TABLE 1.

—Shook, Hardy & Bacon's Draft GEP Resolution for a “Sound Science” Coalition54

THE EXECUTIVE COMMITTEE RECOMMENDS that the members of the Sound Science Coalition adopt and actively promote in the scientific community at large and within their individual disciplines, appropriate and specific professional standards for epidemiological research, to be carried out by accredited individuals and institutions, reflecting the following principles:
1. The study design should clearly define all objectives and hypotheses. Possible problems in design and data interpretation should be described and the intended method for addressing each fully set out. Every effort should be made to address possible confounders to avoid the need for subsequent adjustments.
2. In case–control studies, special attention should be given to how the control group will be selected and what matching procedures will be utilized. Also, case selection should be explained with emphasis on efforts to ensure a high participation rate. A pre-calculation of required sample size should be carried out to ensure that the sample is sufficient to produce meaningful results.
3. Statements of study design should contain a description of statistical techniques. This should include underlying assumptions for distribution, variance, correlation and regression procedures. The degree to which violation of these assumptions would invalidate the analysis should be specified whenever possible.
4. Adherence to the study protocol should be as close as possible. Any deviations (e.g., errors in randomization, low participation rate, suspected confounders, possible misclassification) should be documented.
5. Special care should be given to the training and monitoring of those administering questionnaires and surveys; blinded techniques are preferred.
6. After the study is conducted, the results should be analyzed as specified by the study protocol. Two-sided hypothesis tests are encouraged. If a one-sided test is employed, this should be noted and the rationale for using it provided. The presentation of confidence intervals for the estimate of risk gives more information than a single point value with an associated p value. Generally, 95% confidence intervals are preferred.
7. An adequate description of the raw data should precede and complement formal statistical analysis. If the data are not supportive of the stated hypotheses, no further analysis is necessary. Subsequent treatment of the data should only be for hypothesis generating purposes.
8. Odds ratios of 2 or less should be treated with caution, particularly when the confidence intervals are wide. There is a likelihood that the odds ratio is artefactual and the result of problems with case or control selection, confounders or bias.
9. Meta-analysis and pooling techniques are best used for homogenous data gathered under a uniform pool.
10. Observations that are inconsistent with the main body of the data should not be excluded from the analysis.
11. Journal articles and scientific conferences are the appropriate forum for the presentation of research results. Every effort should be made to publish and report on all completed research, regardless of outcome. Only by such efforts can the entire sample of conducted research be made available to the scientific community and publication bias minimized.
12. Generally, hypotheses tests not specified by the study protocol should not be reported. When many hypotheses tests are performed on data from a single study, a number of positive results can be expected to arise by chance alone, creating serious problems of interpretation.
13. Recognizing that a statistically significant association does not in itself provide direct evidence of causal relationship between the variables concerned and that causation can only be established on nonstatistical grounds, particular care should be taken when comparing two variables that have changed over time. Such comparisons often produce apparent associations.
14. Graphic display of results and figures that show individual observations are to be encouraged. For example, when appropriate, fixed regression lines should be presented together with a scatter diagram of the raw data. Any complex statistical methods should be communicated in a manner that is comprehensible to the reader.
15. Rigorous scientific objectivity should be the standard when reporting on epidemiological results. Defects in study design, conduct and analysis should be frankly admitted. It is helpful for abstracts accurately to reflect any study deficiencies. Advocacy and objectivity rarely comfortably coexist.