Rheumatology Advance Access originally published online on February 20, 2009
Rheumatology 2009 48(5):462-463; doi:10.1093/rheumatology/kep025
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||
EDITORIALS |
Statistical guidelines for submission of papers—revised 2008
Most studies, whether laboratory or clinical, generate numerical data and interpretation of the findings depends as much on the statistical handling of the results as it does on the quality of the methods used for data collection. The major weakness in most publications is the absence of relevant pieces of information that allow the reader to interpret the study and compare it with the findings of others.Reporting guidelines for specific types of research studies are readily available from the Enhancing the QUAlity and Transparency Of health Research (EQUATOR) Network website [1]. This website has links to up-to-date guidelines for the reporting of randomized clinical trials (with extensions for reporting harms, cluster randomized trials, non-inferiority and equivalence trials) (CONSORT), systematic reviews (QUOROM/PRISMA), diagnostic accuracy studies (STARD), observational studies (STROBE) and meta-analysis of observational studies (MOOSE). Since 1988, the International Committee of Medical Journal Editors (ICMJE) has provided basic guidelines for presenting and writing about statistical aspects of research in the Uniform Requirements for Manuscripts Submitted to Biomedical Journals: Writing and Editing for Biomedical Publication [2]. Authors are requested to adhere to these guidelines when preparing manuscripts for submission to Rheumatology.
The following uses both of the above sources to produce a short and easy-to-follow set of statistical guidelines that should be viewed as a positive attempt to enhance the value of published work. They outline the approach that will be viewed by Rheumatology, and its reviewers, as constituting an appropriate standard of presentation.
Objective(s) of the study
These should be clearly stated in the Introduction section. The nature of the main questions to be answered will determine the statistical approach required.
Description of the methods used
The Methods section should include only information that was available at the time the plan or protocol for the study was written; all information obtained during the conduct of the study belongs in the Results section.
Subjects/Patients
- A clear description of the selection procedure together with inclusion and exclusion criteria.
- Justification should be provided for the number of subjects studied including details of any a priori sample size calculation.
Statistical methods
- All statistical methods used should be adequately described, and references provided for any methods that are likely to be unfamiliar to the general reader.
- If it is necessary to use more complex statistical methods, then a brief description of the method and justification for its use should be provided.
- The name and version number of any statistical software used should be stated.
- An indication that the statistical analyses chosen were appropriate in respect of:
- the structure of the study e.g. whether samples were independent, paired (e.g. before/after comparisons) or repeated evaluations of the same subjects;
- any distributional assumptions required e.g. normality and constant variances.
- the structure of the study e.g. whether samples were independent, paired (e.g. before/after comparisons) or repeated evaluations of the same subjects;
- Avoid sole reliance on statistical hypothesis testing, such as the use of P-values, which fail to convey important quantitative information. Wherever possible, use estimation methods and present results with appropriate indicators of uncertainty, such as CIs. For example, the statement that two means are significantly different is less informative than an indication of the range in which the true difference might lie. CIs can be calculated for many estimates, such as means, proportions, in statistical packages including SPSS, Stata and CIA [3–5].
Presentation of results
Present your results in logical sequence in the text, tables and illustrations, giving the main or the most important findings first. A general guideline is that there should be sufficient information presented to justify the conclusions and to enable others to repeat the work.
Data presentation in the text
- Response rates should be given for clinical/epidemiological studies.
- Descriptive statistics should be provided for each variable analysed. Categorical data should be represented by frequency and, where appropriate, percentages. The distribution of numerical data should be examined prior to calculating the appropriate summary measures: normal distribution—means and S.D.; non-normal data—median and inter-quartile range.
- When presenting summary measures, the use of the ± symbol without further quantification is ambiguous. Rather, authors are encouraged to use parentheses, i.e. ‘The mean (S.D.) age of the group was 54.6 (6.7) years.’
- If outliers (extreme values) are likely to distort the results, especially in small studies, this should be discussed.
- Whether using hypothesis tests or estimation methods, the researcher should present sufficient information for the reader. When presenting results from hypothesis tests, appropriate summary measures and sample sizes should be presented together with the associated P-value. Results using estimation methods should be presented as estimated differences together with the associated 95% CI.
- Exact P-values should be presented, rather than merely stating whether a difference is statistically significant at a specified significance level, i.e. ‘P-value not significant’ or ‘P < 0.05’.
- Some statistics packages, including SPSS and Stata [3, 4], present P-values smaller than 0.001 as 0.000. The researcher should be aware of this anomaly and present the output correctly, i.e. P < 0.001.
Data presentation in tables and figures
- Figures used to present raw data rarely have any advantage over appropriate descriptive statistics. Restrict tables and figures to those needed to explain the argument of the paper, using the most appropriate form of data presentation.
- The same information should not be presented in a table and in a figure.
- Pie-charts should be avoided; data should be given either in the text or in a frequency table.
- Bar charts should be presented using the 2D option. The 3D option should only be used when the data being presented has three dimensions.
Interpretation of results
- Any conclusions drawn should be justified based on the results presented.
- Without a CI, a non-significant result might reflect too small a sample size rather than the absence of a clinically relevant effect.
- If many hypothesis tests have been performed, major, pre-specified hypotheses should be distinguished from any explanatory subgroup analyses that might produce a significant result by chance.
References
- The EQUATOR Network website. http://www.equator-network.org/index.aspx?o=1032.
- The ICMJE website. http://www.icmje.org/- prepare.
- Stata website. http://www.stata.com/.
- SPSS website. http://www.spss.com/.
- Altman DG, Machin D, Bryant TN, Gardner MJ. Statistics with confidence. (2000) 2nd edition. London: BMJ Books.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  E. Thomas and C. J. Dore Statistical guidelines for contributors to Rheumatology Rheumatology, May 1, 2009; 48(5): 461 - 461. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||