Discussion Club: FDA/PhUSE CSS Working Group 5 Update Budapest 2012
FDA/PhUSE CSS Working Group 5 Update at the annual PhUSE Conference 2012
The FDA/PhUSE CSS Working Group 5 collaborates on the topic "Development of Standard Scripts for Analysis and Programming". This working group will identify potential standard scripts for data transformations and analyses across and/or within a therapeutic area and exploratory analysis. The goal will be to begin the process of standardizing analyses across the industry, but also include examples of what can be done with a standardized data set. This will be an interactive session using online wiki to create and add input that can be shared openly between industry and regulatory.
Progress made since the CSS meeting in March in Washington
- Planned White Papers on Common Analysis for Clinical Trials
- Draft white paper available for review on Measures of Central Tendency
- Need people to help draft and review additional white papers below:
- Outliers or Shifts from Normal to Abnormal
- Adverse Events and Deaths
- Demographics, Concomitant Medications, and Disposition
- Any other ideas for White Papers?
- Need developers (R and SAS) to code analysis based on white papers
- Actively working with Google Code environment for publishing of scripts
- Looking for developers to give it a try…
Minutes of the discussion
We had a 2 hour session to discuss the FDA/PhUSE collaboration, which consisted of an intro and overview by Chris Decker (D-Wise) and Chuck Cooper (FDA). We then split the audience into three subgroups. Presentations and discussions were held on three of the 6 working groups, including ours, working group 5, which was led by a combination of Kevin Kane, Dirk Spruck and Sascha Ahrweiler.
We had about 70 interested attendees for the discussion about WG5. We had a lot of interest in both the progress that has been made until now, and the future plans. A lot of interesting questions and issues were raised, and Sascha recorded these. A list with e-mail addresses of interested people to support WG5 were alreay added to Listbos and sent to the WG leads.
The following areas were covered during the discussions:
- Update and discussion about subgroup discussing the "What Scripts?" topics
- Update and discussion about subgroup discussing the "Development and Validation" topics
- Update and discussion about subgroup discussing the "Platform" topics
"What Scripts?" topics
Since no CSS WG5 subgroup member attended the discussion in Budapest only high level updates were given. The attendees were informed that there are several white scripts in preparation and available for review.
The following topics wer brought up
- Why do we limit ourselves to complete analysis? We also could align first on standard ways for deriving variables (e.g. calculation of relative days, LOCFs, etc.) and use these than in the analysis later. This would have the big advantage that we can build scripts in a modular way.
"Development and Validation" topics
Kevin Kane gave an update of the current status of this subgroup. Attendees jumped in with questions and proposals during the update.
The following issues were discussed:
- Questions came up about the source for these scripts. The whol group agreed that both SDTMs and ADaMs should be covered.
- Questions came up how scripts are approved and updated. A rating system would be preferred to rate different scripts which serve the same purpose.
- We should have monthly or quarterly requests for script development. That would give the whole group a well defined target for development.
- It was discussed that script development should not be restricted to R and SAS. Also other models for development should be considered.
- Development of scripts should be developed on a continuous bases and therefore improved over time.
- Are interactions between scripts allowed and how can they be managed?
- Why do we plan to validate these scripts at all? A better approach would be to test the scripts and name the test cases/databases on which these scripts were tested. In case users use the scripts for other cases as well they could amend the testcases and increase the usage of the scripts. There was a lot of agreement on this proposal. The V-model was considered to be outdated and not appropriate for this type of collaborative effort.
- A proposal was made top built up databases to perform tests as mentioned above. These testdatabases could be built by using different trial designs in order to meet various user needs
- We should only validate pieces of scripts. Users should be able to access these scripts like validated versioned scripts.
Ian Fleming gave an update on the current status of the subgroup working on the platform.
The following topics were covered during the discussion:
- Everyone agreed with the basics of the platform search: good usable GUI, version control, non proprietory
- Everyone agreed that GitHub is too advanced and not appropriate for this collaborative effort
- The Wiki will be used for publishing of "final" scripts. For the script development the Wiki was considered to have some disadvantages (e.g. no concurrent editing)
- Google Docs is currently the preferred platform as a colloaboration environment fo script development. Several attendees reported bad experience with wrong updates of Google Docs though. A document management system was mentioned as a better platform.
- Look into the development for R as a model for script development.
- The following alternatives were mentioned during the discussions:
- Most of the attendees agreed that the usage of standards scripts by both the FDA and the Industry is a big advantage. Both parties know what results will be when using standard scripts.
- The collaboration with CTSPEDIA was mentioned and considered to be very helpful.
The following topics were brought up during the discussion and require further discussion with the WG:
- Did we set the bar too high? It seems as if too big steps could prevent people from contribution
- Lack of communication is a continuous issue in our WG. Some of this is based on Listbox issues.
- Questions about well defined target for deliveries came up.
- Why don't we grab the low hanging fruits first and google for code which we publish on the Wiki?
This is the current stage of the discussion. You are welcome to visit the Wiki and add your comments, suggestions and improvements to it to keep the content up to date and the discussion alive. Many thanks
Last revision by Dirk Spruck,10/23/2012