PMED: APPM Workshop: Data & Analytics in Precision Oncology- Warren Kibbe, March 14, 2019
1. Data and Analytics in Precision
Oncology
Warren A. Kibbe, Ph.D.
Professor, Biostatistics & Bioinformatics
Chief Data Officer, Duke Cancer Institute
warren.kibbe@duke.edu
@wakibbe
#PredictiveModeling
#ComputationalPhenomics
#PrecisionOncology
3. Fundamental Changes
• Data generation is not the bottleneck
• Most data are now ‘digital first’
• Old statistical models assuming variable
independence are inadequate – systems
and pathways are not independent!
• Project management is critical in scaling
population science
Well-defined experiments are still key
4. Changes in Oncology
• Understanding Cancer Biology
• Anatomic vs molecular classification
• Health vs Disease
10. Big Data Scientist Training Enhancement
Program (BD-STEP)
Graduates of BD-STEP would:
• have skillsets to perform next-generation patient-
centered outcomes research by manipulating and
analyzing large-scale, multi-element, patient data
sets to develop novel disease signatures or unique
performance-based clinical benchmarks
• have an understanding of real-time, performance-
driven health care delivery in the VA systems
Frank Meng, VA Michelle Berny-Lang, NCI
11. Mining the VA Corporate Data
Warehouse
• From 130 clinical sites covering
about 9 current million veterans, 16
million since VistA was put in place in
1990
Work performed by David Winski, PhD
https://www.hsrd.research.va.gov/for_researchers/cyber_seminars/archives/2376-notes.pdf
12. Understanding NSCLC
• What is the impact of new
immunotherapies on the outcomes of
NSCLC patients in the VA?
• Does Mutational Tumor Burden
impact effectiveness?
• Is PD-L1 expression predictive of
response to immunotherapies
13. Mining the VA Corporate Data
Warehouse
Transforming the National Department of
Veterans Affairs Data Warehouse to the OMOP
Common Data Model
Fern FitzHenry ;
Jesse Brannen ;
Jason Denton ;
Jonathan R. Nebeker ;
Scott L. Duvall ;
Freneka F. Minter ;
Jeffrey Scehnet ;
Brian Sauer;
Lucila Ohno-Machado ;
Michael E. Matheny
14. Cancer Registry Tables (“Raw Onc Tables”)
- Set of two T-SQL tables comprised of a “Patient” table and
a “Cancer” table
- When a VA patient is diagnosed with cancer, cancer
registrars will enter a patient record in the Patient table and
a cancer record in the the Cancer table
- Tables structured along North American Association of
Central Cancer Registry (NAACCR) guidelines
- Patient table contains >100 fields containing patient
identifiers, patient demographic data and patient military
service data
- Cancer table contains >500 fields including date of
diagnosis, diagnosis codes, tumor location, tumor histology
and diagnosis-related medications/procedures
Work performed by David Winski, PhD
15. Identify patients receiving
immunotherapy
Work performed by David Winski, PhD
Transforming the National Department of
Veterans Affairs Data Warehouse to the OMOP
Common Data Model
Fern FitzHenry ;
Jesse Brannen ;
Jason Denton ;
Jonathan R. Nebeker ;
Scott L. Duvall ;
Freneka F. Minter ;
Jeffrey Scehnet ;
Brian Sauer;
Lucila Ohno-Machado ;
Michael E. Matheny
16. Building a Tumor-Sequenced Non-Small Cell
Lung Cancer (NSCLC) Cohort
1.Begin with all patients in Precision Oncology
Program (i.e. tumor profiled by NGS) with
associated NSCLC diagnosis (n=2057)
2.Filter to subset of these patients who received
chemo or immuno drugs through VA (n=1457)
3.Filter to those patients whose first date of
immunotherapy treatment was prior to April 2018
to allow enough time for survival analysis
(n=383)
4.Filter to those patients who had NSCLC diagnosis
corroborated in the Cancer Registry (n=330)
17. Lag in Cancer Registry Records
Work performed by David Winski, PhD
18. Lag in Cancer Registry is a Reporting Lag
Work performed by David Winski, PhD
Number of visits vs cancer diagnosis in the ‘Raw Onc’ tables
20. Immunotherapy Drugs of Interest
- Four drugs of interest: Pembrolizumab,
Nivolumab, Atezolizumab and Durvalumab
# of Orders at VA
21. NSCLC POP Dx With Tumor
Profiled by NGS: 2057
patients
NSCLC Verified in
Cancer Registry:
330
Immuno Prior to
April 2018: 383
Chemo/Immuno
Drug Orders at VA:
1472
22. PD-L1 expression and Nivolumab in NSLC
• We also examined PD-L1 testing and
the impact of high expressing tumors
on outcomes
– Inconclusive because many patients
were treated as second line therapy,
where PD-L1 testing is optional.
23. • Retrospective mining still requires
good questions and adequate power
• Even given the size of the VA, the
ability to build a well powered cohort
with good data is difficult