Preface

Clinical data is easier to understand when presented in a visual format. The human brain allocates a large percentage of its resources to take in and process visual information rapidly. Pattern recognition is key to human survival, and we can rapidly and accurately make sense of complex visual information to make decisions. We can make judgments on visual data even when we are not focused on the task explicitly.

In comparison to this remarkable ability developed through sheer necessity for survival, the remembering and processing of numeric data in raw tabular firm requires the explicit and intentional involvement of the cerebral cortex. The human brain is relatively slow in absorbing pure numbers, and remarkably poor in remembering more than a handful at a time. Furthermore, making evaluation of relative magnitude between such numeric data is slow.

Graphical views of the data allows quick processing and evaluation that can help in planning the analysis phase of the project. Results of the analysis are easier to understand when they are delivered in a graphical form. Graphical representation of the data along with the derived statistical information can be a key factor in understanding of the results.

Presenting the data as a simple bar chart or a scatter plot can help in its understanding. In some cases, sophisticated graphs with complex layouts help to understand the trends and see the associations in the data. These include graphs with raw data along with derived statistics and tabular information, classification panels by multiple class variables, scatter plot matrices of multiple measures and ad-hoc layout of dissimilar graphs necessary to display the information using multiple representations of data, often on a uniform scale.

To create such graphs you need a language to systematically describe the complex layouts and the relationships between the different parts of the graph. Individual graphs could be created using extensive annotate functionality, but such graphs are difficult to adapt to different situations, hard to build and hard to maintain. The new graphics software included in Base SAS® such as the Graph Template Language (GTL), the Statistical Graphics (SG) Procedures, and the ODS Graphics Designer provide you with the tools you need to create complex clinical graphs.

GTL is a comprehensive syntax to define the structure of a graph. GTL has a structured and logical syntax, necessary to build complex graphical layouts, with a large set of features. With a high level of features comes some complexity, so GTL has a significant learning curve. Often you just need a simple graph quickly. For such situations we can use the Statistical Graphics (SG) procedures, which provide an easy to use procedure like syntax to the GTL functionality.

GTL was first released with SAS 9.2, and was initially motivated by the needs of the SAS statisticians and procedure writers to create the graphs that are automatically created by the SAS analytical procedures. SG procedures provide a simplified, value added syntax to create graphs using GTL under the covers.

With SAS 9.3, significant new features were added to make the building of complex clinical graphs possible. This set of features was further expanded with SAS 9.4 and the maintenance releases to make clinical graphs easy. While most types of graphs can be made using the SAS 9.3 feature set, they are much easier to make using the SAS 9.4 features, many of which were developed expressly for such use cases.

In this book, I have described how to create many clinical graphs using SAS 9.3 in Chapters 3 and 7. Many new plot types and features have been added with SAS 9.40M3 making clinical graphs much easier to create. So, the recommended way to create clinical graphs is with the SAS 9.40M3 release as shown in Chapters 4 and 8.

Often, the SG procedures are all you need to create a large percentage of the graphs that are commonly used in the HLS domain. The SGPLOT procedure is designed to create "Single-Cell" graphs. These graphs comprise a very large proportion of the graphs in use that display all the data related information in one graphical data display area. Other items necessary to decode and convey the information such as legends, statistics tables, titles and footnotes are also included in the graph. The SGPANEL procedure makes it easy to create classification panels for one or more class variables.

Often, a complex, multi cell layout is necessary to create graphs that that contain a lot of information. The data in each cell has to be displayed on a uniform scale with other data and tabular information. Such graphs need a bit more structure and functionality and are best created using GTL.

In this book, I will organize the graphs in two categories based on complexity. Graphs we can create using SG procedures and complex graphs that require use of GTL. For each case, I will show you how to make the graph using SAS 9.3 features and also SAS 9.4. SAS 9.4 provides you with many new features that will make the task much easier.

Clinical graphs have their own aesthetic requirements which are based on industry standard usage and requirements of scholarly journals for publications or for submissions to regulatory authorities. Such appropriate visual aesthetics are built in by default and you have to do little to get the right graph "out-of-the-box". The graphics system is designed with the principles of effective graphics in mind to convey the information with maximum clarity and minimum clutter. However, extensive customizations can be done to meet your specific requirements.

This book shows you how to create the required clinical graph given the data. Often, the data I use is simulated using mathematical functions and random number generators. The graphs themselves attempt to duplicate the presentation of data as proposed by experts in the clinical domain. My goal here is not to invent new graphical displays for clinical use, but to show you how to create displays commonly used in the industry, and how certain aspects of the displays may be better from the point of view of effectiveness of the graph. Techniques for modeling and analysis of the data itself are beyond the scope of this book.