Types of studies

• We will consider only RCTs as primary studies in this systematic review and meta‐analysis.

• We will exclude quasi‐randomised trials, cross‐over trials, cohort trials and case‐controlled studies. We will include both full‐text and abstract publications.

Types of participants

• We will include acute myeloid leukaemia (AML) patients (both adults and children) who have achieved first complete remission (CR) and have not relapsed.

• We will include studies concerning all risk groups.

  • If trials consist of patients with different haematological malignancies, we will use data from the AML subgroup.

  • If subgroup data for AML patients are not provided, after contacting the author of the trial, we will exclude the trial if less than 80% of patients have AML.

• Furthermore, we will assess the definition of AML used by the investigators. We will evaluate effects of different definitions of AML (i.e. more than 20% blasts versus 30% or more blasts) by subgroup analyses.

Types of interventions

Any of the following comparisons will be allowed:

• IL‐2 maintenance monotherapy (intervention) versus placebo (control) or best supportive care (control) or maintenance chemotherapy (control); and

• IL‐2 plus maintenance chemotherapy (intervention) versus the same maintenance chemotherapy alone (control).

We will exclude studies in which patients received allogenic or autologous stem cell transplantation in the control group. Furthermore, we will assess different doses and time schedules of the IL‐2 regimen.

Types of outcome measures

Electronic searches

We have adapted search strategies from those suggested in the Cochrane Handbook for Systematic Reviews of Interventions (Lefebvre 2011). We will not impose any language restrictions in order to reduce the language bias.

We will search the following databases of medical literature.

1. Cochrane Central Register of Controlled Trials (CENTRAL), The Cochrane Library, latest issue (see Appendix 1 for search strategy).

2. MEDLINE (1950 to present, see Appendix 2 for search strategy).

3. EMBASE (1950 to present, see Appendix 3 for search strategy).

4. LILACS (1982 to present, see Appendix 4 for search strategy).

5. Chinese BioMedical Literature Database (CBM) (1978 to present, see Appendix 5 for search strategy).

Searching other resources

We will search the following conference proceedings from 2000 to the present, if they are not included in CENTRAL.

1. American Society of Hematology (ASH) (2000 to 2012).

2. American Society of Clinical Oncology (ASCO) (2000 to 2012).

3. European Hematology Association (EHA) (2000 to 2012).

4. European Society of Medical Oncology (ESMO).

In addition, we will search the following conference proceedings.

1. International Society for Hematology and Stem Cells (available at: http://www.exphem.org/search) (from 1999 to present).

We will electronically search the metaRegister of Controlled Trials (available at: http://www.controlled‐trials.com/mrct/) for ongoing trials (see Appendix 6 for search strategy) and we will save the search results of each item for further screening.

We will also check the citations of included trials and major reviews for additional studies.

Selection of studies

We will perform the selection of studies according to the guidelines in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011a). After the initial screening of all the titles and abstracts of the identified studies from the above sources, two review authors (CM and YH) will independently exclude all studies that are clearly ineligible. We will assess selected studies by using an eligibility form regarding study design and compliance with inclusion criteria. The forms will contain the following questions.

• Is the study described as randomised?

• Were the patients diagnosed with acute myeloid leukaemia?

• Had the patients achieved first CR and not relapsed?

• Were the patients in the intervention group treated with IL‐2?

In case of any doubt, we will include the full‐text analysis and review authors will discuss eligibility of the studies in order to finalise a decision (preferably including studies rather than losing relevant data). According to PRISMA we will use a flow diagram to show numbers of identified records, excluded articles and included studies (Moher 2009).

Data extraction and management

We will extract data from the included trials in accordance with Chapter 7 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011a) by using a standardised data extraction form. Two review authors (CM, YH) will conduct the data extraction independently. Any disagreement will be resolved by consensus. If the disagreement remains unresolved, a third review author (ZY) will independently extract the data. We will enter the extracted information onto the standardised data extraction form, which contains the following items.

• General information: author, title, source, publication date, country, language, and duplicate publications.

• Quality assessment: sequence generation, allocation concealment, blinding (participants personnel outcome assessors), incomplete outcome data, selective outcome reporting, and other sources of bias.

• Study characteristics: trial design, aims, setting and dates, source of participants, inclusion/exclusion criteria, comparability of groups, subgroup analysis, statistical methods, power calculations, treatment cross‐overs, compliance with assigned treatment, length of follow‐up, and time point of randomisation.

• Participant characteristics: age, gender, number of participants recruited/allocated/evaluated, and participants lost to follow‐up.

• Interventions: setting, dose and duration of IL‐2 treatment, type of additional or comparator chemotherapy, supportive treatment, type of treatment (first‐line treatment, pretreated patients), and additional treatment.

• Outcomes: disease‐feee survival, overall survival, event‐free survival, treatment‐related mortality, adverse events, and quality of life..

Finally, we will enter all the extracted information into the latest Cochrane Review Manager software (RevMan 2011).

Assessment of risk of bias in included studies

Two review authors (CM, ZY) will review the included studies independently according to the recommendations in Chapter 8 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011b) for assessing bias for the following criteria.

• Sequence generation

• Allocation concealment

• Blinding (participants, personnel, outcome assessors)

• Incomplete outcome data

• Selective outcome reporting

• Other sources of bias

We will resolve any disagreement by discussion. If disagreement persists, a third review author (YH) will extract the data independently. When necessary, we will contact the authors of the studies for clarification. If this is unsuccessful, we will report the disagreements.

Measures of treatment effect

For the time‐to‐event data including disease‐free survival, event‐free survival and overall survival, we plan to extract hazard ratios (HRs) with standard errors (SEs). If hazard ratios are not reported, we will use indirect estimating methods as described in Tierney 2007. In reporting results, we will adopt the HR with 95% confidence intervals (CIs) as the measure of treatment effect for each trial.

For binary outcomes, such as relapse rate, treatment‐related mortality and adverse events, we will extract or calculate events plus the total number of participants in each arm of every study. When reporting results, we will calculate the risk ratio (RR) with 95% CIs as measures of treatment effect for each trial.

Unit of analysis issues

We have conducted a pilot search and found that published studies eligible for this review were usually individually randomised, non‐cross‐over trials, without multiple intervention groups. In addition, quasi‐randomised trials, crossover trials and cohort trials will not be included in the review. Therefore, the unit‐of‐analysis issues related to cluster‐randomised trials, cross‐over trials and multiple intervention groups seem unlikely to arise. We will try to avoid any potential unit‐of‐analysis error by extracting and analysing the data carefully according to the methods recommended in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011c).

Dealing with missing data

As suggested in Chapter 16 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011d), there are many potential sources of missing data: at study level; at outcome level; at summary data level; at individual level; and at study‐characteristics level (e.g. for subgroup analysis). We will seek missing data from the study authors to obtain the information which is not reported in the publications.

Moreover, it is important to make assumptions on the missing data between "missing at random" and/or "not missing at random". The assumption will be based on the report of the reasons for missing data and the distribution between allocated groups. If the data are assumed to be missing at random, we will only analyze the available data (i.e. ignoring the missing data). If the data are assumed not to be missing at random, we will input the missing data with replacement values, and treat these as if they were observed (e.g. last observation carried forward, imputing an assumed outcome such as assuming all were poor outcomes, imputing the mean, and imputing based on predicted values from a regression analysis). We will perform sensitivity analysis of the primary outcome on these assumptions to detect the impact of assumption modifications on the results.

Assessment of heterogeneity

We will examine heterogeneity of treatment effects between trials qualitatively with the Chi² test with a significance level at P < 0.1. We will use the I² statistic of inconsistency to quantify possible heterogeneity (Deeks 2011). We will use the following rough guide to interpret heterogeneity.

• 0% to 40%: might not be important;

• 30% to 60%: may represent moderate heterogeneity;

• 50% to 90%: may represent substantial heterogeneity; and

• 75% to 100%: considerable heterogeneity.

We will explore potential sources of heterogeneity through subgroup stratification by age, treatment used before the maintenance randomisation (i.e. whether allogeneic or auto transplant or chemotherapy alone have been used), allocation concealment, blinding and size of studies. We will use the tests for interaction to test for differences between subgroup results.

Assessment of reporting biases

If at least 10 studies are included, we will inspect the funnel plot of the treatment effect against the precision of trials (plots of the log of the RR for efficacy against the SE) and estimate potential asymmetry that may indicate selection bias (the selective publication of trials with positive findings) or methodological flaws in the small studies. We will statistically test by means of a linear regression test. We will consider P < 0.1 as significant for this test. If there are less than 10 studies, we will not adopt funnel plots because the power of the test is too low to distinguish chance from real asymmetry (Sterne 2011).

Data synthesis

We will perform analyses according to the recommendations in Chapter 9 of the Cochrane Handbook for Systematic Reviews of Interventions (Deeks 2011). We will enter the data into Review Manager (RevMan 2011). A second review author will check it for accuracy. For dichotomous data, we will estimate RRs and their CIs using the Mantel‐Haenszel method for the primary meta‐analysis, and repeat the statistical analysis, adopting a random‐effects model using the DerSimonian Laird method in terms of sensitivity analysis (Higgins 2011c). For time‐to‐event data, we will pool the log HR for time‐to‐event outcomes using an inverse variance method. If the HR and its SE or CI are not reported, we will estimate 'O‐E' and 'V' statistics indirectly using the methods described by Parmar 1998 and Tierney 2007. HRs or RRs less than 1.0 are in favour of maintenance treatment.

Subgroup analysis and investigation of heterogeneity

We will assess the heterogeneity of treatment effects by a Chi² test with a significance level at P < 0.1 and we will quantify it with the I² statistic (Deeks 2001). We will explore potential sources of heterogeneity through stratifying the patient subgroups according to:

• age (younger adults age < 60 years; older adults age ≥ 60 years);

• paediatric patients (age < 15 years) versus adult patients (age ≥ 15 years);

• different types of definitions of AML (i.e. > 20% blasts versus >= 30% blasts);

• treatment used before the maintenance randomisation (allogeneic stem cell transplants versus autologous stem cell transplants versus chemotherapy alone);

• Patients receiving interleukin‐2 therapy before the maintenance randomisation versus those not receiving interleukin‐2;

• different types of intervention (whether patients in the intervention group received other maintenance chemotherapy or not); and

• doses of IL‐2 (as reported in the primary literature).

Sensitivity analysis

We will perform sensitivity analyses based on:

• quality components, including full‐text publications versus abstracts, preliminary results versus mature results, and published versus unpublished data;

• random‐effects model; and

• risk of bias (by omitting studies judged to be of high risk of bias).