ADaM Implementation Guide Status Update John K. Troxell John Troxell Consulting LLC Bridgewater, NJ jktroxell@gmail.com June 17, 2013
Current CDISC ADaM Documents 2009 Analysis Data Model (ADaM), Version 2.1 Analysis Data Model (ADaM) Implementation Guide, Version 1.0 2010-2012 CDISC ADaM Validation Checks, Version 1.2 2011 Analysis Data Model (ADaM) Examples in Commonly Used Statistical Analysis Methods 2012 Analysis Data Model (ADaM) Data Structure for Adverse Event Analysis, Version 1.0 The ADaM Basic Data Structure for Time-to-Event Analyses, Version 1.0 IG compliance update of CDISC-FDA Pilot I 2013-06-17 John Troxell - ADaM IG Status Update 2
ADaMIG to Date ADaM team started working on IG in 2006 Concentrated effort in 2009, published IG in Dec. Multiple work streams 2010-2012 Published Validation Checks, Examples in Stat. Analysis, AE, TTE, compliance update of CDISC-FDA Pilot I Worked on metadata guide, general occurrences, integration, multivariate, PK, contributed to Define 2.0 Started to refocus on IG in 2H 2012 Learning from implementation experience Recently dedicating team meetings to IG, doubled frequency 2013-06-17 John Troxell - ADaM IG Status Update 3
ADaMIG Update Primarily driven by Cathy Barrows, GSK (retired) minor assistance from colead John Troxell Compiled list of suggestions/issues/deferred items Triaged: reject, do first, do later IG 1.1 draft for public comment targeted 2013* typos, error fixes clarifications (requires thought and discussion) some relatively easier issues and enhancements (but really none of them are easy and there is lots of deep thought and debate) IG 1.2 targeted 2014+* issues requiring deepest thought and extended debate more extensive document restructuring * Guaranteed by the ADaM Team or double your money back. 2013-06-17 John Troxell - ADaM IG Status Update 4
Processing of Implementation Experience Clarify, clarify, clarify The scrutiny of the text by implementers is intense Impossible to be too precise Implementers misunderstandings Re-examine original thinking, clarify Objections to constraints Re-examine fundamental aspects of the model or data structure Can we ring-fence a proposed relaxation or exception without breaking the model and inviting chaos? Deep thought process and debate Additional features desired Debate, consider impacts, design Typos, errors Fix, considering backwards compatibility 2013-06-17 John Troxell - ADaM IG Status Update 5
Types of Clarification 1. We agree on what we meant and we need to express it more clearly. 2. We agreed on some language but now we realize we understood it differently. We need to agree on the meaning before we can clarify the text. 3. We realize we had not thought things through well enough to begin with. We need to clarify our thinking first. We may need to make some changes here and/or elsewhere in the standard, as well as clarify the text. 2013-06-17 John Troxell - ADaM IG Status Update 6
Example of a Needed Clarification: 1:1 Map of AVAL:AVALC QS1 is a BDS parameter from a standard questionnaire. QSSTRESC contains the coded value of the answer. In SDTM In ADaM QSORRES QSSTRESC QSSTRESN AVISIT PARAMCD AVAL AVALC DTYPE VERY BAD 1 1 Week 8 QS1 1 VERY BAD BAD 2 2 Week 8 QS1 2 BAD GOOD 4 4 Week 8 QS1 4 GOOD We copy the answer from QSORRES into AVALC. We copy (or derive) the score and put it into AVAL. Within QS1, AVAL and AVALC map 1:1 as required by IG 2013-06-17 John Troxell - ADaM IG Status Update 7
Example of a Needed Clarification: 1:1 Map of AVAL:AVALC Per the SAP, we need to analyze the average Score for Week 8, so we derive a record for it. In SDTM In ADaM QSORRES QSSTRESC QSSTRESN AVISIT PARAMCD AVAL AVALC DTYPE VERY BAD 1 1 Week 8 QS1 1 VERY BAD BAD 2 2 Week 8 QS1 2 BAD GOOD 4 4 Week 8 QS1 4 GOOD Week 8 QS1 2.333 AVERAGE This looks good so far still 1:1 but what happens when we add more data.? 2013-06-17 John Troxell - ADaM IG Status Update 8
AVAL:AVALC In SDTM In ADaM QSORRES QSSTRESC QSSTRESN AVISIT PARAMCD AVAL AVALC DTYPE VERY BAD 1 1 Week 8 QS1 1 VERY BAD BAD 2 2 Week 8 QS1 2 BAD GOOD 4 4 Week 8 QS1 4 GOOD Week 8 QS1 2.333 AVERAGE OK 3 3 Week 8 QS1 3 OK VERY BAD 1 1 Week 8 QS1 1 VERY BAD BAD 2 2 Week 8 QS1 2 BAD Week 8 QS1 2.0 AVERAGE Now we don t have a 1:1 correspondence between AVAL and AVALC: AVAL=2.333 / AVALC=null AVAL=2.0 / AVALC=null This is a topic for clarification by the ADaM team. 2013-06-17 John Troxell - ADaM IG Status Update 9
Example of a Needed Clarification: 1:1 Map of AVAL:AVALC Solution: clarify the scope of the 1:1 mapping requirement Clarify that AVAL and AVALC must be a one-to-one map within a parameter on the rows on which both AVAL and AVALC are populated We are making many scope clarifications in IG 1.1 A statement applies within a study? dataset? parameter? subject? Often this was unstated in IG 1.0. 2013-06-17 John Troxell - ADaM IG Status Update 10
Common Misunderstandings and Impacts on Next Versions of IG Sometimes we observe people (including us) misunderstanding the IG or authors intent Misunderstandings point to places where the IG can be improved See Common Misunderstandings about ADaM Implementation, PharmaSUG 2012. N. Freimark, S. Kenny, J. Shostak, J. Troxell a snapshot of personal opinions of the authors not a complete list 2013-06-17 John Troxell - ADaM IG Status Update 11
Example of a Misunderstanding: Mistaking PARCATy as a Qualifier Parameter PARAM has no qualifiers Parameter Category PARCATy groups parameters into categories PARCATy does not subdivide or qualify PARAM ADaM Validation Check Error: There is more than one value of PARCATy for a given value of PARAMCD. Some have misunderstood PARCATy to be a qualifier of PARAM Misunderstanding often results from carrying over SDTM thinking ADaM SDTM PARCATy --CAT Basic Data Structure Findings Domain + AVAL/AVALC --STRESN/--STRESC 2013-06-17 John Troxell - ADaM IG Status Update 12
Example of a Misunderstanding: Mistaking PARCATy as a Qualifier Compliant but not scalable: PARAM AVAL Cumulative Dose of Drug A (mg) 1,234 Cumulative Dose of Drug B (mg) 8,765 Non-compliant but scalable: PARAM PARCAT1 AVAL Cumulative Dose (mg) Drug A 1,234 Cumulative Dose (mg) Drug B 8,765 For a company core controlled terminology for PARAM, it would be nice if we could have one parameter, not dozens. SO, wouldn t it be nice to misunderstand PARCATy? 2013-06-17 John Troxell - ADaM IG Status Update 13
Example of a Misunderstanding: Mistaking PARCATy as a Qualifier How should ADaM IG be modified? Clarify and emphasize that PARCATy is not a qualifier on PARAM and re-emphasize that PARAM has no qualifiers? Find a carefully ring-fenced way to permit limited qualifier functionality without permitting degeneration to SDTM TEST + qualifiers approach? Is this even possible? Deeper question than just PARCATy Need to address one way or the other 2013-06-17 John Troxell - ADaM IG Status Update 14
What you really wanted to hear about: Exactly what is changing? Not appropriate to publicize details until release for public comment Still evolving Some clues about 1.1 corrections of typos and minor errors clarifications (e.g. scopes, definitions, text) additional timing and other variables some features useful for oncology more Word tables minor restructuring of document 1.2 more profound issues addressed more restructuring coordinated model document update 2013-06-17 John Troxell - ADaM IG Status Update 15
Energy Around ADaM? Join Us! 2013-06-17 John Troxell - ADaM IG Status Update 16