Seeing is believing: Making clinical trial statistical data from medical product testing easy to understand

Posted on September 23, 2015 by FDA Voice

By: Richard A. Forshee, Ph.D.

If you heard that some FDA scientists were helping people pick out colors and designs, you might wonder if the agency had added interior decorating to its responsibilities.

What they’re really doing is helping scientists craft statistical graphs and plots of clinical trial safety data so that they tell clear, compelling stories. Key to success? Creating tables and graphs that aren’t so dense with numbers, boxes, lines, and words that extracting meaning from them is like excavating hard rock for minerals.

To do this, the Safety Graphics Working Group, a team from FDA, industry, and academia, created a web-based, publicly available database of graphical designs for reporting clinical trial safety data from tests of new medical products. Graphics expertise and funding provided by Vanderbilt University enabled the development of this web site, which was launched in 2013. More recently, the team published details of the concept of improving statistical graphics design in the journalStatistics in Medicine.

The collaborators developed a step-by-step process for creating statistical graphs and plots that quickly explain important findings by taking advantage of the way peoplenaturally look for patterns in images, not just in pictures of nature, but also in representations of statistical data. They also provide computer codes for re-creating the models available on the website. The choices include some fanciful names, such asspaghetti and lasagna graphs, and violin and forest plots.

Designing good graphs and plots for specific types of data requires thoughtful approaches to illustrating how that data can be most clearly displayed. This chart is a guide to choosing the most effective design for displaying data, depending on how much detail is to be displayed and what aspect of the data you want to emphasize. In this case, the chart provides options for displaying a type of data called a “continuous variable,” which could be, for example, how a specific drug has affected blood pressure as measured at the end of treatment.

The authors of the article also demonstrate how to use effective, self-explanatory symbols on graphs to explain the data instead of overwhelming a reader with it. Such data can include any of a wide variety of measurements, from liver enzymes to prolonged Q-T intervals (abnormal electrocardiogram results associated with heart beat abnormalities, fainting, and sudden death).

Of particular importance for FDA regulators, the designs make it easier to clearly illustrate statistical data representing a wide variety ofpotential adverse effects (AEs) observed during clinical trials of medical products or during post-licensure use (when products are on the market). For example, they might show which AEs are elevated in treatment versus control groups, which AEs are elevated in specific patient subgroups, and which might be a safety signal (potential link between a medical product and an adverse effect).

Creating those compelling graphs and plots is important to other scientists who read journal articles to keep up with the latest research in their field. And it’s also very important to a lot of people who make key decisions based on that data: editors of journals deciding whether a researcher’s paper is important enough to publish, and FDA regulatory officials reviewing clinical trial results of medical products submitted by industry.

Most scientists might think it’s impossible to turn statistical illustration into an art form. But the work of the Safety Graphics Working Group shows that “seeing is believing.”

Richard Forshee, Ph.D., is FDA’s Associate Director for Research at the Office of Biostatistics and Epidemiology in the Center for Biologics Evaluation and Research