Burak Bahar1, Stefan E Pambuccian2, Güliz A Barkan2, Yasemin Akdas3. 1. Department of Laboratory Medicine, Yale University School of Medicine, New Haven, CT. 2. Department of Pathology, Loyola University Medical Center, Maywood, IL. 3. College of Liberal Arts and Sciences, University of Florida, Gainesville.
Abstract
BACKGROUND: The misuse of statistical methods in diagnostic accuracy studies has been criticized in many publications. OBJECTIVE: To assess the use and misuse of statistical methods in medical journal articles. METHODS: We reviewed 141 original articles from 6 cytopathology journals published in 2014. RESULTS: In total, 16 articles used no descriptive statistics and 66 articles used no inferential statistics. Also, 82 articles did not report using any tests for diagnostic accuracy. The most commonly adopted statistical measure for diagnostic accuracy was sensitivity, followed by specificity. The most common inferential statistical methods were χ2 testing, the Fisher exact test, and the Cohen kappa coefficient, respectively. Only 78 articles were free of statistical errors. "No statistics were used although statistical methods were required" was the most common type of error, followed by "failure to select proper statistical methods." CONCLUSION: Our assessment of use of statistical methods in this subsection of pathology demonstrated a need for improvement.
BACKGROUND: The misuse of statistical methods in diagnostic accuracy studies has been criticized in many publications. OBJECTIVE: To assess the use and misuse of statistical methods in medical journal articles. METHODS: We reviewed 141 original articles from 6 cytopathology journals published in 2014. RESULTS: In total, 16 articles used no descriptive statistics and 66 articles used no inferential statistics. Also, 82 articles did not report using any tests for diagnostic accuracy. The most commonly adopted statistical measure for diagnostic accuracy was sensitivity, followed by specificity. The most common inferential statistical methods were χ2 testing, the Fisher exact test, and the Cohen kappa coefficient, respectively. Only 78 articles were free of statistical errors. "No statistics were used although statistical methods were required" was the most common type of error, followed by "failure to select proper statistical methods." CONCLUSION: Our assessment of use of statistical methods in this subsection of pathology demonstrated a need for improvement.