About the Scenarios in This Book

This book is divided into two examples, and each describes an extended scenario that is intended to familiarize you with the many features of SAS Text Miner. Within each example, the current topic builds on the previous topics, so you must work through the chapters in sequence. Several key components of the SAS Text Miner process flow diagram are covered. In these step-by-step examples, you learn to do basic tasks in SAS Text Miner, such as how to create a project and build a process flow diagram. In your diagram, you perform tasks such as accessing data, preparing the data, building multiple predictive models using text variables, and comparing the models. The extended examples in this book are designed to be used in conjunction with SAS Text Miner software.

The first example illustrates the use of the Text Mining node and uses the Vaccine Adverse Event Reporting System (VAERS) data. This data is publicly available from the U.S. Department of Health and Human Services (HHS). Anyone can download this data in comma-separated value (CSV) format from http://vaers.hhs.gov. There are separate CSV files for every year since the U.S. started collecting the data in 1990. This data is collected from anybody, but most reports come from vaccine manufacturers (42%) and health care providers (30%). Providers are required to report any contraindicated events for a vaccine or any very serious complications. In the context of a vaccine, a contraindication event would be a condition or a factor that increases the risk of using the vaccine. Please see the “Guide to Interpreting Case Report Information Obtained from the Vaccine Adverse Event Reporting System (VAERS)” available from HHS (http://vaers.hhs.gov/data/index).

See the following in the Getting Started Examples zip file:

The following figure shows the first 8 columns in the first 10 rows in the table of VAERS data for 2005. Included is a unique identifier, the state of residence, and the recipient's age. Additional columns (not in the following figure) include an unstructured text string SYMPTOM_TEXT that contains the reported problem, specific symptoms, and a symptom counter.

In analyzing adverse reactions to medications, both in clinical trials and in post-release monitoring of reactions, keyword or word-spotting techniques combined with a thesaurus are most often used to characterize the symptoms. The Coding Symbols for Thesaurus of Adverse Reaction Terms (COSTART) has traditionally been the categorization technique of choice, but it has been largely replaced by the Medical Dictionary for Regulatory Affairs (MedDRA). COSTART is a term developed by the U.S. Food and Drug Administration (FDA) for the coding, filing, and retrieving of post-marketing adverse reports. It provides a keyword-spotting technique that deals with the variations in terms used by those who submit adverse event reports to the FDA.

In the case of vaccinations, the COSTART system has been used. The FDA has used a program to extract COSTART categories from the SYMPTOM_TEXT column. Here are some of the variables used by the program:

If you open the VAERS data for 2005, you can see that VAERS_ID 231844 has SYMPTOM_TEXT of 101 fever, stiff neck, cold. The program has automatically extracted the COSTART terms that appear in column SYM01 to column SYM20 in the data file.

The VAERS table contains other columns, including a variety of flags that indicate the seriousness of the event (life-threatening illness, emergency room or doctor visit, hospitalized, disability, and recovered), the number of days after the vaccine that the event occurred, how many different vaccinations were given, and a list of codes (VAX1-VAX8) for each of the shots given. There are also columns that indicate where the shots were given, who funded them, what medications the patient was taking, and so on.

The README file taken from the VAERS Web site decodes the vaccine abbreviations. Note that some vaccinations contain multiple vaccines (for example, DTP contains diphtheria, tetanus, and pertussis). Here is a portion of the README file:

As you go through this example, imagine you are a researcher trying to discover what information is contained within this data set and how you can use it to better understand the adverse reactions that children and adults are experiencing from their vaccination shots. These adverse reactions might be caused by one or more of the vaccinations that they are given, or they might be induced by an improper procedure from the administering lab (for example, an unsanitized needle). Some of the adverse reactions will be totally unrelated. For example, perhaps someone happened to get a cold just after receiving a flu vaccine and reported it. You might want to investigate serious reactions that required a hospital stay or caused a lifetime disability or death, and find answers to the following questions:

When you are finished with this example, your process flow diagram should resemble the one shown here:

After you complete the VAERS data example, you will use the Text Parsing, Text Filter, and Text Topic nodes to analyze the abstracts from a collection of SAS Users Group International (now called the SAS Global Forum) papers. The goal of this example is to determine whether any themes are present in the papers. You will use the Text Topic node to create a set of topics that will describe the document collection. Additionally, you will create a user-defined topic that finds all abstracts related to dynamic Web pages. This example is independent of the VAERS data example and will result in a simpler process flow diagram.