11

The CDISC-HL7 StandardThe CDISC-HL7 StandardAn FDA PerspectiveAn FDA Perspective

Armando Oliva, M.D.Armando Oliva, M.D.President, Avilo Consulting LLCPresident, Avilo Consulting LLC

Former Deputy Director for BioinformaticsFormer Deputy Director for BioinformaticsOffice of Critical Path ProgramsOffice of Critical Path ProgramsOffice of the CommissionerOffice of the Commissioner

22

OutlineOutline Data standards in clinical researchData standards in clinical research What does CDISC offer?What does CDISC offer? What does HL7 offer?What does HL7 offer? How the CDISC-HL7 project will How the CDISC-HL7 project will

benefit regulatory review of clinical benefit regulatory review of clinical researchresearch

Implementation IssuesImplementation Issues

33

Take Home MessagesTake Home Messages

““The World is Round”The World is Round” • Clinical data are not flat and cannot be exchanged using Clinical data are not flat and cannot be exchanged using

flat two-dimensional files without significant loss of flat two-dimensional files without significant loss of meaningmeaning

FDA is transitioning to a “round view of the FDA is transitioning to a “round view of the world” of clinical researchworld” of clinical research• CDISC-HL7 standard will get us thereCDISC-HL7 standard will get us there

SDTM is here to staySDTM is here to stay• Will transition from a standard submission format to an Will transition from a standard submission format to an

standard view of the data in support of simple analyses standard view of the data in support of simple analyses (e.g. distribution, means, etc.)(e.g. distribution, means, etc.)

44

Data ExchangeData Exchange

Definition:Definition: Data exchange is the process of Data exchange is the process of

sending and receiving data in such a sending and receiving data in such a manner that the information content manner that the information content or or meaningmeaning assigned to the data is assigned to the data is not altered during the transmission. not altered during the transmission.

55

The ProblemThe Problem

Most clinical trials …Most clinical trials …• don’t employ a standard for data exchangedon’t employ a standard for data exchange• don’t use standardized analytic tools or techniques don’t use standardized analytic tools or techniques

Result: Result: • Analyzing clinical trial data efficiently and Analyzing clinical trial data efficiently and

systematically is difficult and time consuming, systematically is difficult and time consuming, especially across many trials especially across many trials

• e.g. How many women participate in clinical trials? e.g. How many women participate in clinical trials?

66

Study #1 – demog.xptStudy #1 – demog.xpt

SUBJIDSUBJID SEXSEX00010001 MM00020002 FF00030003 FF00040004 MM00050005 FF

IDID GENDERGENDERA1A1 MaleMaleA2A2 MaleMaleA3A3 FemaleFemaleA4A4 FemaleFemaleA5A5 MaleMale

Study #2 – dmg.xptStudy #2 – dmg.xpt

USUBIDUSUBID SEXSEX0001100011 000001200012 110001300013 110001400014 000001500015 11

PTIDPTID GENDERGENDER00010001 1100020002 1100030003 2200040004 2200050005 11

Study #3 – axd222.xptStudy #3 – axd222.xpt

Study #4 – dmgph.xptStudy #4 – dmgph.xpt

77

How do you automate counting women?How do you automate counting women? No standard file namesNo standard file names

• How does a computer find the file that contains How does a computer find the file that contains the data?the data?

No standard variable namesNo standard variable names• How does a computer find the column that How does a computer find the column that

contains sex information?contains sex information? No standard terminologyNo standard terminology

• How does a computer know which code How does a computer know which code represents which sex?represents which sex?

88

NET RESULT: NET RESULT: Analysis of study data is . . .Analysis of study data is . . .

Difficult

Time Consuming

Expensive

99

CDISCCDISC

CClinical linical DData ata IInterchange nterchange SStandards tandards CConsortiumonsortium Non-profit standards development organizationNon-profit standards development organization Standards for clinical trial data for submission (SDTM), for Standards for clinical trial data for submission (SDTM), for

collection (CDASH), derived data for analysis (ADaM), and collection (CDASH), derived data for analysis (ADaM), and non-clinical data for submission (SEND)non-clinical data for submission (SEND)

CDER adopted CDISC SDTM 7/04, now piloting SENDCDER adopted CDISC SDTM 7/04, now piloting SEND CBER evaluating SDTM in a pilot settingCBER evaluating SDTM in a pilot setting CVM intends to pilot SEND for animal dataCVM intends to pilot SEND for animal data

1010

Example: Women in Clinical TrialsExample: Women in Clinical Trials

Demographics File: DM.xptDemographics File: DM.xpt Variable Name: SEXVariable Name: SEX Acceptable Codes: M, F, UAcceptable Codes: M, F, U

1111

Better Tracking of Women In Clinical TrialsBetter Tracking of Women In Clinical Trials

Applicants submit standardized Applicants submit standardized (CDISC) clinical trial data and are (CDISC) clinical trial data and are loaded into Janus* loaded into Janus*

New analysis tools (e.g. New analysis tools (e.g. ‘demographics counter’) can quickly ‘demographics counter’) can quickly and easily extract demographic and easily extract demographic information from all trials in the information from all trials in the repository and analyze the data. repository and analyze the data.

*Janus: FDA’s enterprise data warehouse for study data

1212

Standardized Data Flow:Standardized Data Flow:

sponsor

CDISC Datavia ESG/cEDR

Janus

Custom Analytic Tool:“Demographics Counter”

How manyWomen?

1313

Sample CDISC “DM” DatasetSample CDISC “DM” Dataset

1414

Standard Report on “DM” Standard Report on “DM”

FM

SEX

40504506

N

47.2952.71

% of Total

67.383456867.4962273

Mean

5959

Min

8080

MaxAGE

737833

ASIAN

4042

BLACK

186237

HISPANIC

7667

OTHER

30113327

WHITERACE

Created using JMP

1515

With the right bioinformatics With the right bioinformatics infrastructure (data standards, data infrastructure (data standards, data warehouse, analysis tools):warehouse, analysis tools):

Monitoring women in clinical trials is easy!Monitoring women in clinical trials is easy! Assessing sex differences is almost as Assessing sex differences is almost as

easyeasy

1616

CDISC is a tremendous advance!!CDISC is a tremendous advance!!

Big improvement from the pastBig improvement from the past Easy to find the dataEasy to find the data Easy to understand the dataEasy to understand the data Easy to analyze the dataEasy to analyze the data

1717

But…But…

We’ve learned some lessons along the way…We’ve learned some lessons along the way…

““Flat” 2-dimensional files are not the best Flat” 2-dimensional files are not the best way to exchange clinical dataway to exchange clinical data

Some Some meaningmeaning is lost when exchanging flat is lost when exchanging flat files, making certain analyses difficult or files, making certain analyses difficult or impossibleimpossible

1818

1919

A flat map is usefulto map a route fromWashington to Texas.

2020

Washington, DC

London

But a flat map of the North Atlantic cannot be used to map the shortest flight path from London to Washington. In the same way, we need to recognize that flat 2-dimensional clinical trial data files, though very useful for certain analyses, can’t be expected to support all analyses that FDA wants to perform.

2121

Clinical Observations: Clinical Observations: Highly Relational and HierarchicalHighly Relational and Hierarchical

Observation#1

Observation#2

Observation#3

Observation#4

Observation#5

Observation#6

•We are currently not capturing these relationships well.

Planned Observation

#1

2222

Example 1 (from my clinic)Example 1 (from my clinic)

AE=Tremor

LAB=Thyroid Test

(normal)

EEG=(abnormal) Fam Hx=Tremor

RX=Stop DepakoteStart Neurontin

AE=TremorResolved

RX=Depakote

Conclusion: Tremor reasonably due to Depakote

2323

Think of Clinical Observations as Think of Clinical Observations as nested folders in a tree structurenested folders in a tree structure

2424

Flat Files Don’t Flat Files Don’t Inherently Capture Inherently Capture the Tree Structure, the Tree Structure,

which is itself which is itself important to important to

understand the understand the datadata

2525

One approach: capture the relationships separately One approach: capture the relationships separately and add them to the flat file structureand add them to the flat file structure

Better approach: data model that inherently captures relationships at the point of collection and can transmit them.

2626

Example 2Example 2 Study drug: known association with Study drug: known association with

thrombocytopenia in nonclinical studiesthrombocytopenia in nonclinical studies Planned clinical assessment: complete blood Planned clinical assessment: complete blood

count every 3 monthscount every 3 months

Consider two clinical observations, both Consider two clinical observations, both observed the same day – reported in flat files observed the same day – reported in flat files as:as:

AE: BLEEDINGAE: BLEEDING LB: PLATELETS = 35K (LOW)LB: PLATELETS = 35K (LOW)

These observations can represent two clinical These observations can represent two clinical scenarios.scenarios.

2727

Two ScenariosTwo Scenarios

#1 - Subject arrives for a planned #1 - Subject arrives for a planned study visit and laboratory test: study visit and laboratory test: complete blood count (CBC). Results complete blood count (CBC). Results show low platelets 35K. Investigator show low platelets 35K. Investigator asks about abnormal bleeding asks about abnormal bleeding episodes. Patient recalls a mild episodes. Patient recalls a mild nosebleed earlier that morning. nosebleed earlier that morning.

Hypothesis: Monitoring Hypothesis: Monitoring schedule is adequate (?)schedule is adequate (?)

#2 – Subject calls investigator to #2 – Subject calls investigator to report a mild nosebleed. Investigator report a mild nosebleed. Investigator advises subject to undergo advises subject to undergo unscheduled complete blood count. unscheduled complete blood count. CBC shows low platelet count of 35K. CBC shows low platelet count of 35K.

Hypothesis: Monitoring Hypothesis: Monitoring schedule is inadequate; more schedule is inadequate; more frequent CBC monitoring is frequent CBC monitoring is needed (?)needed (?)

AE=Nosebleed

AE=Nosebleed

LB = dec platelets

Planned LB = platelets count

Unplanned AEassessment

LB = dec platelets

2828

Review Questions?Review Questions? How many unplanned CBCs occurred?How many unplanned CBCs occurred? What observations led to unplanned CBCs?What observations led to unplanned CBCs? How often is bleeding associated with unplanned How often is bleeding associated with unplanned

CBC? CBC? How often is bleeding associated with planned How often is bleeding associated with planned

CBC?CBC? Is monitoring frequency adequate… ? Is monitoring frequency adequate… ?

• ……For the trial (IND)? For the trial (IND)? • ……For labeling (NDA/BLA)?For labeling (NDA/BLA)?

Current flat file exchange format makes these Current flat file exchange format makes these analyses difficult.analyses difficult.

A more robust data model will facilitate answering A more robust data model will facilitate answering these questions. these questions.

2929

HL7HL7 Health Level 7 is the world’s leading standards Health Level 7 is the world’s leading standards

development organization for healthcare development organization for healthcare information exchange. information exchange.

ANSI accreditedANSI accredited Cooperation agreements with ISO and CEN (the Cooperation agreements with ISO and CEN (the

European Standards Body)European Standards Body) HL7 Version 2 is widely implemented in over 20 HL7 Version 2 is widely implemented in over 20

countriescountries HL7 standards are called “messages” because they HL7 standards are called “messages” because they

support healthcare information exchange between support healthcare information exchange between systems or organizationssystems or organizations

3030

What is the HL7 RIM?What is the HL7 RIM? HL7 Version 3 is based on the “Reference Information Model” HL7 Version 3 is based on the “Reference Information Model”

and has been developed for complex health IT systemsand has been developed for complex health IT systems Over 10 years in developmentOver 10 years in development The fundamental information model from which all HL7 The fundamental information model from which all HL7

messages are basedmessages are based HL7 RIM (V3) has significant international implementation:HL7 RIM (V3) has significant international implementation:

• United Kingdom (National Health Service - NHS)United Kingdom (National Health Service - NHS)• NetherlandsNetherlands• CanadaCanada• MexicoMexico• GermanyGermany• CroatiaCroatia

3131

What does HL7 V3 offer?What does HL7 V3 offer? More “multi-dimensional” or “multi-More “multi-dimensional” or “multi-

relational” representation of the datarelational” representation of the data cost-effective method to exchange healthcare cost-effective method to exchange healthcare

information; less manual processinginformation; less manual processing In the healthcare domain:In the healthcare domain:

• Increase patient safety; fewer clinical errorsIncrease patient safety; fewer clinical errors Interoperability among loosely coupled Interoperability among loosely coupled

systemssystems Ability to collect increasing amounts of coded Ability to collect increasing amounts of coded

data in EHRsdata in EHRs

3232

CDISC-HL7 StandardCDISC-HL7 Standard

What What is it?is it? WhyWhy do we need a CDISC-HL7 do we need a CDISC-HL7

exchange standard?exchange standard? HowHow is the standard being is the standard being

developed?developed? WhenWhen will the standard be available? will the standard be available? … …will FDA implement it?will FDA implement it?

3333

WhatWhat Is It? Is It? A project in HL7 to develop HL7 XML exchange A project in HL7 to develop HL7 XML exchange

standard (“messages”) for CDISC structured standard (“messages”) for CDISC structured study data. study data. • Sponsored by FDA and CDISCSponsored by FDA and CDISC

Four messages:Four messages:• Study Design (What will be done?)Study Design (What will be done?)• Study Participation (Who is involved?)Study Participation (Who is involved?)• Subject Data (What was observed?)Subject Data (What was observed?)• HL7 ICSR (expedited AE reporting)HL7 ICSR (expedited AE reporting)

3434

WhyWhy CDISC-HL7? CDISC-HL7?1.1. Improved Data ManagementImproved Data Management

- Facilitate more analysesFacilitate more analyses- Transition from XPT to HL7 XMLTransition from XPT to HL7 XML- More multi-dimensional representation of the dateMore multi-dimensional representation of the date

2.2. Harmonize with other HL7 standards for Harmonize with other HL7 standards for regulated medical product information: regulated medical product information: Structure Product Labeling (SPL), Individual Structure Product Labeling (SPL), Individual Case Safety Report (ICSR), Case Safety Report (ICSR),

3.3. Better long-term integration with EHRs as Better long-term integration with EHRs as they start being used for both Clinical they start being used for both Clinical Research and SurveillanceResearch and Surveillance

3535

CDISC-HL7 and JanusCDISC-HL7 and Janus Janus – FDA’s enterprise logical data model for Janus – FDA’s enterprise logical data model for

structured scientific data and physical data structured scientific data and physical data warehouseswarehouses

CDISC-HL7 XML will facilitate loading study data CDISC-HL7 XML will facilitate loading study data into the Janus study data warehouseinto the Janus study data warehouse

HL7 ICSR will be used to load post-marketing HL7 ICSR will be used to load post-marketing observational data into enterprise repositoriesobservational data into enterprise repositories

Data checker/loader can be leveraged to work Data checker/loader can be leveraged to work with all HL7 exchange standardswith all HL7 exchange standards

Harmonization of CDISC-HL7 (pre-market) and Harmonization of CDISC-HL7 (pre-market) and HL7 ICSR (post-market) will allow integration of HL7 ICSR (post-market) will allow integration of pre- and post-market databases and facilitate pre- and post-market databases and facilitate analyses across a product’s entire lifecyle. analyses across a product’s entire lifecyle.

Better safety assessmentsBetter safety assessments

3636

Example: Pre- and Post-Market AEsExample: Pre- and Post-Market AEs AE signal identified post-marketingAE signal identified post-marketing Question:Question: How does the AE post-marketing How does the AE post-marketing

experience compare with clinical trials experience compare with clinical trials experience?experience?• Current data standards/systems make this analysis Current data standards/systems make this analysis

extremely difficult/time-consuming to performextremely difficult/time-consuming to perform WhyWhy did clinical trials miss the signal (if that’s did clinical trials miss the signal (if that’s

the case)?the case)? HowHow can future trials / development programs can future trials / development programs

be improved to detect similar problems earlier?be improved to detect similar problems earlier?

3737

JReview(Software Application)

Janus Data PyramidJanus Data PyramidStructured Scientific Data Management SystemStructured Scientific Data Management System

RIM Database

Data Mart n

Exchange Layer

Persistence Layer

Results Layer Results

Database Layer

Data Mart & Special Purpose Layer

RIM Database(Data Warehouse)

ClinicalDatabase

(NCI)

Nonclinical Database(NCTR)

CDISC SEND(Analysis Views)

CDISC ADaM(Analysis Views)

FAERS Database(Analysis Views)

CDISC SDTM(Analysis Views)

SAS(Software Application)

Exchange Layer

Persistence Layer

Analysis Layer

Results Layer Results

HL7 Messages (XML)

Database Layer

Data Mart & Special Purpose Layer

Product InformationDatabase (SPL)Database n

Array Track(Software Application)

FAERS COTSPackage

(Software Application)WebVDME

(Software Application)Analysis

Software n

Data

Managem

entA

nalysis

3838

Better Long-Term Integration with EHRs to Better Long-Term Integration with EHRs to support Clinical Research and Surveillancesupport Clinical Research and Surveillance

Health Care Sponsors/CROs

EHR

EHR

EHR

EHR

EHR

Janus

Clinical ResearchDatabase

Surveillance

ICSR

3939

CDISC-HL7 – Integrating Clinical CDISC-HL7 – Integrating Clinical Research with HealthcareResearch with Healthcare

Now Future

Healthcare(HL7)

Clinical Research(CDISC)

Clinical Research

(CDISC-HL7)

Healthcare(HL7)

4040

HowHow is the standard being developed? is the standard being developed? Fall 2007 – CDISC-HL7 Project launchedFall 2007 – CDISC-HL7 Project launched Development of 4 messages for CDISC Development of 4 messages for CDISC

content: content: • study designstudy design• study participationstudy participation• subject data (Care Record)subject data (Care Record)• HL7 ICSR (expedited AE reporting)HL7 ICSR (expedited AE reporting)

4141

CDISC HL7 Standards DevelopmentCDISC HL7 Standards Development

Based on CDISC standardsBased on CDISC standards Leverages existing HL7 standards (e.g. Leverages existing HL7 standards (e.g.

HL7 ICSR, Clinical Statement CMET*, HL7 ICSR, Clinical Statement CMET*, SPL)SPL)

*CMET – Core Message Element Type (reusable XML across messages)

4242

CDISC-HL7 StandardCDISC-HL7 StandardFour MessagesFour Messages

MessageMessage UseUseStudy DesignStudy Design Study ProtocolStudy Protocol

Study ParticipationStudy Participation Add/changes to participant Add/changes to participant informationinformation

Subject Data (Care Record)Subject Data (Care Record) Study result submissionStudy result submission

ICSRICSR Expedited AE reportingExpedited AE reporting

4343

WhenWhen … … …will the standard be available? …will the standard be available? …will FDA implement it?…will FDA implement it?

4444

*DRAFT**DRAFT* Timeline TimelineActual Timeline will be documented/updated in rolling PDUFA 4 IT PlanActual Timeline will be documented/updated in rolling PDUFA 4 IT Plan

2009 – 2012 Testing, FDA accepts both SDTM and 2009 – 2012 Testing, FDA accepts both SDTM and CDISC-HL7 dataCDISC-HL7 data

2013 and beyond – (PDUFA V?) – CDISC-HL7 XML 2013 and beyond – (PDUFA V?) – CDISC-HL7 XML onlyonly

2014 – EHRs fully deployed nationwide2014 – EHRs fully deployed nationwide

2020 (?) – EHRs fully integrated into clinical 2020 (?) – EHRs fully integrated into clinical researchresearch

4545

Common MythsCommon Myths SDTM is going awaySDTM is going away

• False: SDTM is a very useful standard view of the collected data to support basic False: SDTM is a very useful standard view of the collected data to support basic analyses (e.g. means, S.D., distribution)analyses (e.g. means, S.D., distribution)

• SDTM will transition at FDA from an exchange standard to a standard analysis view that SDTM will transition at FDA from an exchange standard to a standard analysis view that Janus will produceJanus will produce

CDISC-HL7 is just for human drug trialsCDISC-HL7 is just for human drug trials• False: CDISC-HL7 is being designed for studies involving any substance (drug, biologic, False: CDISC-HL7 is being designed for studies involving any substance (drug, biologic,

device, combination) on any organism (human, animal), groups of organisms, part of device, combination) on any organism (human, animal), groups of organisms, part of organisms, and also for studies where the medical product is the subject of the organisms, and also for studies where the medical product is the subject of the experimentexperiment

FDA wants study data submitted as flat filesFDA wants study data submitted as flat files• False: Current flat file submissions evolved from paper “tabulations”False: Current flat file submissions evolved from paper “tabulations”• FDA now recognize the limitations associated with flat filesFDA now recognize the limitations associated with flat files• FDA intends to move to a more ‘multi-dimensional,’ ‘multi-relational’ model for data FDA intends to move to a more ‘multi-dimensional,’ ‘multi-relational’ model for data

submissionssubmissions

CDISC-HL7 is an XML wrapper for SDTM flat filesCDISC-HL7 is an XML wrapper for SDTM flat files• False: CDISC-HL7 is a new relational model for study data that is able to capture and False: CDISC-HL7 is a new relational model for study data that is able to capture and

convey more meaningconvey more meaning• CDISC-HL7 will support the creation of additional views of the data to support more CDISC-HL7 will support the creation of additional views of the data to support more

analysesanalyses• CDISC-HL7 will enable us to answer more scientific questions and make better CDISC-HL7 will enable us to answer more scientific questions and make better

regulatory decisions from study dataregulatory decisions from study data

4646

Implementation ConsiderationsImplementation Considerations FDA mindful of tremendous resources currently expended in creating SDTM / SAS FDA mindful of tremendous resources currently expended in creating SDTM / SAS

Transport FilesTransport Files

Transition plan to CDISC-HL7 needs to consider costs and IT investment lifecycle Transition plan to CDISC-HL7 needs to consider costs and IT investment lifecycle

We expect to continue to accept SAS transport files for quite some time (at least We expect to continue to accept SAS transport files for quite some time (at least thru end of PDUFA4 - 2012?)thru end of PDUFA4 - 2012?)

Updates to this Implementation Plan will be made to the PDUFA4 IT PlanUpdates to this Implementation Plan will be made to the PDUFA4 IT Plan Initial implementationInitial implementation of CDISC-HL7 should be a mapping exercise from SDTM of CDISC-HL7 should be a mapping exercise from SDTM

XPT to CDISC-HL7 XML but full benefit of new exchange format will not yet be XPT to CDISC-HL7 XML but full benefit of new exchange format will not yet be realized (flat XPT realized (flat XPT flat XML) flat XML)

Ultimate implementation takes clinical observations from EHRs as input to CDISC-Ultimate implementation takes clinical observations from EHRs as input to CDISC-HL7 messages round XML in EHRs HL7 messages round XML in EHRs round XML for submission (2014 and round XML for submission (2014 and beyond)beyond)

Full benefitFull benefit of CDISC-HL7 realized when EHRs fully integrated into Clinical of CDISC-HL7 realized when EHRs fully integrated into Clinical Research as the standard data collection instrument for clinical study data, post-Research as the standard data collection instrument for clinical study data, post-marketing AE data, and exposure data. marketing AE data, and exposure data.

4747

Take Home MessagesTake Home Messages

““The World is Round”The World is Round” • Clinical data are not flat and cannot be exchanged using Clinical data are not flat and cannot be exchanged using

flat two-dimensional files without significant loss of flat two-dimensional files without significant loss of meaningmeaning

FDA is transitioning to a “round view of the FDA is transitioning to a “round view of the world” of clinical researchworld” of clinical research• CDISC-HL7 standard will get us thereCDISC-HL7 standard will get us there

SDTM is here to staySDTM is here to stay• Will transition from a standard submission format to an Will transition from a standard submission format to an

standard view of the data in support of simple analyses standard view of the data in support of simple analyses (e.g. distribution, means, etc.)(e.g. distribution, means, etc.)

4848

XPT vs. CDISC-HL7 XMLXPT vs. CDISC-HL7 XMLa more robust model of the world of clinical researcha more robust model of the world of clinical research

SAS Transport XPT CDISC-HL7 XML

Washington

Dallas

Washington

Rio de Janeiro