Why ADaM for a statistician?

5

Click here to load reader

Transcript of Why ADaM for a statistician?

Page 1: Why ADaM for a statistician?

Why ADaM(Analysis Data Model)?

Kevin Lee, Cytel, Chesterbrook, PA

ABSTRACT The pharmaceutical industry is embracing CDISC as the standard of clinical trial process and ADaM is the part of the process that statisticians and clinical trial programmers is developing and using to analyze the clinical trial data. But, one can ask why ADaM? Why do we need to create ADaM? What are the benefits to use ADaM Data Structure? How can statisticians use them? The paper is intended for statisticians who are interested in ADaM. The paper will discuss the data structure of ADaM ADSL and ADaM BDS (Basic Data Structures). It will also discuss the role of ADaM in CDISC and its benefits in clinical trial. It will also discuss the roles of Clinical Trial SAS programmers and statisticians in ADaM(Usually, it depends on the company). The paper will provide the example of ADaM Data Set and SAS codes that generate the final numbers in the analysis. The example will display how ADaM Data set is built and how it can be used in the analysis.

THE FLOW OF CDISC CLINICAL TRIAL PROCESS

SDTM ADaM TFL

SAP Mock Up tables

eCRF ADaM Metadata

Protocol

EDC

ADaM Related Activity

INSTRUCTION OF ADaM ADaM(Analysis Data Model) currently has two different structures. One is ADSL, subject-level data. Another one is BDS(Basic Data Structure), multiple record per subject basic data. The naming convention of ADaM is ADxxxxxx. STRUCTURE OF ADSL The ADSL has one record for each subject. Usually, it contains subject level information such as demographic, treatment, disposition information and so on. STRUCTURE OF ADaM BDS The ADaM BDS(Basic Data Structure) can be used for all the data sets and consists of the following class variables.

• Subject Identifier Variables – STUDYID, ADDOMAIN and etc • Treatment Variables – TRTA, TRTAN and etc • Timing Variables – AVSITN, AVISIT and etc • Analysis Parameter Variables - PARMCD, PARAM, AVAL, AVALC, BASE, CHG and etc • Analysis Descriptor Variables – DTYPE and etc • Indicator Variables – PPROTFL, ABLFL and etc • Analysis Enabling Variables • Data Point Traceability Variables – SRCDOM, SRCVAR and SRCSEQ. • SDTM Variables

The special BDS such as ADAE(Adverse Event) and ADTTE(Time to Event) data structures will be available in public soon.

1

Page 2: Why ADaM for a statistician?

EXAMPLE Below examples will display the order and steps of ADaM Related Activity. The example is based on the time to event survival analysis. Mock Up Table

Table 14.2.1 Days to Death Through the end of Study by Treatment

PARAM = ‘Days to Death’, TRTP Analysis Population: Intent to Treat

ITTFL=‘Y’ Drug 1 Drug 2 p-value __________________________________________________________________________________________ N xxx xxx Censored Observation xxx xxx Median xx xx x.xx Q1, Q3 xx, xx xx, xx Min, Max xx, xx xx, xx __________________________________________________________________________________________ AVAL(Days to Event), CNSR(Censor information) PROC LIFETEST DATA=ADTTEOS; TIME AVAL*CNSR(1); STRATA TRTP; RUN; As seen in above mock up table, we need the following variables in ADTTEOS dataset.

• Analysis Parameter Variable - PARAM = “Days to Death (Days)”, AVAL, CNSR • Treatment Variables - TRTP • Indicator Variables – ITTFL

ANALYSIS DATASET METADATA

Dataset Name

Dataset Description

Dataset Location

Dataset Structure Key Variables of Dataset

Class of Dataset

Documentation

ADTTEOS Overall Survival Time to Event Analysis Dataset

adtteos.xpt

One Record per Subject per Parameter

USUBJID, PARAM

BDS adtteos.sas

ANALYSIS VARIABLE METADATA including ANALYSIS PARAMETER VALUE-LEVEL METADATA

Dataset Name

Parameter Identifier

Variable Name

Variable Label Variable Type

Display Format

Codelist / Controlled

Terms

Source/Derivation

ADTTEOS **ALL** STUDYID Study Identifier Char $12 ADSL.STUDYID

ADTTEOS **ALL** ADDOMAIN Analysis Domain Abbreviation

Char $8 ADLB Derived

ADTTEOS **ALL** USUBJID Unique Subject Identifier

Char $20 LB.USUBJID

ADTTEOS **ALL** ITTFL Intent-To-Treat Population Flag

Char $1 ADSL.ITTFL

ADTTEOS **ALL** TRTP Plan Treatment Group

Char $20 ADSL.TRT01P

ADTTEOS **ALL** TRTPN Planned Treatment Number

Num 8 ADSL.TRT01PN

ADTTEOS PARAM PARAM Parameter Char $100 Days to Death (Days)

ADTTEOS PARAMCD PARAMCD Parameter Code Char $8 OS

ADTTEOS **ALL** PARAMTYP Parameter Type Char $8 DERIVED

2

Page 3: Why ADaM for a statistician?

ADTTEOS **ALL** AVAL Analysis Value Num 8 ADT – STARTDT + 1

ADTTEOS **ALL** STARTDT Time to Event Origin Date for Subject

Num YYYY-MM-DD

ADSL.TRTSDT

ADTTEOS **ALL** ADT Analysis Date Num YYYY-MM-DD

Date of DS.DSDTC

ADTTEOS **ALL** CNSR Censor Num 8 0,1 0 for ‘DEATH’, 1 for any other study completion

ADTTEOS **ALL** EVNTDESC Event or Censoring Description

Char $40 DEATH, COMPLETED STUDY, LOST TO FOLLOW-UP, AE

DS.DSDECOD

RESULTS METADATA Metadata Field Metadata

DISPLAY IDENTIFIER

Table 14.2.1

DISPLAY NAME Days to Death Through the end of Study by Treatment, Analysis Population: Intent to Treat

RESULT IDENTIFIER Days to Death (Days)

PARAM Days to Death (Days)

PARAMCD OS

ANALYSIS VARIABLE

AVAL

REASON Primary efficacy analysis

DATASET ADTTEOS

SELECTION CRITERIA

ITTFL=“Y” and PARAM=“Days to Death (Days)’

DOCUMENTATION See SAP Section XX for details. Program: t_14.2.1_OS.sas

PROGRAMMING STATEMENTS

PROC LIFETEST DATA= ADTTEOS; WHERE ITTFL=‘Y’ and PARAM = “Days to Death (Days)”; TIME AVAL*CNSR(1); STRATA TRTP; RUN;

THE FLOW OF ADaM CREATION As seen above example, the followings could be the steps of ADaM Creation 1. Starts from Analysis (Mock Up Tables)

• What kind of analysis (ex, survival) • What SAS procedure is needed (ex, proc lifetest) • What is the subset condition (where statement) • What variables are needed(ex, ITTFL, TRTP, PARAM, AVAL and CNSR)

2. Create ADaM Metadata • Analysis Dataset • Variable • Parameter • Result

3. Create ADaM Datasets(ex, ADTTEOS) 4. Create Table(ex, Table 14.2.1)

The sample codes for report using ADaM BDS data The followings could be other examples to use ADaM to get the efficacy analysis.

3

Page 4: Why ADaM for a statistician?

• COX model with covariates of SEX and AGE, proc phreg data=ADTTEOS; where PARAM = ‘OVERALL SURVIVAL’; model AVLA*CNSR(1) = TRTPN SEX AGE; run;

• two sample t-test(comparing two treatment effects from baseline to week 6 on ‘Forced Expiratory Volume’) proc ttest data=ADFEV; where PARAM = ‘FORCED EXPIRATORY VOLUME’ and AVISIT=’WEEK 6’; class TRTPN; var CHG; run;

• wilcoxon rank-sum test(Non-parametric comparison on two treatment effects) proc npar1way wilcoxon data=ADFEV; where PARAM = ‘FORCED EXPIRATORY VOLUME’ and AVISIT=’WEEK 6’; class TRTPN; var CHG; run;

• Fisher’s exact test(Comparing two treatment effects on Hospital Readmission) proc freq data=ADHS; where PARAM = ‘READMISSION TO HOSPITAL’; table TRTPN*AVALC / fisher; run;

• ANOVA(comparing more than two treatment effects from baseline to week 6 on ‘Blood Pressure’) proc glm data=ADVS; where PARAM = ‘SYSTOLIC BLOOD PRESSURE’; class TRTPN; model CHG = TRTPN; lsmeans TRTPN / pdiff stderr; run;

TIPS TO USE ADaM BDS STRUCTURE As seen in above examples, the purpose of analysis is parameter. PARAM and PARAMCD are created in ADaM to select the analysis. AVAL or CHG is the analysis variable and the dependent variables in the model such as t-test, anova, ancova, wilcoxon, kruskal-wallis, regression, logistic regression, chi-square, fixer exact, log-rank test and so on. Treatment variables such as TRTP or TRTA are usually used as one of independent or explanatory variables in the general model. Indicator variables such as ITTFL, PPROTFL, SAFTY, and ANLY01FL are used for the subset condition. LIMITATION OF USING ADaM BDS STRUCTURE Beside ADSL, all the ADaM datasets are the vertical data structure, so data could be big. All the analyses can be done with one SAS procedure in ADaM except multivariate analysis. Since multivariate analysis requires more than one dependent variable, some data manipulation will be required. ADaM Team is currently developing Multivariate Data Structure. ADaM is new, so statisticians will have a learning curve just like anything else. BENEFITS OF USING ADaM First of all, ADaM is analysis-ready, which means that if one uses SAS, all the results in TFL(Tables, Figures, and Listings) report will be derived with one SAS procedure. Statisticians do not need any data manipulations in TFL programs. It could be easily validated by statisticians. Second, ADaM provides the traceability. Statisticians will be able to see where ADaM data come from or how ADaM data are derived.

4

Page 5: Why ADaM for a statistician?

5

In addition, ADaM BDS follows a general structure, so statisticians do not need to keep on learning the new database structure in clinical trials once they learn about ADaM. It will be a lot easier to discuss about ADaM with the fellow statisticians, programmers and even agency. Unlike SDTM, ADaM provide the flexibility, so it can accommodate any analysis. CONCLUSION ADaM is the new concept in the clinical trial study. Since they are new, it won’t be easy to learn and implement. But, once ADaM is learned and implemented, it will be much easier for statisticians to use. ADaM is usually created by programmers and it is easily used by statistician for any analysis. Unlike the pre-CDISC derived datasets, ADaM can be easily used by statisticians. ADaM is closer to analysis than to SDTM and any analysis is planned and reviewed by statisticians. Therefore, ADaM is for statisticians. REFERENCES The ADaM Implementation Guide, Version V 1.0 (ADaMIG v1.0) The Analysis Data Model, Version 2.1 (ADaM 2.1) CONTACT INFORMATION Your comments and questions are valued and welcomed. Please contact the author at Kevin Lee Cytel Inc. 640 Lee Road, Suite 201 Chesterbrook, PA 19087 610-994-9840 [email protected] TRADEMARKS SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product names are registered trademarks or trademarks of their respective companies.