A well-designed graphic can effectively communicate a message to a range of audiences and help identify patterns in data that might otherwise be missed.
In clinical trials, when analysing adverse events where there is often an abundance of complex data, graphical visualisations can be particularly useful to identify potential adverse treatment reactions. Trial reporting guidelines such as the CONSORT harms statement and the 2016 recommendations from Lineberry et al. encourage the use of visualisations for exploring adverse event data.
But use, at least in journal articles, is limited. A systematic review we performed in 2018 found that only 12% of journal articles made use of visual summaries for adverse event data, and this finding has been supported by a recent survey of UKCRC CTU statisticians (results to be published soon).
There are an abundance of potential visualisations for adverse event data (work presented at ICTMC 2019 summarised here) and a repository of some useful ideas can be found here. Two underused but potentially useful plots to summarise the body of emerging adverse event data collected in a clinical trial are the volcano and dot plots.
The volcano plot (first proposed for analysing adverse event data by Zink et al.) displays individual events as a bubble, with the size, colour and position of the bubbles used to illustrate unique pieces of information about the event frequency, size of the treatment effect and direction of effect respectively (see figure 1). The name comes from its typical resemblance to an image of a volcanic eruption with the most significant adverse events as bubbles at the top. Outlying bubbles can be used as a signal to flag an event for further scrutiny (i.e. a potential adverse drug reaction).
Figure 1: Example volcano plot
The x-axis represents the difference in proportions of participants experiencing each adverse event between the treatment arms (intervention – placebo). The y-axis represents the p-value from a Fisher’s exact test on the -log10 scale. The centre of the bubble indicates the coordinates for each adverse event. The size of the bubble is proportional to the total number of events for both treatment groups combined. Colour is used to indicate direction of treatment effect with red indicating greater risk in the intervention group and blue indicating greater risk in the placebo group. Colour saturation corresponds to the -log10(p-value) for each event.
The dot plot (first proposed by Amit et al. for adverse event data) displays absolute treatment group differences for individual events (such as risk difference) and relative measures (such as risk ratio) with an indication of precision (such as 95% confidence interval) alongside each other (see figure 2). The dot plot can be a useful, space efficient, visual alternative to the traditional two-by-two table adverse event data is typically displayed in.
Figure 2: Example dot plot
The left side of the figure displays the percentage of participants experiencing an adverse event (labelled on the y-axis) in the intervention group with a red triangle and placebo group with a blue circle. The right side of the figure displays the relative risk and corresponding 95% confidence interval on the log10 scale.
Producing informative visualisations is one of the many tasks busy trial statisticians carry out. Statistical software, such as Stata, eases some of the burden facing trial statisticians. But an effective reproducible visualisation can take a lot of sophisticated code. To date production of the volcano and dot plot in Stata would require lengthy, complex coding. To encourage wider use of these visualisations we have developed simple to use Stata commands for production.
Commands aevolcano and aedot for individual participant data and commands aevolcs and aedots for summary level data can now be downloaded here or by typing ssc install commandname in the command line within Stata.
aevolcano (or aevolcs) and aedot (or aedots) allow users to produce the volcano and dot plots with one simple line of code. We have written accompanying help files for both of these plots and will be hosting training sessions on how to use them in the coming months. In the meantime, please do reach out via email (r.phillips@imperial.ac.uk) if you find any bugs in the code and please do feel free to share your images on how you implement these visualisations.
Happy plotting – Rachel & Suzie.
Comments