 I'm going to talk today about graphical displays in R for clinical trials. I'm going to briefly survey the history of statistical graphics, discuss the essential role of visualization of clinical trial data, the use of R in creating specialized graphics for clinical trials. And then I'll talk briefly about major pitfalls to be avoided in graphical displays for clinical trials and guidelines for well-designed powerful graphics. Graphical displays are often, as we all know, much more informative than numerical tables and lists. Visualization of clinical trial data is essential for evaluating the progress of the trial and for first determining the outcome, then understanding what we found and finally explaining it to a wider audience. The creation of effective graphical displays involves art as well as science. The psychology of human perception matters greatly in how a graphic will be understood by the viewer and we'll talk more about that later. So this process requires careful thought and also careful execution. It's valuable to keep in mind always what information would be helpful to the audience and how can we show them this information accurately and without misleading. The capabilities of R let us develop and produce clinical trial graphics for many purposes, exploring the data, monitoring and assessing safety and efficacy. I think I'll skip this list now because I will go into it in more detail in my discussion of R. For clinical trial data, we must take great care to avoid pitfalls and presentation errors in creating our graphics and to follow guidelines and best practices. The history of statistical graphics might begin in the classical period. Some highlights include play fairs development in the late 1700s of the first line plot bar chart and pie graph. These methods are still widely used today. Snow's work on the cholera epidemic in London in which he plotted cholera deaths by neighborhood to locate the source of the disease. And Florence Nightingale's use of statistical graphics to demonstrate to the British Army that sickness caused many more deaths than battlefield injuries. Consequent improvements in army hygiene after the presentation of her work dramatically reduced the overall death rate in the army. In the first half of the 20th century, as quantitative methods like regression analysis and analysis of variants were developed and in their ascendancy, graphical methods were relegated to a lesser role. They were commonly considered less precise and less important than the new quantitative methods. Anscombe, an insightful paper in 1973, commented about statistical packages at that time that most of them originated in the previsual era that is in the first half of the 20th century. And Anscombe said, the user is not showered with graphical displays. We can get them only with trouble, cunning and a fighting spirit. It's time that was changed. Anscombe presented a striking example for simple data sets, each with just two coordinates X and Y and 11 points. If you don't look at the plot that I'm showing you, but if you just take the numbers off and put them into your favorite statistical package, the four data sets will all give identical results in their regression analyses and their analyses of variants. You can look at the quantitative output, the pages of output and say, well, these all look very similar. They're identical. But when you look at the plot, you see how very different graphical patterns can produce the same quantitative results. This drives home the essential need for statistical graphics in order to distinguish these four cases that are illustrated here from one another. And the same is true even to a much, much greater degree when we have dozens or hundreds or thousands or perhaps millions of variables and equally large or larger numbers of observations. Graphics play an essential role in showing the structure in finding and showing the structure of large data sets. In the 1950s and the 1970s, Tukey developed a philosophy of exploratory data analysis. EDA uses graphics, quantitative analyses, and everything else we know, different kinds of analyses that don't fall into either of these categories and knowledge from every source to determine as best we can the underlying structure of the data, figure out the key questions, answer them and interpret them. The work of Edward Tufty put graphics on the statistical map, ending what Anscombe called the previsual era. In his first book, the visual display of quantitative information came out in 1983. His books are considered as a review in the Journal of the American Statistical Association, a review of the second edition says, the cornerstones of today's information visualization research. I've listed Tufty's table of contents in visual display of quantitative information. I don't have time to go into the details. A current account is in a book published in this year, 2021, an account of the crucial role of data visualization in informed decision making. And I also want to call your attention to the statistical thinking blog of Frank Harold. Frank has developed an extensive body of work on all aspects of statistical thinking and analysis and methodology, and the impact of statistical work on science and everyday life. The range of topics is both broad and deep. Those of particular interest to us today are statistical graphics, computing and our and modeling and clinical trial design analysis and reporting. And there's much more of great value here. Well conceived and well constructed graphs and plots avoid the syndrome. Niko, my eyes glaze over that can occur in evaluating a clinical trials progress and determining understanding and explaining its outcome. As we noted earlier, art as well as sciences involved and careful thought and execution are required. Now are is particularly well suited to developing and creating both the standard graphics for clinical trials and innovative graphics for tasks that are that are necessary for clinical trials, but not commonly found elsewhere. These aspects of a trial include to name a few exploring the data detecting irregularities outliers anomalies, monitoring and assessing operational success. The performance of clinical research sites variability within and between these sites on key metrics. And assessing patient recruitment and retention and the production of complete and accurate data from patients. Monitoring and assessing the safety and efficient efficacy treatments, displaying variability and other points we've talked about before are is particularly well suited because its flexibility allows us to develop the methodology we need. The methodology that no pull down menu, even with an extensive list of options ever can. So, the our programming environment is enormously helpful for the broad suite of algorithms that we require to accomplish these tasks are deals with diverse data types. Text and binary data, single measurements time series event based outcomes and lots more. This flexibility is a crucial asset for our. I've listed here a few of the wide variety of our packages available for creating and developing. Graphics from the task view graphics are computation for graphics website, both general packages like lattice and gg plot to, and others that are more specialized. Let's discuss now major pitfalls. We must avoid in graphical displays for clinical trials. The objective of displays is to convey the outcomes of clinical trials clearly concisely, completely and correctly. Data visualization should never mislead or misinform. We can't afford to make honest mistakes because they have the same results as bad intentions. So here are a few common errors made in data visualization. High color contrast leads to a perception of great differences among the values, often greater disparities that are actually present. Conversely, if a graphic has a low color contrast, it can make differences that are substantial seem to be much smaller. So the choice of colors and contrast, I recommend always steering those manually, not using the default and assuming that it will be effective, it often won't. Three dimensional graphics are appealing, but often introduced distortion, we have to be aware of that elements in the foreground appear larger. Some objects in the background appear smaller. Some objects in the front obscure those behind them. The scale relationships are confusing. So I would say use 3D graphics, ideally not at all, but if you use them with great caution, only when clearly beneficial. To present too much information. They can overwhelm and overload the viewer. So just don't show too much super imposing two or more variables on the same graph can show a correlation that gives a strong visual impression of causation. There are two very similar plots of say values of variables across a period of, you know, 10 or 11 years or 20 years, and if the plots are very similar, and the correlations are high. It may seem as if there is causation going on, it sort of leads us in that direction. So, when the variables really are related, we can be. It can be suggested to us we can be convinced that there's causation going on when there isn't, if we're not careful. So the inattention to the scale values of X and Y. In order to show data, the fluctuations in more detail or to improve the visual appeal can really be highly misleading. So the Y axis, only from 200 to 300. The difference between the value 205 and the value 290 will seem enormous, much greater than it actually is for example. Guidelines for well designed graphics for clinical trials and other uses. Could start with Tufty's six fundamental principles of design. I don't have time to discuss all of them, but I want to focus on the strong element of subjectivity in them. The variables we choose to perform our analysis, for example, but particularly on the last two, establishing credibility and focusing on content. Okay. Major goals at various stages of clinical trial can include data display exploration and analysis and reporting the results, all of which require different kinds of graphics to address the needs of these different tasks, and of different audiences. And work in progress with collaborators requires different graphics from regulatory reports, peer reviewed papers and articles addressed to the general public. So choosing graphics best suited to your goals from your immediate goal to your final goal is of vital importance. I've listed a few examples here that I will go through the list and move on. So, graphics must be chosen carefully designed carefully, and their contents must be specified carefully. If you include error bars in your graphics, they could mean standard error, or 95% confidence interval, or several other things your viewer won't know, unless you tell what those error bars mean. So titles, labels and legends should be informative and appropriate, but not too informative, fully informative but not, you know, an over abundance of detail. From an overview of the data set down to small individual chunks of data can be viewed with interactive graphics, viewing or hiding components at different levels as needed. And I've listed a few are tools for producing, you know, reports of this kind here. So, in summary, make sure your graphics represent the data accurately and deliver the intended message. And make sure that they support suitable interpretation of the trial data that they address the key questions of the study clearly and accurately. And that they keep everything as simple as possible the display should be as simple as possible, but no simpler. In other words, don't oversimplify. In conclusion, graphical displays are a vital component of the conduct and analysis of clinical trials. A lot of progress has been made, but much more remains to be done. The importance of specialized graphics for specialized tasks of which I've mentioned a few in clinical trials. The importance of graphics for implementing analyzing understanding and explaining clinical trials can hardly be overstated. Because of its flexibility and power are as ideally positioned to lead in both refining standard graphics and developing new innovative graphics for clinical trials. Thank you very much.