Types of studies

Randomised controlled trials (RCTs) with a parallel, single‐stage or cross‐over design, including studies with a quasi‐randomised allocation in cases where allocation concealment was described.

Types of participants

Study participants were adults (aged 18 years or older) with a histological or clinical diagnosis of cancer, meeting international criteria for cancer cachexia of any stage (Fearon 2011), which included:

• precachexia, defined as weight loss of 5% or less with anorexia and metabolic changes;

• cachexia, defined as weight loss greater than 5% in the past six months or body mass index (BMI) less than 20 kg/m² and ongoing weight loss greater than 2% or sarcopenia, anorexia or systemic inflammation; and

• refractory cachexia, defined as active catabolism, ongoing weight loss, unresponsive to treatment and life expectancy of less than three months.

We contacted study authors to seek additional data where baseline demographic data were insufficient to assess participants against these criteria.

We considered studies that were relevant to the review objectives but which were not performed specifically to address cancer cachexia (e.g. studies in people with advanced cancer). As such, we included studies in which at least half of the study population fell within the cachexia definitions above. Participants could be studied in any hospital or community setting. We excluded studies relating to participants during or following treatment with curative intent, or with no evidence of current disease.

Types of interventions

Studies examining any programme of exercise offered as a sole intervention or in combination with another intervention were eligible. We considered programmes using aerobic/endurance training, resistance training or a combination of both. We expected programmes to vary in terms of session length (minutes) and frequency (sessions/week), intensity of training (low, moderate, high) and overall duration (weeks). There were no restrictions on these or other programme characteristics including the setting in which the programme was offered (hospital/centre/home) and level of supervision (none, minimal, close). Interventions could be compared to either usual care, no treatment or an active control group (e.g. a nutritional or drug intervention). We used the definition of exercise by the American College of Sports Medicine as a "planned, structured and repetitive bodily movement done to maintain or improve one or more components of physical fitness" (Thompson 2010).

Types of outcome measures

We considered only outcome measures from the primary studies that were validated and used count/rate of events.

Electronic searches

We developed an electronic search strategy using a combination of terms based on the target population, intervention, comparator and outcomes. See Appendix 1 for the search strategies. We adapted these where necessary for the other databases listed below. We searched the following electronic databases from their start date until March 2020:

• CENTRAL (the Cochrane Library) 2020, Issue 3;

• MEDLINE (Ovid) 1946 to March 2020;

• Embase (Ovid) 1974 to March 2020;

• DARE (the Cochrane Library) 2015, Issue 2. It was not searched after 2015, since it is no longer updated/available;

• HTA – Health Technology Assessments (the Cochrane Library) 2016, Issue 4;

• ISI Web of Science (SCI‐Expanded and CPCI) 1900 to March 2020;

• LILACS (Latin American and Caribbean Health Sciences) (BIREME) 1985 to March 2020;

• PEDro (the Physiotherapy Evidence Database) 24 March 2020;

• SciVerse SCOPUS 24 March 2020;

• Biosis Previews PreMEDLINE 1969 to March 2020;

• Open Grey (System for Information on Grey Literature) 24 March 2020.

Searching other resources

We identified ongoing studies using:

• ClinicalTrials.gov (clinicaltrials.gov/);

• International Clinical Trials Registry Platform (ICTRP) (www.who.int/ictrp/en/);

• MetaRegister of Controlled Trials (mRCT) (www.controlled-trials.com/) updated to ISRCTN registry (www.isrctn.com/);

• Pan African Clinical Trials (www.pactr.org); and

• the EU Clinical Trials Register (www.clinicaltrialsregister.eu/).

We handsearched the following sources: the Society on Sarcopenia, Cachexia and Wasting Disorders (SCWD); the American Cancer Society; the British Association for Cancer Research (BACR) and the European Clinical Guidelines. We checked reference lists of relevant studies and reports citing all retrieved studies. In addition, we contacted corresponding authors of retrieved studies, experts and organisations in the field to seek potentially relevant research material, including unpublished and ongoing studies.

Selection of studies

We used reference management software to merge results from different electronic databases and remove duplicate studies. Two review authors (AJG, VS) independently assessed titles and abstracts of articles for relevance (Higgins 2011). We obtained full‐text reports of potentially relevant studies for assessment against the inclusion criteria. We contacted the study authors by email to clarify the necessary information if missing information impaired the study selection. The two review authors discussed any disagreement in the selection of studies and resolved it by consensus. In cases of persistent disagreement, they consulted a third review author (MSP or MM). We applied no language restrictions in the selection of studies.

Data extraction and management

Two review authors (AJG and VS) independently extracted data from the included studies. We developed an online extraction form to store data relating to the study source and eligibility, methods and bias (study design, sequence generation, allocation sequence concealment, blinding), participants (number, age, sex, ethnicity, diagnosis, disease severity, setting) and intervention (exercise type and intensity, session length and frequency, and overall programme duration), adherence to the exercise programme (either self‐reported or objective) and the occurrence of any adverse events. We discussed and resolved any disagreements by consensus.

Outcome data collected at baseline immediately following a programme of exercise and at first follow‐up were:

• lean body mass, generally assessed by anthropometry (e.g. skin fold thickness, or imaging, e.g. dual x‐ray absorptiometry and expressed as a weight (e.g. kilograms, kg), cross‐sectional area (square centimetres, cm²) or volume (cubic centimetres, cm³) normalised to height);

• muscle strength, either isometric or isotonic, generally assessed using myometry and expressed as a measure of force (e.g. kilograms, kg, or Newton metres, Nm); muscle endurance, generally assessed as time or number of repetitions to a specified decline in muscle performance;

• maximal and submaximal exercise capacity, generally assessed by a walking or cycling test and expressed as a measure of oxygen uptake (VO₂) or performance (e.g. distance walked in metres, m);

• fatigue, generally assessed on a numerical or categorical scale with a higher score representing more severe fatigue;

• health‐related quality of life, generally assessed on a numerical or categorical scale with a higher score representing a better quality of life.

Assessment of risk of bias in included studies

Two review authors (AJG and VS) independently assessed risk of bias for each study using the criteria outlined in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2017), with any disagreements resolved by discussion. We completed a 'Risk of bias' table and summary using the 'Risk of bias' tool in Review Manager 5 (Review Manager 2014).

We assessed the following for each study.

• Random sequence generation (checking for possible selection bias): we assessed the method used in the studies to generate the randomised sequence process. We assessed the methods as: low risk of bias (e.g. random number table, using a computer random number generator, coin tossing, shuffling cards or envelopes, throwing dice, drawing of lots); unclear risk of bias (method not clearly described); high risk of bias (e.g. sequence generated by odd or even date of birth, date (or day) of admission, hospital or clinic record number; or allocation by judgement of the clinician, preference of the participant, results of a laboratory test or a series of tests or availability of the intervention).

• Allocation concealment (checking for possible selection bias): we assessed the method used to conceal allocation to interventions prior to assignment determines whether intervention allocation could have been foreseen in advance of, or during recruitment, or changed after assignment. We assessed the methods as: low risk of bias (e.g. telephone or central randomisation; consecutively numbered, sealed, opaque envelopes); unclear risk of bias (method not clearly stated); high risk of bias (studies that did not conceal allocation (e.g. open list)).

• Blinding of participants and personnel (checking for possible performance bias): we assessed the methods used to blind trial participants and personnel from knowledge of which intervention they received. We assessed the methods as: low risk of bias (trial stated that personnel were blinded to participant's condition and intervention); unclear risk of bias (trial did not mention it or provide an adequate description of how it was achieved); high risk of bias (trial stated that participants or personnel were not blinded).

• Blinding of outcome assessment (checking for possible detection bias): we assessed the methods used to blind outcome assessors from knowledge of which intervention a participant received. We assessed the methods as: low risk of bias (study stated that it was blinded and described the method used to achieve blinding, e.g. outcome assessor was someone not involved in the study); unclear risk of bias (trial stated that it was blinded but did not provide an adequate description of how it was achieved); high risk of bias (e.g. the study mentioned that it was not blinded).

• Incomplete outcome data (checking for possible attrition bias due to the amount, nature and handling of incomplete outcome data): we assessed the methods used to deal with incomplete data as: low risk (less than 10% of participants did not complete the study or missing data imputed using appropriate methods, or both); unclear risk of bias (used 'last observation carried forward for analysis without detailed time point' or lacking information to judge it); high risk of bias (used 'completer' analysis).

• Selective reporting (checking for reporting bias): we assessed the methods used to report the outcomes of the study as: low risk of bias (if all planned outcomes in the study protocol or methods were reported in the results of the published study); unclear risk of bias (if there was no clear distinction between planned outcomes and reported outcomes); or high risk of bias (if some planned outcomes from the study protocol or methods were clearly not reported in the results).

• Other bias (checking for possible bias not explored in the categories above): we assessed the methods as low risk of bias (e.g. if the study appeared free of other sources of bias (e.g. study funding described, balanced groups for characteristics, and carryover effect or blocking described in the study); unclear risk of bias (e.g. lack of information regarding participants, study funding not described, and carryover effect or blocking not clearly reported); high risk of bias (unbalanced groups and carryover effect or blocking not being conducted).

Measures of treatment effect

We considered the review objectives and extracted relevant counts, dichotomous, categorical and continuous data from included studies. We extracted dichotomous data and calculated risk ratio (RR) with 95% confidence intervals (CI) for adherence to prescribed programmes and occurrence of adverse events. We extracted continuous data and calculated mean difference (MD) with 95% CI for the outcomes of lean body mass, fatigue and health‐related quality of life, muscle strength and muscle endurance, maximal and submaximal exercise capacity (Deeks 2011). If studies had used different scales, we would have expressed the treatment effect as standardised mean difference (SMD) and 95% CI.

Unit of analysis issues

In the parallel‐group RCTs, we considered the individual participant as the unit of analysis. If we had included cross‐over RCTs, we would have analysed both periods, separated by periods and together. If we had included cluster‐RCTs, we would have considered each cluster group as the unit of analysis.

Dealing with missing data

We contacted authors of included studies via email to request further information when the published study report did not provide sufficient information (e.g. did not describe randomisation or intention‐to‐treat analysis, or had missing data). If no answer had been obtained from study authors, we planned to present the findings in the main discussion.

Assessment of heterogeneity

We explored clinical, methodological and statistical heterogeneity. We conducted an initial assessment of clinical heterogeneity (participants, intervention and outcome), as well as methodological heterogeneity (study design). Based on this analysis and the decision to include exercise as part of multicomponent interventions, we chose a random‐effects model to better estimate effects (Deeks 2011; Higgins 2002; Higgins 2003). We planned visual inspection of any forest plots and testing for statistical heterogeneity for any meta‐analysis to inform the interpretation of the findings.

Assessment of reporting biases

We planned to use a funnel plot and examine for asymmetry to assess for evidence of reporting bias if there were 10 or more included studies.

Data synthesis

We planned to meta‐analyse data from similar comparisons using a random‐effect model in Review Manager 5 (Review Manager 2014). Due to the limited available data, we used Review Manager 5 to report individual studies with different time points, random effects and without group totals.

Subgroup analysis and investigation of heterogeneity

We planned to use descriptive comparisons to consider differences in effect between subgroups according to the following participant characteristics: type of cancer (lung, pancreatic, etc.); stage of cachexia (precachexia, cachexia, refractory cachexia (Fearon 2011)); and intervention characteristics: type of exercise (aerobic, resistance, combined), intensity of exercise (low, moderate, high), duration of exercise programme (six weeks, 12 weeks, etc.), where there were sufficient data.

Sensitivity analysis

We planned to perform a sensitivity analysis to consider the difference in pooled effects when studies were at high or unclear risk of bias, or those with substantial (greater than 20%) missing data, were omitted from analyses, where there were sufficient data. However, we did not perform such analysis due to the number of included studies, which did not allow us to explore the results any further.

Summary of findings and assessment of the certainty of the evidence

Two review authors (AJG, VS) independently rated the certainty of the evidence for each outcome using the GRADE system and GRADEprofiler Guideline Development Tool software (GRADEpro GDT), and the guidelines provided in Chapter 14 of the CochraneHandbook for Systematic Reviews of Interventions (Schünemann 2019).

The GRADE approach uses five considerations (study limitations, consistency of effect, imprecision, indirectness and publication bias) to assess the certainty of the body of evidence for each outcome, using the following criteria (Guyatt 2008):

• high: we are very confident that the true effect lies close to that of the estimate of the effect;

• moderate: we are moderately confident in the effect estimate; the true effect is likely to be close to the estimate of effect, but there is a possibility that it is substantially different;

• low: our confidence in the effect estimate is limited; the true effect may be substantially different from the estimate of the effect;

• very low: we have very little confidence in the effect estimate; the true effect is likely to be substantially different from the estimate of effect.

The GRADE system uses the following criteria for assigning a certainty level to a body of evidence (Chapter 14 of the CochraneHandbook for Systematic Reviews of Interventions; Schünemann 2019).

• High: RCTs or double‐upgraded observational studies.

• Moderate: downgraded RCTs or upgraded observational studies.

• Low: double‐downgraded RCTs or observational studies.

• Very low: triple‐downgraded RCTs or downgraded observational studies; or case series/case reports.

Factors may decrease the certainty level of a body of evidence such as downgrading the evidence by one (–1) or two (–2) levels for the five considerations:

• serious (–1) or very serious (–2) limitations in the design and implementation of available studies suggesting high likelihood of bias;

• serious (–1) or very serious (–2) indirectness of evidence (indirect population, intervention, control, outcomes);

• serious (–1) or very serious (–2) unexplained heterogeneity or inconsistency of results (including problems with subgroup analyses);

• serious (–1) or very serious (–2) imprecision of results (wide CIs);

• serious (–1) or very serious (–2) high probability of publication bias.

We included two 'Summary of findings' tables to present the main findings in a transparent and simple tabular format, for the comparisons:

• exercise plus usual care compared with usual care;

• multimodal intervention with exercise plus usual care compared with usual care.

In particular, we included key information concerning the certainty of the evidence, the magnitude of effect of the interventions examined and the sum of available data on the outcomes lean body mass (eight weeks for comparison 1 and up to 12 weeks for comparison 2); adherence to prescribed programmes (eight weeks for comparison 1 and no data for comparison 2), occurrence of adverse events (eight weeks for comparison 1 and six weeks for comparison 2); muscle strength and endurance (no data for comparison 1 and six ‐ 12 weeks for comparison 2); maximal and submaximal exercise capacity (no data for comparison 1 and six ‐ 12 weeks for comparison 2); fatigue (eight weeks for comparison 1 and six weeks for comparison 2); and health‐related quality of life (eight weeks for comparison 1 and 12 weeks for comparison 2).

In circumstances where there were no data for an outcome, the certainty of the evidence was unknown and we reported this as 'no evidence to support or refute'.