[Use and presentation of statistics methods in original articles published in MEDICINA CLINICA in 1993]

In the last few years a marked increase has been observed in the use of statistical techniques in biomedical publications. Some characteristics of the statistical quality of the first 100 articles consecutively published in 1993 in the section of originals and surveys of the journal MEDICINA CLINICA are presented in this study.

In each original one reviewer identified errors and/or criticisms in statistical methodology. An adaptation of the protocol of statistical revision made by the journal The Lancet since november 1990 was used in addition to its own classification. Likewise, an error and/or criticism involving minimum statistical quality necessary was considered as major and that which only had a decrease in optimum statistical level was considered as minor. Statistical analysis consisted in descriptive tables of the number of originals in which each of the classified errors and/or criticisms were observed.

Sixty-seven percent of the originals (CI 95%: 64;70) presented major errors and/or criticisms in design, analysis and inference. The most frequent were found in power and sample size (design), need for better analysis (analysis) and lack of confidence intervals (inference). Only 15% (CI 95%: 13;17) had major errors in statistical presentation and 82% (CI 95%: 80;84) minor errors among which the so-called orphaned p was of note.

The need for accompanying the p values by the confidence intervals or taking the calculation of the required sample size into account are of note. Furthermore, the diffusion of explicit recommendations concerning the carrying out and presentation of statistical analysis is necessary.