Reanalyses of RCTs Can Lead to Different Conclusions
Thirty-five percent of published reanalyses led to changes in findings that implied different conclusions
WEDNESDAY, Sept. 10, 2014 (HealthDay News) -- More than one-third of the small number of reanalyses of randomized clinical trials (RCTs) have implied conclusions different from those of the original articles, according to a study published online Sept. 9 in the Journal of the American Medical Association.
Shanil Ebrahim, Ph.D., from Stanford University in California, and colleagues characterized methodological and other differences between the original trial and reanalysis of RCT data. Thirty-seven reanalyses were identified in 36 published articles, of which five were performed by entirely independent authors.
The researchers found that the most frequent differences in reanalyses were in statistical or analytical approaches (18 reanalyses) and in definitions or measurements of outcomes of interest (12 reanalyses). The direction of effect was changed in four reanalyses, and the magnitude of treatment effect was altered in two; changes in statistical significance of the findings were noted in four reanalyses. Interpretations were different from that of the original article in 35 percent of reanalyses: 8 percent showing that different patients should be treated, 3 percent showing that fewer patients should be treated, and 24 percent showing that more patients should be treated.
"Authors of confirmatory reanalyses may choose not to publish the results or, alternatively, They may have difficulty publishing their article because many journals may not consider it interesting. Thus, our observed estimate of different conclusions (35 percent) is probably an overestimate," the authors write.