DCDISC Users Group Nate Freimark Omnicare Clinical Research Presented on 2011-05-12 1
Disclaimer The opinions provided are solely those of the author and not those of the ADaM team or Omnicare Clinical Research 2
What is ADaM? Datasets created to support analyses analysis-ready Rich with metadata support DEFINE file Computational methods document/reviewer s guide Results-metadata also a component Elements for DEFINE identified DEFINE schema does not yet support Can be produced in non-xml fashion 3
ADaM vs SDTM We know in CDISC there is SDTM and ADaM Some think of SDTM as collected data and ADaM as analysis/derived data Distinction not 100% accurate and lines may get more blurred over time SDTM has many derived components AGE (unless collected on CRF) RFSTDTC/RFENDTC --DY (study day anchored by RFSTDTC) Patient Populations (SUPPDM) Rows added with DRVFL=Y (total score in QS) 4
Why Not Just Use SDTM? Components not gathered in one place Populations Treatment values Values from multiple domains (AEs by baseline APACHE score) Fixed and Rigid Structure Does not allow addition of supportive variables Does not allow for all transformations of values (categorization (RESCAT has been added but only allows one ), groupings, etc.) Unless SUPP is used as primary storage space However - Additions in SUPPs which are not 100% user-friendly Not a lot of controlled terminology huge learning curve Values can only be captured as character and not numeric No standard tools to put things together 5
Why Not Just Use SDTM? (cont.) Analysis Components not part of standard Visit windows Record selection flags Derivation methodology captured as free text Supportive variables have to be supp ed Metadata for SUPPs are not user-friendly 6
Simple Summary From SDTM Age Univariate by Treatment Group Safety Population Age from DM Treatment Group ARM from DM QNAM=SAFETY, QVAL=Y from SUPPDM Easy to create and reproduce if standard tool is available to put DM and SUPPDM together 7
Less Simple Summary From SDTM Disposition by Treatment Group Safety Population Completion Status from DS Treatment Group ARM from DM QNAM=SAFETY, QVAL=Y from SUPPDM Requires: Subsetting of DS records Merge of DM/SUPPDM Merge of DS and DM Assumes all subjects have a disposition record does not account for ongoing subjects 8
Slightly Complicated Summary From SDTM Vital Signs Summary by Analysis Visit and Treatment Group Vital Sign Values from VS Treatment Group ARM from DM QNAM=SAFETY, QVAL=Y from SUPPDM Requires: Windowing of VS records per SAP Data or metadata to support windowing decisions Selection decision if more than one record in window Merge of DM/SUPPDM Merge of VS and DM 9
So Now What? We ADaM! Decide which ADaM datasets to create Gather source data Gather source documentation Create specifications Create ADaM datasets QC datasets Create DEFINE (.doc,.pdf,.xml,html depending on requirements) Outputs and documentation we will leave for another time 10
SDTM/ADaM Relationship How do they interact? Creation Process SDTM -> ADaM ADaM Superdataset -> SDTM and ADaM ADaM -> SDTM Documentation ADaM has to point back to SDTM Reconstruction/consistency is an issue that has to be considered 11
Which ADaM Datatsets Do I Create? ADSL required Anything more is up to sponsor and reviewer ADaM Model Document: consist of the optimum number of analysis datasets needed and have enough self-sufficiency to allow analysis and review with little or no additional programming or data processing " Some possibilities: Support only selected summaries Key safety and efficacy Support all summaries Support all outputs 12
Where Does ADaM Come From? Source data and documentation: ADaM Model has ADaM datasets built from SDTM domains as the default methodology Even if SDTM is not source for ADaM, ADaM metadata has to indicate how to create ADaM from SDTM External input: Data that may not be captured in SDTM (dictionaries, IVRS variables, Evaluation Committee data) Decision spreadsheets Disallowed medications Sponsor over-rides 13
Where Does ADaM Come From? (cont.) Documents: Protocol SAP Table shells Evaluability plan (definitions for populations and end-point determinations with supportive review outputs) SDTM define 14
Analysis Dataset Specifications Everyone has their own methodology One method is to create excel spreadsheets Provides a priori direction Dataset/variable names attributes algorithms Captures post facto documentation Special rules implemented during dataset creation Decision spreadsheet values, etc. Can be used as input into Define creation Populate source/derivation column Provide content for computational method document 15
What Goes Into ADaM? ADSL Subject level population flags Planned and actual treatment variables Demographic information Randomization factors Subgrouping variables Important dates 16
What Goes Into ADaM? (cont.) BDS (Basic Data Structure) PARAM/PARAMCD (required) AVAL/AVALC (at least one is required) other analysis variables (CHG, AVALCAT1) by group variables (AGEGRP1, SEX) Selection flags (ITTRFL, ANL01FL) Treatment variables (TRTP, TRTA) Can be grouped by topic Labs, vitals, time-to-event parameters 17
More ADaMs Standards: ADAE post-public review ADTTE post-public review Some other possibilities: ADEFF ADLB ADRESP No current controlled terminology (besides for ADSL) 18
How Do You Evaluate ADaM Datasets? Fundamental Principles : Analysis datasets and their associated metadata must: facilitate clear and unambiguous communication provide traceability between the analysis data and its source data (ultimately SDTM) be readily useable by commonly available software tools Analysis datasets must: be accompanied by metadata be analysis-ready 19
How Do You Evaluate ADaM Datasets? Some principles are testable and some are not Does data and metadata allow a user to reproduce the results? Does data comply to the standard? ~180 checks defined to measure compliance Metadata for ADaM does not currently fit into define schema (parameter identifier and results metadata) Define.xml for ADaM can be created using SDTM schema but is not required 20
How Do You Document ADaM Databases? ADaM Metadata Dataset metadata Variable metadata Parameter-level metadata Results level metadata 21
Dataset Metadata 22
Dataset Metadata example 23
Variable Metadata 24
Variable Metadata example 25
Parameter-Level Metadata 26
Results Level Metadata 27
28
Results Level Metadata example 29
Does Anything From ADaM Go Back Into SDTM? Patient Populations Can go seamlessly from ADSL into SUPPDM Allows for the creators to create and the users to use Input into population creation needs a process and is a subject for another day Complicated Endpoint Values Capturing in ADaM only means it is not available with SDTM data Capturing as an SDTM data fact is possible but traceability/documentation becomes a hunting exercise Very careful process has to be followed to ensure linear process and no circular logic is followed 30
What Next? Posted Compliance checks 180 checks OpenCDISC has created ADaM Validator ADAE Draft posted >450 comments received ADTTE Draft posted >150 comments received Near future General Examples Metadata 31
Questions? Thank you all for attending!! 32
Source material: ADaM Model Document V2.1 ADAMIG V1.0 Nate.freimark@omnicarecr.com 33