STATISTICAL INACCURACIES IN
PEER-REVIEWED ARTICLES
Shelley Johnson and Adel Mikhail
Medical Incorporated, 9605 West Jefferson Trail, Inner Grove Heights, MN 55077, USA
Objective: Clinical studies are published with the assumption that the statistics are valid. However, careful dissection of articles from peer-reviewed journals frequently expose simple statistical errors.
Design: Review of MEDLINE identified 27 articles published from 1989 through 1993 reporting clinical follow-up of patients with prosthetic heart valves. Data dealing with morbidity and mortality were extracted from each report. This study focused on accuracy of the calculation of statistics.
Results: Errors discovered were separated into 3 categories: addition/subtraction errors, incorrect calculation of simple percentages, and mistakes in calculating linearized rates. The majority of articles (16/27, 59%) contained no errors. However, 22% of the articles (6/27) were found to have errors in addition/subtraction, and simple percentages were inaccurate in an equal number of articles. Miscalculation of linearized rates were detected in 15% (4/27) of the articles. Five of the publications had errors in 2 categories.
Conclusions: Review of prosthetic heart valve articles published during a 5-year period frequently demonstrated statistics that are inconsistent with the data reported. Errors are widespread, and the current peer review system apparently does not detect them. Ironically, the errors are usually simple mistakes that could be eliminated through careful final review before publication. Unfortunately, when easily calculated rates are incorrectly reported, doubt is cast on the accuracy of the entire clinical investigation.
Return to Session Information
