Catherine Klersy, Angela
Pistorio, Moreno Curti, Carmine Tinelli, Donatella De Amici, Gabriella Gabutti*, Andrea
Zeccato*.
Scientific Direction, Biometry and *Scientific Library, IRCCS Policlinico San
Matteo, Pavia, Italy
|
|
The use of quantitative methods for biomedical
research is mandatory in order to generalize results obtained in the study sample of the
general population. Moreover statistical tools, nowadays largely availale on personal
computers, can lead to misleading conclusions when applied inappropriately or when the
design itself is not appropriate. The arrhythmologic literature being particularly rich of
original articles, which can have a large impact on the way of treating arrhythmic
patients, a study has been performed to determine the design and the statistical methods
used as well as the most common types of statistical errors encountered. At that purpose a
Medline search has been performed on 4 of the major cardiologic journals for year 1996: A)
Circulation; B) J Am Coll Cardiol; C) Am J Cardiol; D) Eur Heart J, with
"tachycardia" as a keyword. The search strategy has gathered 29 articles on
tachycardia in journal A, 36 in journal B, 59 in journal C and 8 in journal D. Among
these, 104/132 are original articles and are considered for the analysis. Study design was
transversal (or case series) in 58 papers (56%), a case-control in 5 (5%), longitudinal in
26 (25%) and a randomized clinical trial in 15 (14%). Elementary statistical techniques
(descriptive, contingency tables, t test, correlation and linear regression) have been
applied at most in 41 papers (61.6% of methods used); intermediate level analysis
(multiple regression, ANOVA, use of multiple comparison techniques, nonparametric methods,
multiway tables, power considerations and transformation) was performed in 34 papers (32%)
and advanced statistical analysis (including epidemiologic statistics, survival analysis
and other methods of statistical modeling) was used in 29 papers (28%). At least one
statistical error was found in 61 papers (59%). These findings point to the need of
continuing statistical tuition for medical researchers.
|
