Description and reporting of surgical data--scope for improvement?

BACKGROUND

Surgical research articles utilise a variety of descriptive and inferential methods to present and analyse data. The aim of this study was to determine the appropriateness of descriptive methods (e.g. mean, median, SD, range) and survey the use of inferential methods (statistical tests) in key general surgical journals.

METHODS

Articles were retrieved from the first three issues of four general surgical journals (Annals of Surgery, British Journal of Surgery, The Surgeon and the Annals of the Royal College of Surgeons of England) in the year 2005. Only original articles were included. Study characteristics, use and appropriateness of descriptive statistics and the number and types of statistical methods employed were evaluated.

RESULTS

Of the 144 articles analysed, 119 made use of continuous data. Of these, the use of descriptive methods was inadequate or inappropriate in 50%. The use of '+/-' to describe data dispersion without further explanation occurred in 6.7% and standard error of mean (SEM) was inappropriately used in a further 9.2% of the articles. No statistical methods were described in 23% of the articles. The six most common tests used (Pearson's chi-square, Fisher's exact test, unpaired t test, Mann-Whitney test, log rank and the Cox proportional hazards) accounted for the majority (69%) of the statistical methods employed.

CONCLUSION

These results highlight the need for training surgical researchers in 'data description and reporting' and for improved screening by journal reviewers/editors. Critical appraisal of surgical research can be undertaken with a comprehensive understanding of only a limited number of statistical methods.