Types of studies

We included studies if they were RCTs of interventions. Quasi‐randomised trials and crossover trials were excluded. No restrictions on language or publication status were applied.

Types of participants

Studies involving adults aged 18 years and older were included regardless of gender, stage of disease, tumour type, and type of treatment. Participants could have been receiving curative or palliative treatment or long‐term follow‐up, or could have had no evidence of active disease.

Types of interventions

To be included, studies must have stated that they aimed to evaluate the effect of educational interventions designed specifically to manage cancer‐related fatigue, or educational interventions targeting a constellation of physical symptoms or quality of life where fatigue was the primary focus. For the purpose of this review, educational interventions were defined as any advice, information, or self‐management education (verbal, written, or audiovisual) provided in order to help people understand and manage cancer‐related fatigue. These may have incorporated information and advice about non‐pharmacological strategies (e.g. information about relaxation, nutrition, CBT, or exercise), but cannot have actually used these interventions. Studies may have used techniques such as discussion, coaching, goal‐setting, feedback, and reinforcement that may also be used in psychological therapies such as CBT, but the intervention would not be classified as CBT itself. Interventions did not need to be delivered face to face but may have included interventions delivered via telephone, post, or the Internet. The intervention may have taken place in any setting, been delivered either to a group or an individual, involved a single education session or a series of sessions or delivered in either groups or with individuals. The studies could have compared educational interventions with no intervention or wait list controls, attention controls, or an alternative intervention for cancer‐related fatigue.

Types of outcome measures

Included studies must have considered fatigue or its management as the primary outcome of interest; this included fatigue if it was measured as a main outcome within a constellation of physical symptoms or quality of life, provided separate data for fatigue were available. Outcome measures of fatigue or its management were via self‐report, as fatigue is a subjectively experienced symptom. Self‐report of fatigue or its management was measured through questionnaires or diaries.

Electronic searches

We searched:

• The Cochrane Central Register of Controlled Trials (CENTRAL) via CRSO on 1/11/16;

• MEDLINE (OVID; 1950 to October week 3 2016);

• Embase (OVID; 1966 to October 2016 week 4);

• CINAHL (EBSCO; 1982 to October 2016);

• PsycINFO (OVID4; 1840 to October 2016);

• ERIC (1966 to October 2016);

• OTseeker (to October 2016);

• PEDro (to October 2016).

We developed the search strategy for MEDLINE using the Cochrane filter for the identification of RCTs, as published in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2008), and added relevant terms for this topic. The search strategies used can be found in Appendix 1, Appendix 2, Appendix 3, Appendix 4, Appendix 5, Appendix 6, Appendix 7 and Appendix 8. Non‐English language studies were considered if they had an abstract published in English.

Searching other resources

In an effort to identify further published, unpublished, and ongoing trials we:

• checked reference lists of all relevant studies;

• searched ongoing trials and research registers including ClinicalTrials.gov (www.clinicaltrials.gov/; accessed October 2016) and the Australian New Zealand Clinical Trials Registry (accessed October 2016);

• contacted investigators known to be involved in research in this area;

• handsearched nine relevant journals.

Selection of studies

One review author (CD) initially screened titles and abstracts and eliminated those obviously not relevant to this review. Two review authors (SB and PM) independently screened the remaining titles and abstracts for their eligibility for inclusion in accordance with the above defined criteria. When the title and abstract did not provide all the information necessary to assess relevance, we retrieved full paper copies for screening. We retrieved full text copies of studies when either review author determined that the study possibly or definitely met the inclusion criteria (Figure 1).

Figure 1

---

---

Study flow diagram.

For a trial to be included it must have contained fatigue as a primary outcome measure and one treatment arm must have been an educational intervention with the primary aim being management of fatigue.

We included a study if all of the following were met.

1. It was an RCT.

2. The primary aim of the study was to evaluate the effect of educational interventions to manage cancer‐related fatigue.

3. Participants were 18 years of age or older.

4. Participants were diagnosed with cancer.

5. At least one of the study arms received an educational intervention designed specifically to manage cancer‐related fatigue, or an educational intervention targeting a constellation of physical symptoms or quality of life where fatigue was the primary focus.

6. The primary outcome of interest included the measurement of fatigue or its management. Measurement could be as a separate measure of fatigue, or as part of a quality of life measure providing separate data for fatigue were available (e.g. as a sub scale).

We contacted study authors where information was unclear. We resolved disagreement about the selection of a study by consensus.

Data extraction and management

Two review authors (SB and TH) independently extracted data from the studies using a standard data extraction form. A third independent person not associated with any of the included trials checked the data extraction. We contacted authors in order to obtain any missing data. We collected the following data.

Assessment of risk of bias in included studies

Two review authors (SB and CD) independently assessed the risk of bias of the selected studies (Figure 2, Figure 3), using the Cochrane 'Risk of bias' tool in accordance with methods recommended in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). Ascertaining risk of bias involved considering the following seven domains.

Figure 2

---

---

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Figure 3

---

---

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

• Random sequence generation.

• Allocation sequence concealment.

• Blinding of outcome assessment.

• Incomplete outcome data.

• Selective reporting.

• Study size.

• Other bias.

Blinding of participants and personnel was not included as a domain in the 'Risk of bias' assessment because it is not achievable for any studies of educational interventions due to the nature of the intervention.

We assessed the size of the study in line with the Cochrane Pain, Palliative and Supportive Care (PaPaS) Group's policy: low risk of bias for studies with 200 participants or more per treatment arm; unclear risk of bias for studies with 50 to 199 participants per treatment arm; or high risk of bias for studies with fewer than 50 participants per treatment arm.

We assessed each study as being at 'low risk of bias', 'high risk of bias', or 'unclear risk of bias' for each of the 'Risk of bias' items, based on the study reports and/or additional information provided by the study authors. We conducted these assessments in accordance with methods recommended in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). When we could not determine from the report if the criteria had or had not been met, we indicated an 'unclear risk of bias'.

Measures of treatment effect

We used Cochrane's Review Manager software, RevMan 5 (RevMan 2014), for all analyses. For continuous data, we used unadjusted post‐scores (outcomes that were recorded immediately after the end of the intervention period). We did not analyse data from follow‐up time points. We did not calculate change over time scores for the purposes of meta‐analyses in this review, but we reported change scores provided by study authors in the narrative description of results if post‐score data were not available. We calculated the mean difference (MD) and the 95% confidence interval (CI) for continuous data. For continuous outcomes where no standard deviations (SD) were reported, we planned to calculate the SD using the methods described in the Cochrane Handbook for Systematic Reviews of Interventions. We calculated the standardised mean difference (SMD) and the 95% CI for continuous data that measured the same outcome using different measurement tools as per the plan for data synthesis described below. We planned to calculate the relative risks (RRs) and 95% CI and the number needed to treat (NNT) for dichotomous data.

Unit of analysis issues

Unit of analysis was the participant.

Dealing with missing data

We analysed data for all participants in the group to which they were allocated, regardless of whether or not they received the allocated intervention. If, in the original reports, participants were not analysed in the group to which they were randomised, we had planned to attempt to restore them to the correct group using data from the report or from the authors. However this was not possible as relevant data were not available.

Assessment of heterogeneity

We first assessed the studies for clinical homogeneity with respect to the population, intervention, and outcomes. We planned to describe studies that we judged to be too clinically heterogeneous separately and not combine them in a meta‐analysis. We tested studies without substantial clinical heterogeneity for statistical heterogeneity using the I² statistic (I² greater than 50% was considered substantial heterogeneity). If we suspected heterogeneity, we explored possible causes using subgroup analyses (where sufficient studies were available) and used the random‐effects model.

Assessment of reporting biases

Where we suspected reporting bias, we attempted to contact study authors to ask them to provide missing outcome data. Where this was not possible, and we suspected that the missing data introduced serious bias, we did not include the data for these outcomes in the meta‐analysis. 

Data synthesis

We pooled clinically and statistically homogeneous studies using the fixed‐effect model, and clinically homogeneous and statistically heterogeneous studies using the random‐effects model. We combined continuous data only where (i) means and SDs were available or calculable and (ii) there was no clear evidence of skew in the distribution (using methods described in the Cochrane Handbook for Systematic Reviews of Interventions). When able to be combined, we used SMDs of scales measuring the same clinical outcomes in different ways in order to combine results across scales; otherwise we used MDs.

Subgroup analysis and investigation of heterogeneity

Where sufficient data were available, we undertook subgroup analysis based on type of cancer treatment, timing of intervention relative to completion of cancer treatment (e.g. intervention during cancer treatment or intervention delivered following completion of cancer treatment), tumour type, stage of disease, and group versus individual intervention. We completed subgroup analysis using the Deeks method (Deeks 2001). We restricted analyses to effects on the primary outcome (fatigue).

Sensitivity analysis

We planned to carry out sensitivity analysis, where appropriate, to explore the effects of risk of bias. We had planned to exclude studies at high risk of bias for concealment of allocation from the analysis in order to assess any substantive change in the overall result. We chose to focus on concealed allocation because, according to Pildal 2007 (p 854) "most conclusions favouring an intervention would lose support if trials with unclear or inadequate allocation concealment were excluded from the meta‐analysis." If no substantive difference existed, we left the studies in for the main analysis. We planned to conduct this sensitivity analysis for the primary outcome only.