Appendix A Guidelines for "Statistical Considerations" Sections

A well-developed statistical considerations section persuades reviewers that solid skill and effort have gone into framing the research questions, planning the study, and forming an appropriate team. The writing should be crafted for the clinical researcher who is a good "para-statistician," as well as for the professional biostatistician. The "Statistical Considerations" section should be mostly self-contained and thus may reiterate information found elsewhere in the proposal.

A.1 Components

Design. Summarize the study design. It may be helpful to use appropriate terms such as randomized, double blind, crossover, controlled, comparative, case-control, prospective, retrospective, longitudinal, and cohort.

Research questions. Strictly speaking, not all studies are driven by testable hypotheses, but all studies have research questions that should be delineated in your Specific Aims. Summarize the outcome measures and describe how you expect them to be related to the components of the study design and other predictor variables. Restate/translate your primary research questions into specific estimates of effects and their confidence intervals, and/or into statistical hypotheses or other methods. Similar descriptions regarding secondary questions are valuable, too.

Statistical analysis plan. Specify what statistical methods will be used to analyze the primary and secondary outcome measures. Cite statistical references for non-routine methods. (Example: The two groups will be compared on KMOB830430 and its metabolites using estimates and 95% confidence limits for the generalized odds ratio (Agresti, 1980), which is directly related to the common Wilcoxon rank-sum test.) These sections often state what statistical software package and version will be used, but this usually provides little or no information about what actually will be done.

Randomization (if appropriate). Specify how the randomization will be done, especially if it involves blocking or stratification to control for possible confounding factors.

Sample-size analyses. State the proposed sample size and discuss its feasibility. Estimate the key inferential powers, or other measures of statistical sensitivity/precision, such as the expected widths of key confidence intervals. Strive to make your sample-size analyses congruent with the statistical methods proposed previously, and discuss any incongruencies. State how you arrived at the conjectures for all the unknowns that underlie the sample-size analysis, citing specific articles and/or summarizing analyses of "preliminary" data or analyses presented in unpublished works. If a sample-size analysis was not performed, state this categorically and explain why. For example, the proposal may only be a small pilot study.

Data management. Summarize the schema for collecting, checking, entering, and managing the data. What database software will be used? How will the database be tapped to build smaller analysis datasets? Note how you will meet modern standards for data security.

Technical support. Who will perform the necessary database and statistical work? If such people are less experienced, who will supervise the work?