Scribd Logo
Upload

Welcome to Scribd!

  • 268 views

    Todd Mike CDISC SDTM Mapping 2009

    Uploaded by

    Barath R Baskaran
    Drug information association, DIA and DIA logo are registered trademarks or trademarks of DIA. Current SDTM mapping methodology is well-established but limited. We need a fully automated expert system to convert clinical trial data to SDTM.

    Copyright:

    Attribution Non-Commercial (BY-NC)
    Download as PDF, TXT or read online from Scribd

    Todd Mike CDISC SDTM Mapping 2009

    Uploaded by

    Barath R Baskaran
    268 views33 pages
    Drug information association, DIA and DIA logo are registered trademarks or trademarks of DIA. Current SDTM mapping methodology is well-established but limited. We need a fully automated expert system to convert clinical trial data to SDTM.

    Copyright

    © Attribution Non-Commercial (BY-NC)

    Available Formats

    PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    Drug information association, DIA and DIA logo are registered trademarks or trademarks of DIA. Current SDTM mapping methodology is well-established but limited. We need a fully automated expert system to convert clinical trial data to SDTM.

    Copyright:

    Attribution Non-Commercial (BY-NC)
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    268 views33 pages

    Todd Mike CDISC SDTM Mapping 2009

    Uploaded by

    Barath R Baskaran
    Drug information association, DIA and DIA logo are registered trademarks or trademarks of DIA. Current SDTM mapping methodology is well-established but limited. We need a fully automated expert system to convert clinical trial data to SDTM.

    Copyright:

    Attribution Non-Commercial (BY-NC)
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    of 33

    SDTM Mapping: Current Technology and Expert Systems

    Michael Todd and Thomas Jablonski


    Nth Analytics

    Disclaimer
    The views and opinions expressed in the following PowerPoint slides are those of the individual presenter and should not be attributed to Drug Information Association, Inc. (DIA), its directors, officers, employees, volunteers, members, chapters, councils, Special Interest Area Communities or affiliates, or any organization with which the presenter is employed or affiliated. These PowerPoint slides are the intellectual property of the individual presenter and are protected under the copyright laws of the United States of America and other countries. Used by permission. All rights reserved. Drug Information Association, DIA and DIA logo are registered trademarks or trademarks of Drug Information Association Inc. All other trademarks are the property of their respective owners.
    2

    Introduction
    Current SDTM mapping methodology is well-established but limited
    Many companies use some version of metadata-driven ETL mapping system However, it requires a mapping expert to define the metadata
    the number of experts is limited

    We need a fully automated expert system to convert clinical trial data to SDTM on a massive scale
    3

    Current Technology

    CDISC Implementation Goals


    Design a strategy such that:
    No knowledge needed of system that originally produced the legacy data Applicable to files from any system Implementation is flexible enough to adapt to different study designs Minimal programming support required for maintenance Reasonable cost
    5

    Implementing an ETL Process


    Programs read table-driven metadata to translate the analysis data into SDTM formats
    Tells the SAS code which analysis variables populate the SDTM variables Indicates when specialized code is required

    All code is developed to be generic using the metadata to indicate when variations are required New studies only require changes to metadata
    6

    ETL Transformation Process


    Define how raw/analysis data fits into SDTM domains and variables Match data to required, permitted and expected SDTM data when possible Provide an automated mechanism for specifying the data sources and algorithms
    Metadata for the SDTM files Basis for the FDA-mandated DEFINE.XML documentation
    7

    Sample SDTM Metadata

    Process Without Automation


    Project 1
    Data 1 Data 1

    Data
    Data 1

    Project Programs

    Reports

    Project 2

    Data etc.
    9

    Project Programs

    Reports

    ETL Process

    10

    Why it Works
    Role of standards
    Standards drive the process. Target has standard structure so can be standardized. While source variables differ, commonalities can be exploited

    Knowledge required
    CDISC Standards Understanding of raw data issues Study design Limited derivation
    11

    Tasks and Job Roles


    Task
    Annotated CRF SDTM mapping specifications Create trial design datasets Development of conversion jobs in ETL Environment QC of the SDTM files Statistician Data Integration Specialist QC Specialist

    Job Role Requires


    Mapping Specialist In-depth knowledge of SDTM V3.1.1 IG Expert knowledge of clinical data Ability to translate abstract concepts into datasets SAS programming Knowledge of ETL tool In-depth knowledge of SDTM V3.1.1 IG Expert knowledge of clinical data In-depth knowledge of system, SDTM, and clinical data

    System maintainance Project-level metadata/macros

    Technical Lead
    12

    Dataflow
    SDTM V3.1.1 Guide Protocol Statistical Analysis Plan Raw Data Annotated CRF Raw Data Data Integration Specialist Mapping Specialist SDTM Data Annotated CRF Trial Design Datasets: TE, TA, TV, TI, TS SDTM Datasets and SAS Programs
    SE.XPT SE.SAS

    Statistician
    SV.XPT SV.SAS

    Study-Level Metadata ETL System Study-Level SAS Macros

    AE.XPT

    AE.SAS

    SUPPAE .XPT

    SUPPAE .SAS

    ...
    Project-Level Metadata Project-Level SAS Macros

    CM..XPT

    CM.SAS

    Technical Lead

    SUPPCM .XPT

    SUPPCM .SAS

    DM..XPT

    DM.SAS

    ...

    DEFINE.XML Stylesheet (from CDISC)

    DEFINE.XML

    13

    Limitations
    Requires experts Severely limited throughput, relative to amount of clinical trial data Converting legacy data on a systemic scale is infeasible

    14

    Future Directions

    15

    Requirements Going Forward


    Without legacy data, goals of meta analyses, etc. will be limited and incomplete In order to effectively use SDTM, the FDA warehouse must include all data for a compound, not just new data going forward Converting legacy on this scale is simply infeasible with current techniques.
    16

    Challenge
    Convert unstructured information such as text into relational tables that can be used to generate code to create SDTM & DEFINE.XML To create this system, imagine thinking like a computer.
    You have sources of information You have a set of rules You have a storage of knowledge available. Apply heuristics to create SDTM datasets with a certain probability of accuracy.
    17

    Sources of Information
    Data
    Main source of information Can assume data exists, while protocol & CRF may not for legacy studies.

    CRF
    Usually this is an image, can it be processed?

    Protocol and Study Report


    Possibly use text-mining techniques to extract information to help organize data
    18

    Why SDTM is Amenable to an Expert-System Approach


    SDTM represents a well-defined, rule-based structure
    Expert system assumptions rely on well-defined structure and meaning Assume that data have organization and meaning, however hard to determine

    Certain things make it easier


    Assume characteristics for clinical data, as opposed to exponentially more possibilities for any arbitrary data Limited set of target SDTM domains
    19

    Role of the SDTM Expert


    Recognize the kinds of data coming in Redistribute data to SDTM
    Have to recognize the type of data without being told what it is How do you know something is lab data, if you cant rely on variable names and labels? How would a machine recognize the type of data just by the structure and values?

    20

    Role of the Expert (continued)


    Expert can recognize tests even if the data are not labeled
    Experts can differentiate the data Under a well-defined set of rules, data has logical and mathematical place to be.

    System must handle tests may not exist today, but would still fit into findings, events, or interventions.
    21

    Examples of Reliable Assumptions


    Data are in English Each dataset contains the same type of data
    AEs and conmeds are not in the same dataset

    Each dataset contains keys: variable(s) that enable datasets to be joined together Dates and times have a sequence
    Discoverable by sorting
    22

    How to Think Like an Expert (Machine)

    23

    Identifying Dataset HMZ11

    What is this? How do we know?


    No obvious visit or timing variable, other than C3 C4 and C5 may be controlled terminology C1 and C2 look like keys
    24

    Identifying C1

    Left-most column often is a protocol Mixture of letters, numbers, and special characters: probably a code No hits for dictionary lookup for meaningful terms If the sponsor is known, there may be a list of protocols for lookup
    25

    Identifying C2

    The same things we noted for C1 also apply to C2. FAB-10 is as likely to be a protocol number as X312. It is only because C1 has the same value for all records in the dataset that we can conclude with a high probability that FAB-10 is a protocol number.
    26

    Protocol and Subject Numbers


    Assume if a possible protocol number has only one value in a dataset, it very likely is a protocol number. If there are two values (FAB-10, FAB-11), possibly the dataset contains results from two protocols. If it contains only one value in multiple datasets, this boosts our confidence in it being a protocol number If the list of subject numbers is consistent across datasets, we can assume with more certainty they are subject numbers We assume datasets contain the same subjects, for the most part

    27

    Identifying C3

    Is it a sequence number?
    C3 contains only integers Increasing series from 1 to n with some gaps and some ties Most subjects have the same number of records Implies series of checkboxes on CRF, preprinted choices If we select distinct C2, C3, there should only be one record for each combination..
    28

    Identifying C4

    Is this a list of body systems?


    Terms would match known systems in a dictionary lookup for body systems We would expect most subjects to have the same terms Should be controlled terminology Usually corresponds to the sequence number (C3), although not always
    29

    Identifying C5

    If C4 is the body system, is C5 the status?


    We assume that if there is a body system, there should be a result for that system Terms appear to be a finite set, implying controlled terminology HISTORY / NOT ACTIVE and CURRENTLY ACTIVE suggest medical history
    30

    Identifying C6 and C7

    If C4 is the body system and C5 the status, the remaining columns probably are verbatim descriptions
    There are several disease-related words Appears to be verbatim text Unclear why there are multiple columns of information.
    Probably legacy data structure with each description in a separate column. 31

    Definitive Identification

    If we cheat and look at the CRF, it is obviously Medical History


    32

    Summary
    Current SDTM mapping technology depends on experts Severely limits throughput relative to all legacy data needed for a comprehensive clinical trial database A fully automated expert system that can perform SDTM conversions with a high probability of accuracy is a promising approach.
    33

    You might also like