Statistical Analysis Plan

The SDBC uses statistical analysis plans to ensure effective communication between statisticians and clinicians and to help document the analyses that accompany a manuscript. This page details a typical statistical analysis plan, why they are important, and pitfalls of not using a statistical analysis plan.

A typical Statistical Analysis Plan consists of the following:

1. Introduction - Statistician’s understanding of the study design and overview of project goals
2. Investigator’s Description - Copied from SDBC request form
3. Data - Description of how data was collected, inclusion/exclusion criteria, important predictor and outcome variables
4. Research Objectives - Primary (and secondary) analysis goals that are well-defined and testable
5. Analyses - Statistical methods that will be used to accomplish the research objectives

* To maximize the benefit of the Statistical Analysis Plan, sections 1-4 must be carefully read and approved by the investigator. Investigators should ask questions about the sections above. The Statistical Analysis Plan should be approved by investigators via email. *

Why Use a Statistical Analysis Plan?

1. Facilitates communication between the statisticians and the clinical investigators
2. Enables reproducibility – other statisticians could pick up this plan and reproduce the analysis
3. Helps prevent false positive results by potentially minimizing the number of statistical tests
4. Journals are starting to require statistical analysis plans to be submitted with the manuscript

For more information, please see "The Value of Statistical Analysis Plans in Observational Research" (Thomas et al. 2012)

Pitfalls of Not Using a Statistical Analysis Plan?

Most false findings are due to poor study design and/or research practices, in particular:

1. Small studies (insufficient power/spurious results)
2. Small effect sizes (there is only a small expected difference between groups)
3. Lesser preselection of tested relationships (i.e. not using an analysis plan!)
4. Greater number of tested relationships
5. Using multiple variable definitions, outcome measures, and analysis methods

For more information, please see "Why most published research findings are false" (Ioannidis et al. 2006)

We are experts in study design and will help you to avoid these errors!