MAED Service: A SAS Tool for Implementing SMQs and ...

1 PharmaSUG2010 - Paper AD13 MAED Service: A SAS Tool for Implementing SMQs and Performing meddra -Based Analyses of AE Data P. Chris Holland; MacroGenics, Inc.; Rockville, MD ABSTRACT The MAED Service is an award-winning SAS -based tool that was developed for FDA medical and statistical reviewers of adverse event (AE) data coded to the Medical Dictionary of Regulatory Activities ( meddra ). MAED (pronounced like maid ) stands for meddra -based Adverse Event Diagnostics. The MAED Service allows reviewers to look at AE data by each level of the meddra hierarchy and both narrow- and broad-scope standardized meddra queries (SMQs). All SMQs, including those with algorithms, are derived using each patient s AE data.

2 For each level of the meddra hierarchy and each SMQ class (narrow- or broad-scope), terms are sorted and displayed in a summary report by the degree of disparity between comparison groups ( test drug vs. placebo). So the first term displayed in the MAED Service report represents the term, for the given level, with the largest degree of disparity between the groups being compared. This paper describes the first incarnation of the MAED Service and makes available the code needed to run it. A subsequent paper will describe a more advanced version of the tool that runs the code as a SAS stored process behind a web application. INTRODUCTION The meddra dictionary was first released for implementation in 1997.

3 It was initially developed under the auspices of an ICH working committee focused on replacing the multiple coding terminologies that were being used worldwide with a single international standard. meddra is now widely used in the biopharmaceutical industry. One notable advantage of meddra is its hierarchical structure. The 5 levels of the hierarchy include the lower level term (LLT), which is a very granular level used for coding verbatim terms; the preferred term (PT) level, which represents single medical concepts and is commonly used in AE summaries; the high level term (HLT) level; the high level group term (HLGT) level; and the System Organ Class (SOC), which is also commonly used in AE summaries.

4 Figure 1 depicts this hierarchical structure showing the number of unique terms at each level. As shown, there are now over 18,000 PTs. The large degree of granularity at the PT level has its advantages, but it has its limitations as well. With so many PTs, important safety signals have the potential to go unnoticed when the same signal gets coded across multiple PTs. For example, a PT search on the term rash could yield over 30 results including terms such as Rash papular , Rash macular , Rash maculo-papular , and just Rash . In any given submission, each of these terms and more could be spread across many subjects and a few different SOCs.

5 Differences between groups may be insignificant for any one term and safety signals may therefore go unnoticed until related terms are grouped together more logically. Figure 1: meddra Hierarchy Pyramid for meddra Version 1 A remedy to these problems is to summarize AE rates at the HLT and HLGT level, rather than just the PT and SOC level as is commonly done. Another remedy is to take advantage of the SMQs provided by meddra . As defined by the meddra Introductory Guide Version , SMQs are groupings of meddra terms, ordinarily at the Preferred Term (PT) level that relate to a defined medical condition or area of interest. SMQs are intended to aid in the identification and retrieval of potentially relevant individual case safety reports.

6 SMQs allow for the grouping of like terms that may not otherwise be grouped even at the HLT and HLGT level. Usually, when sponsor companies do apply SMQs to their clinical AE data, they do it only for a selected query or two that corresponds to a suspected safety issue. Despite these resources available to sponsor pharmaceutical companies who license the meddra dictionary, they often go underutilized in safety summaries and reports that get packaged in New Drug Applications (NDAs) and Biologics License Applications (BLAs) submitted to the FDA for marketing approval. FDA personnel responsible for reviewing these applications are charged with the task of determining whether these products are safe and effective enough to be sold in the United States.

7 In doing their evaluations and safety assessments, many have begun to realize the limitations of standard AE summaries that only provide information at the PT and SOC level. They have been in need of a tool that would allow them to easily evaluate AE data in a way that takes advantage of these underutilized meddra features. In order to address this need, the author, who was working at the time as a statistical safety reviewer in the Center for Drug Evaluation and Research (CDER) at FDA, worked with medical reviewers to develop the MAED Service as a tool that could take meddra -coded AE data and produced summaries at each level of the hierarchy and for each and every SMQ (for both narrow- and broad-scope SMQs, including those with algorithms).

8 For each summary, terms are sorted and displayed by the degree of disparity between comparison groups, such as test drug vs. placebo. P-values are used to measure the degree of disparity. So the first term displayed in the summary report for a given level represents the term with the smallest p-value relating to the groups being compared. The results are provided in a bookmarked PDF report that allows for easy navigation through each level of the hierarchy. This paper describes the framework and makes available to readers the code needed to develop their own in-house version of the MAED Service. Although the tool was later developed into a purely web-based tool that runs SAS behind a web-server, this paper describes the initial version of the tool that does not require a web server but does require that users have access to the SAS System.

9 Although the tool was developed for the FDA reviewer environment, it could be of use to any organization with a large amount of meddra -coded AE data from controlled trials and an interest, or regulatory responsibility, to monitor the safety of the products to which the AE coding was applied. RUNNING THE MAED SERVICE Once the MAED Service has been set-up, it can be run in two steps. The first step involves filling out the request form, and the second step involves cutting and pasting the generated SAS code into SAS and running it. STEP 1: FILLING OUT THE MAED SERVICE REQUEST FORM Before the MAED Service can run, it needs to know certain bits of information about the data on which it will be running.

10 This information is captured on web form. The form does not need to run on a web server. Rather, it can reside on any network drive to which users will have access. The page uses JavaScript to read the data entered and to then generate SAS code. Figure 2 depicts this web form. Two data sets are required-- an adverse event data set that contains one record per adverse event and a demographic/denominator data set that contains one record per subject. If the data sets are in CDISC SDTM format, then the MAED Service only needs to know the directory where these data sets exist and whether they are in a native SAS format or a SAS transport file format.