Presentation of continuous outcomes in randomised trials: an observational study

OBJECTIVE To characterise the percentage of available outcome data being presented in reports of randomised clinical trials with continuous outcome measures, thereby determining the potential for incomplete reporting bias. DESIGN Descriptive cross sectional study. DATA SOURCES A random sample of 200 randomised trials from issues of 20 medical journals in a variety of specialties during 2007-09. MAIN OUTCOME MEASURES For each paper's best reported primary outcome, we calculated the fraction of data reported using explicit scoring rules. For example, a two arm trial with 100 patients per limb that reported 2 sample sizes, 2 means, and 2 standard deviations reported 6/200 data elements (1.5%), but if that paper included a scatterplot with 200 points it would score 200/200 (100%). We also assessed compliance with 2001 CONSORT items about the reporting of results. RESULTS The median percentage of data reported for the best reported continuous outcome was 9% (interquartile range 3-26%) but only 3.5% (3-7%) when we adjusted studies to 100 patients per arm to control for varying study size; 17% of articles showed 100% of the data. Tables were the predominant means of presenting the most data (59% of articles), but papers that used figures reported a higher proportion of data. There was substantial heterogeneity among journals with respect to our primary outcome and CONSORT compliance. LIMITATIONS We studied continuous outcomes of randomised trials in higher impact journals. Results may not apply to categorical outcomes, other study designs, or other journals. CONCLUSIONS Trialists present only a small fraction of available data. This paucity of data may increase the potential for incomplete reporting bias, a failure to present all relevant information about a study's findings.