRs and CIs using other accepted methodologies;

• by excluding studies that used any imputation methods for missing data

The sensitivity analysis was not possible due to the small number of studies, but will be considered for future updates of this review.