MAXIMIZING THE USE OF PATIENT DATA WAREHOUSE VIA …€¦ · MAXIMIZING THE USE OF PATIENT DATA...
Transcript of MAXIMIZING THE USE OF PATIENT DATA WAREHOUSE VIA …€¦ · MAXIMIZING THE USE OF PATIENT DATA...
MAXIMIZING THE USE OFPATIENT DATA WAREHOUSE VIA VISUAL ANALYTICS
Presented by Simson Alex & Sandy LeiJohnson & Johnson
2018-10-10 CDISC US Interchange
Agenda
1. Patient Data Warehouse (PDW)
2. PDW Study Browser
3. PDW Visual Analytics
4. PDW Data Conformance
5. Conclusion
PATIENT DATA WAREHOUSE
ENABLING EXPLORATION OF PATIENT DATA ACROSS JANSSEN TRIALS
10/17/2018 4
PATIENT DATA WAREHOUSE DATA SOURCE (IDAR SAS Data Folders)
a SDTM SAS Dataset Folder
SDTM: AE SAS Dataset
Dataset Metadata
Data and metadata browsed from a SDTM
SAS dataset folder via SAS Universal Viewer
10/17/2018
Patient Data Warehouse – PDW
• A centralized repository for:• Study documents : Protocol | SAP | CSR
• Clinical trial data : SAS datasets
• A technical capability for:• Data driven decision making
• Search, subset, integrate, export, analyze
Early-2013 Late-2013 2015 2016 2017 2018
ImplementStabilize Optimize
5
Conceptualized
and funded
First apps
released.
Partial SQL
Stable
applications.
Fully SQL
Visual Analytics
Development. Platform
Stabilization
PDW System Architecture
10/17/2018 6
Project Data
Internal and
External Sources
Patient Level
Data
SDTM, ADaM,
Legacy…
Relational Database
Automated
Loading
Processes
PDW System Architecture
10/17/2018 7
Database Application
Server
SAS Server
Trial
Catalog
Study
Browser
Visual
Analytic
s
Various Web
Applications
SQL
Data Response
Request
10/17/2018 8
PATIENT DATA WAREHOUSE
CONTENTS
467 Trials
With
Patient
Data
609
Reporting
Efforts
1251
Folders
40447
SAS
Datasets
~ 1.1
Billion SAS
Data Rows
All SAS datasets were loaded into database as is
10/17/2018 9
Data Divided Into Two Broad Sections
MetaData
SAS Value
Is the knowledge about
~33 Million Records
~1.1 Billion SAS Rows
Metadata is 3% of SAS data
10/17/2018 10
PDW Study Browser and Visual Analytics
Search
(Metadata in Database)
Select
(Metadata in Memory or in
Database)
Query
(Data, On-Demand)
Analysis Results
Visualization (Analysis
Templates)
Select
(Metadata In Database)
Output Datasets/Metadata
(sas, xpt, csv on file server)
Visual Analytics
(In Development)
Study Browser
(Production V2)
Retrieve Data in Database
(Query via a SAS Program)
Batch Job Submit & SAS
Server
10/17/2018 11
PDW Use Cases1. Signal Detections (9 studies) - PDW Safety Review proof of concept for Global
Medical Safety
2. Placebo Data Extraction (10 studies) - EMIF Metabolic: reuse of placebo data from completed clinical trials
3. Placebo Response (10 studies) – identify possible factors which resulted in high placebo response rate
4. Explore opportunities to increase trial participation (252 studies)
5. Hy’s Law by compound (17 studies in a compound)
6. Time-to-event for Major Adverse Events (10 studies)
7. Efficacy Analysis by Biological Age (all PDW studies) – Janssen Prevention Center proposal biological age in clinical trials
8. Concomitant Medication & PK response (80 studies) - Impact of a concomitant medication co-administration on PK of approved Janssen mAbs
9. Overall patient disposition rate by reasons of discontinuation (304 studies)
PDW STUDY BROWSER
ENABLING AUTOMATED DATA INTEGRATION & SUBJECT SUBSETTING
PDW Study Browser (SB) - Data Retrieval Tools
• Data Integration: based on metadata selections, datasets are stacked into one file for the datasets with the same dataset name. Conformance to data standard is the key.
• Subject Subsetting: based on metadata conditions/criteria set up, output all datasets for subjects meeting the conditions/criteria.
Note: work flows for both tools are very similar.
10/17/2018 13
10/17/2018 14
Study Browser – Data IntegrationCross-Study Dataset, Variable, Row Selection
Users’ decision for analysis & reporting tools to perform research/analysis based on extracted data.
Datasets
Studies
Variables
Studies
Values,
Conditions
10/17/2018 15
A visualization of a Hy’s Law subjectAST/ALT > 3xUL and TB > 2xUL in 30 days
PDW VISUAL ANALYTICS
ENABLING AUTOMATED DATA INTEGRATION & ANALYSIS
10/17/2018 17
PDW Visual Analytics Data Integration & Analysis(in development )
End-to-End platform by enabling automated analysis from metadata browsing to data extraction to final integrated analysis:
• Set up the connections from a visualization tool to PDW database
• Set up on-demand framework for data query from PDW database to facilitate analysis
• Design and create reusable analysis templates which will populate analysis results based on the data query created by the selected metadata.
10/17/2018 18
Figure 1: 46 SDTM studies were selected from 3 compounds Figure 2: Select interested datasets (AE,CM, DM)
Figure 3: Select variables to facilitate analysis Figure 4: refreshed analysis results from an analysis template
PDW Visual Analytics
An Example Application Design
10/17/2018 19
10/17/2018 20
PDW STUDY DATA CONFORMANCE
10/17/2018 21
PDW Data Integration Use CasePooling 10 Phase 2/3 Studies from 5 Compounds
Data Standardization Efforts Required
• Mapping: map datasets, variables and values
• Derivation: variable or value required by analysis but not created in the original analysis dataset
• Verification: data comparisons, reproduce CSR outputs using integrated data, etc.
Data Integration (by PDW tools):
• ADSL : Subject-Level (50 standard variables)
• ADEFF: Efficacy (44 standard variables, 40 standard efficacy parameters)
10/17/2018 22
Baseline XYZ DosePooling 10 Phase 2/3 Studies from 5 Compounds
STUDYID DATA SOURCE DATASET VARIABLE XYZ DOSE RANGE
S0001 SDTM CM CMDOSTXT (2,25)
S0002 SDTM CM CMDOSTXT (7.5,25)
S0003 ANALYSIS MEDRVIEW MRXYZDS (0,…,30)
S0004 ANALYSIS EXPHX EHSCRDOS (15,25)
S0005 ANALYSIS SUBJSF XYZDOSBL (0.5,30)
S0006 ANALYSIS MEDRVIEW MRXYZDOS (0,25)
ANALYSIS SUBJEF XYZBL (15,25)
ANALYSIS EXPHX EHSCRDOS (15,25)
S0007 ANALYSIS ADCM CMDOSTWK (7.5,20)
S0008 ANALYSIS ADSL XYZBL (12,5,15,17,…,25)
S0009 ANALYSIS WNLOBS XYZD (10,25)
ANALYSIS EXPHX EHSCRDOS (10,25)
S0010 ANALYSIS ADSL BLXYZDS (10,25)
10/17/2018 23
Data Standardization ChallengesPooling 10 Phase 2/3 Studies from 5 Compounds
• Same information created in different places - Baseline XYZ Dose
• Same variable applied different derivation - Eg: CHG=AVAL-BASE in some studies and CHG=BASE-AVAL in other studies
• Lack of metadata or variable flags for duplicate records:Without straightforward algorithm to select from duplicate observations/records for analysis is commonly observed in this PDW use case.
• Repetitive efforts for verification – verification tools for both analysis data and outputs need to be created due to the data manipulation above, reports comparisons are manual and time consuming.
10/17/2018 24
Conclusion
• The new capabilities to search, query and perform analysis over hundreds of studies from PDW platform is a big achievement.
• Adherence to clinical data standards is key to address diverse questions accurately, confidently, and in a timely manner.
• As PDW data content increases, we are continuously improving our technological solutions to address performance issues.
• The PDW Visual Analytics platform technologies will shape the clinical trial processes to make “real-time” analysis possible during any time point of the drug development cycle.
Thank You!