### Department of Biostatistics Seminar/Workshop Series

# **A New Model for Reproducible Clinical Trial Statistical Reporting**

## Frank Harrell, PhD

##### Professor and Chair, Department of Biostatistics, Vanderbilt University School of Medicine

Statisticians and statistical programmers spend a great deal of time analyzing data and producing reports for clinical trials, both for final trial reports and for interim reports for data monitoring committees. Point and Click interfaces and copy-and-paste are now believed to be bad models for reproducible research. Instead, there are advantages to developing a high-level language for producing common elements of reports related to accrual, exclusions, descriptive statistics, adverse events, time to event, and longitudinal data.

It is well appreciated in the statistical and graphics design communities that graphics are much better than tables for conveying numeric information. There are thus advantages for having statistical reports for clinical trials that are almost completely graphical. For those reviewers of clinical trial reports who insist on seeing tables, and for those who occasionally like to have tables to see "exact" figures for certain data elements, supporting tables can be placed in an appendix. These tables are hyperlinked to the main graphics. Small tables can also pop-up when one hovers the mouse over a graphic. These two approaches are facilitated by features of Adobe Acrobat Reader. Reviewers who prefer printed reports can print the appendix in order to have a complete document.

In this talk I will describe a new R package called greport ("graphical report") that marries R, the R Hmisc and lattice packages, knitr, and

LaTeX to produce reproducible clinical trial reports with a minimum of coding. greport composes all figure captions and makes heavy use of analysis file annotations such as variable labels and units of measurement. Some new graphical elements are introduced such as special dot charts that replace tables, extended box plots, split violin plots for longitudinal continuous variables, half confidence intervals for differences, new charts for representing patient flow, and pop-up tooltips. Supporting tables are hyperlinked to graphics, and the graphics are hyperlinked back from the tables. Figure captions contain supporting table numbers, and tables contain figure numbers. Two example reports will be shown.

For more information see

http://biostat.mc.vanderbilt.edu/Greport