Missing data reduce the power and precision of the results of a randomised controlled trial (RCT). Sample size inflation will avoid this problem, but will not avoid the threat to internal validity introduced by differential rates and reasons for missing data. In palliative care trials, data missing because of death and disease progression are expected, but have not been quantified. This study assessed the risk that missing data pose to the power and validity of trials testing palliative interventions in patients with terminal disease. We systematically searched CENTRAL, Medline, and EMBASE with no language restrictions for RCTs published between Jan 1, 2009, and April 30, 2014, of palliative interventions in participants with life-limiting disease. Random-effects meta-analysis was performed. The primary outcomes were proportion of missing data at the primary endpoint of the trial, covariates associated with missing data (meta-regression analysis), and differential rates and reasons for missing data between the intervention and control arms. 108 RCTs representing 15 560 patients were included (mean age 64 years [SD 8·4], ECOG performance status 2). The weighted estimate for missing data at the primary endpoint was 23·1% (95% CI 19·3–27·4). Larger proportions of missing data were associated with increasing numbers of questions asked or tests requested (odds ratio [OR] per doubling of questions or tests requested 1·19, 95% CI 1·05–1·35) and with longer study duration (OR per days doubling 1·09, 1·02–1·17). Meta-analysis showed evidence of differential rates of missing data between trial arms, which varied in direction (OR 1·04 [95% CI 0·90–1·20], I2=35·9; p=0·001). Despite randomisation, missing data in the intervention arms were more likely than in the control arms to be attributed to disease-progression unrelated to the intervention (OR 1·31, 95% CI 1·02–1·69), but not for data missing because of death (0·92, 95% CI 0·78–1·08). This review of international RCTs testing a range of palliative interventions found that the overall weighted proportion of missing data is at a level that poses a substantial risk to the validity of trial results. Trial burden and duration need consideration when adjusting sample-size calculations for missing data. Differential reasons and rates of missing data at trial-level also present a risk of bias. Our review only included published trials, so it probably represents an overoptimistic picture. National Institute for Health Research.