Criteria for considering studies for this review
Types of studies
We included all randomised clinical trials assessing beneficial and harmful effects of transarterial (chemo)embolisation irrespective of publication status, language, or blinding. Quasi-randomised and other controlled studies that came up with the search results were considered only for the report of data on harm.
Types of participants
Patients with liver metastases no matter the location of the primary tumour.
Types of interventions
Transarterial (chemo)embolisation compared with no intervention or placebo intervention. Co-interventions were allowed if provided equally to the experimental and control groups of the individual randomised trial.
Types of outcome measures
1. Mortality at last follow-up.
2. Time to mortality.
3. All adverse events and complications, separately and in total. The International Conference on Harmonisation (ICH) Guidelines (ICH-GCP 1997) defines adverse events as serious and non-serious. A serious fatal or non-fatal adverse event is any event that leads to death, is life-threatening, requires in-patient hospitalisation or prolongation of existing hospitalisation, results in persistent or significant disability, and any important medical event which may have jeopardised the patient or requires intervention to prevent it. All other adverse events are considered non-serious.
4. Quality of life.
1. Failure or proportion of patients with recurrence.
2. Time to progression of liver metastasis.
3. Tumour response measures (complete response, partial response, stable disease, disease progression).
Search methods for identification of studies
We searched the Cochrane Hepato-Biliary Group Controlled Trials Register (Gluud 2013), the Cochrane Central Register of Controlled Trials (CENTRAL) in The Cochrane Library, MEDLINE, EMBASE, Science Citation Index Expanded, LILACS, and CINAHL (Royle 2003) as well as the World Health Organization (WHO) International Trial Registry platform (WHO 2011).
One global search was used for all non-surgical ablation methods for primary malignant liver tumours and liver metastases. Search strategies with the time spans of the searches are given in Appendix 1 (up to December 2012). There was no need to improve the search strategies.
In addition, we assessed for inclusion all United States Food and Drug Administration (FDA) approvals and investigational device exemptions as found on the FDA website (FDA 2011).
Searching other resources
We searched reference lists of reviews (such as Schwartz 2004 and Lopez 2006), Health Technology Assessment (HTA) reports (such as ASERNIP-S 2006), all Cochrane reviews, and all trials that were included for relevant studies.
Data collection and analysis
We performed the systematic review following the recommendations of The Cochrane Collaboration (Higgins 2011) and the Cochrane Hepato-Biliary Group module (Gluud 2013) using Review Manager 5 (RevMan 2012).
Selection of studies
Two authors independently evaluated titles and abstracts for ordering papers (RR and MB). Any differences in opinion were resolved by discussion or, if necessary, by consulting a third author (JK). For titles and abstracts that potentially fitted our inclusion criteria, full papers were ordered. These papers were assessed by two independent authors (RR and MB) and differences in opinion, if any, were resolved using the above-mentioned procedure.
Data extraction and management
We extracted the relevant information on participant characteristics, interventions, study outcome measures, and data on the outcome measures for our review as well as information on the design and methodology of the trials. Quality assessment of the trials, assessment for fulfilling the inclusion criteria, and data extraction from the retrieved final evaluation trials were done by one author (RR, MB, or RW) and checked by a second author (RR, MB, or RW).
Assessment of risk of bias in included studies
We assessed the risk of bias of the trials based on the domains described below (Schulz 1995; Moher 1998; Kjaergard 2001; Gluud 2008; Wood 2008; Lundh 2012; Savović 2012a; Savović 2012b). This assessment was presented by trial and was used to describe the results of each trial in relation to reliability.
Allocation sequence generation
Low risk of bias: sequence generation was achieved using computer random number generation or a random number table. Drawing lots, tossing a coin, shuffling cards and throwing dice are adequate if performed by an independent adjudicator.
Uncertain risk of bias: the trial is described as randomised, but the method of sequence generation was not specified.
High risk of bias: the sequence generation method is not, or may not be, random. Quasi-randomised studies, those using dates, names, or admittance numbers in order to allocate patients, are inadequate and will be excluded for the assessment of benefits but not for harms.
Low risk of bias: allocation was controlled by a central independent randomisation unit, using sequentially numbered, opaque and sealed envelopes, or similar, so that intervention allocations could not have been foreseen either in advance of or during enrolment.
Uncertain risk of bias: the trial was described as randomised but the method used to conceal the allocation was not described so that intervention allocations may have been foreseen either in advance of or during enrolment.
High risk of bias: if the allocation sequence was known to the investigators who assigned participants or if the study was quasi-randomised. Quasi-randomised studies will be excluded for the assessment of benefits but not for harms.
Blinding of participants, personnel, and outcome assessors
Low risk of bias (blinding was performed adequately, or the outcome measurement is not likely to be influenced by lack of blinding).
Uncertain risk of bias (there is insufficient information to assess whether the type of blinding used is likely to induce bias on the estimate of effect).
High risk of bias (no blinding or incomplete blinding, and the outcome or the outcome measurement is likely to be influenced by lack of blinding).
Incomplete outcome data
Low risk of bias (the underlying reasons for missingness are unlikely to cause treatment effects departure from plausible values, or proper methods have been employed to handle missing data).
Uncertain risk of bias (there is insufficient information to assess whether the missing data mechanism in combination with the method used to handle missing data is likely to induce bias on the estimate of effect).
High risk of bias (the crude estimate of effects (eg, complete case estimate) will clearly be biased due to the underlying reasons for missingness, and the methods used to handle missing data are unsatisfactory).
Selective outcome reporting
Low risk of bias: pre-defined or clinically relevant and reasonably expected outcomes are reported on.
Uncertain risk of bias: not all pre-defined or clinically relevant and reasonably expected outcomes are reported on or are not reported fully, or it is unclear whether data on these outcomes were recorded or not.
High risk of bias: one or more clinically relevant and reasonably expected outcome was not reported on; data on these outcomes were likely to have been recorded.
Other sources of bias
Low risk of other bias: the trial appears to be free of other components that could put it at risk of bias.
Uncertain risk of other bias: the trial may or may not be free of other components that could put it at risk of bias.
High risk of other bias: there are other factors in the trial that could put it at risk of bias, eg, for-profit involvement, authors have conducted trials on the same topic etc.
Trials judged as having 'low risk of bias' in all of the above specified individual domains were considered to be 'trials with low risk of bias'. All other instances led to classifying the trials in the group of trials with high risk of bias.
Measures of treatment effect
For dichotomous variables, we planned to calculate the relative risk (RR) with 95% confidence interval (CI). For continuous variables, we planned to calculate the standardised mean difference (SMD) (for outcomes such as quality of life when different scales could be used) with 95% CI. For outcomes such as hazard ratio for death, we planned to use the generic inverse variance method for the meta-analysis.
Unit of analysis issues
The number of the randomised participants was to be used to calculate estimates of intervention effects and CIs. In cluster randomised trials, the unit of analysis would have been the cluster. For cross-over trials, we planned to include only data from the first intervention period (Higgins 2011). We calculated pooled estimates using the random-effects model (DerSimonian 1986) and the fixed-effect model (Mantel 1959; Greenland 1985). We planned to present both results if there were discrepancies in the results. If not, we planned to report the random-effects model. We planned to measure the quantity of heterogeneity using the I2 statistic (Higgins 2011).
Dealing with missing data
Data were planned to be analysed using the intention-to-treat principle, that is, patients with missing data (in all treatment groups of a trial) were to be considered as treatment failures and all randomised patients were to be included in the denominator.
Assessment of heterogeneity
Heterogeneity was to be assessed using the Chi2 and I2 statistics. Any plausible, possible causes of heterogeneity were to be discussed.
We followed the instructions given in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011) and the Cochrane Hepato-Biliary Group Module (Gluud 2013).
The evidence synthesis was done in a narrative way, and it was not possible to do meta-analyses.
In principle, all data are suitable for meta-analysis. We planned to calculate measures of effect, as relevant (hazard ratios, odds ratios, relative risks, risk differences, mean differences, and standardised mean differences). Where possible, we planned to calculate hazard ratios using methods described by Parmar and Tierney (Parmar 1998). We planned to extract information (for example, hazard rates, P values, events, ratios, curve data, and information on follow-up) from the publication and, if necessary, to enter data into a Microsoft Office Excel 2003 spreadsheet to calculate hazard ratios (Tierney 2007). Where data were available for the same outcomes using similar methods, meta-analyses were to be performed. If data could not be meta-analysed statistically, for example in the case of extreme heterogeneity, we planned to present results in a forest plot, without the estimate, in order to show the variance of the effects (Egger 1997). We planned to include cross-over trials using the results of the first period only (before cross-over), as if they were parallel trials.
In cases without heterogeneity and yet with meta-analysis not being possible, we planned to present the results in a narrative way, including text, tables, and figures to summarise the data and to allow the reader to judge the results based on the differences and similarities of the included trials and their risk of bias assessment. We planned to group the trials by intervention, patient characteristics, and outcomes and to describe the most important characteristics of the included trials, including a detailed review of the methodological shortcomings of a trial.
We planned to use funnel plots to identify any possible small trial biases, such as publication bias, if data were available (Egger 1997). We planned to discuss the possible implications of our findings if bias was present.
Where possible, we planned to examine apparent significant beneficial and harmful intervention effects with trial sequential analysis (CTU 2011; Thorlund 2011) in order to evaluate if these apparent effects could be caused by random error (‘play of chance’) (Brok 2008; Wetterslev 2008; Brok 2009; Thorlund 2009, Wetterslev 2009; Thorlund 2010).
We planned to create a 'Summary of findings' table including, where possible, survival, response, recurrence, and adverse events, using GRADEpro (http://ims.cochrane.org/revman/other-resources/gradepro).
Subgroup analysis and investigation of heterogeneity
We planned to perform subgroup analyses, where possible, based on prognostic indicators such as age, sex, tumour size, location of primary tumour, and use of any co-interventions.
For an extra subgroup analysis, a trial with a lower risk of bias was to be defined where three or more domain items were met, including sequence generation and allocation concealment.
We planned to summarise the separate outcomes after intervention, at six months or less, six to 12 months, and one year or more.