Division of Cancer Control and Population Sciences

1
CISNET Cancer Intervention and Surveillance Mod
eling Network
Eric J. (Rocky) Feuer, Ph.D. Chief, Statistical R
esearch and Applications Branch, Division of
Cancer Control and Population Sciences, National
Cancer Institute
2
Overview

• NCI Sponsored Consortium Focused on Why and
  What If Questions in Cancer Trends
  
• Statistical Modeling of the Impact of Cancer
  Control Interventions (Screening, Treatment,
  Prevention) on Current and Future Trends
  
• Optimal Cancer Control Planning

• Two Rounds of Funding (4 year grants) U01
  Cooperative Agreement
  
• Sept. 2000 - 9 Funded Grants- Breast (7),
  Prostate (1) and Colorectal Cancer (1)
  
• Aug. 2002 8 Funded Grants - Prostate (1),
  Colorectal (2), and Lung Cancer (5)
  
• RFA to be reissued shortly

3
Funded CISNET Grantees
Three CRC Models Zauber (MSK), Kuntz (Harvard),
Rutter (Group Health)
4
CISNET Goals

• Develop and enhance multi-cohort based population
  models (including new methodology)
  
• Real cohorts representing the necessary birth
  cohorts to construct cross- sectional rates for
  specified years and age groups
  
• Ideally can address the full range of
  interventions
  
• Establish infrastructure to facilitate
  communication and understanding among modelers
  
• Model Profiler
• Comparative Modeling Projects (Base Cases)
• Gain access to data sets that might not otherwise
  be available
  
• Develop liaisons and provide assistance to
  outside groups to address questions amenable to
  modeling
  
• AHRQ/CMS CE of immunochemical FOBT
• CDC/NCI Healthy People 2010 mid-course
  correction studies
  

5
Types of Questions Population Models Can Address

• Responsive to Challenges Due to Increasing Pace
  of Technology.
  
• Provide short term answers while Randomized
  Controlled Trials (RCTs) are still in progress
  (e.g. PSA testing for prostate cancer)
  
• Address emerging questions while they are still
  being debated in the scientific/policy forum.
  
• Impact of smokeless tobacco products.
• Virtual colonoscopy esp. policies/practices in
  polyp-size threshold for sending patients on for
  conventional colonoscopy
  
• Translate RCT evidence to the population
  setting.
  
• Population impact of adjuvant therapy and
  mammography for breast cancer.
  
• Provide estimates of quantities that will never
  be derived from RCTs.
  
• Impact of smoking cessation in the US on future
  lung cancer trends.
  

6
Making Results of Modeling Efforts More
Transparent

• Modeling has been marred by
• Difficulty of understanding and comparing
  different model assumptions and structure
  
• Lack of comparability of inputs, outputs, basic
  definitions
  
• Addressed with Model Profiler and Base Cases

7
CISNET Model Profiler
8
(No Transcript)
9
CISNET Model Profiler
10
CISNET Model Profiler

11
CISNET Model Profiler
12
CISNET Model Profiler
cisnet.cancer.gov (public site)
13
CISNET Base Cases

• Groups jointly decide to address a common
  question
  
• Common population-based inputs are used
• Modelers maintain the unique deeper aspects of
  their models (e.g. assumptions and formulation of
  the natural history of disease)
  
• Common computer runs and content/format of
  outputs are specified
  
• Provides a chance to reach common consensus on
  important questions, and to better understand
  differences between models

14
CISNET Breast Cancer Base QuestionWhat is the
Impact of Mammography, Adjuvant Therapy, and the
Combination on U.S. Breast Cancer Mortality
1975-2000?
15
Colorectal Cancer Base Case Questions

• Graduated approach
• Base Case I What is the impact of different
  natural history models on incidence and mortality
  
  
• Hypothetical situation with conditions in 1978
  frozen in time with no interventions
  
• Base Case II What is the impact of a single
  screen (FOBT, colonoscopy, flex sig) at age 65
  w/wo follow-up surveillance after detection of an
  adenoma

16
Modeling Hypothetical Cohorts vs. Population
Cohorts

• Cost effectiveness models usually conducted on
  hypothetical cohorts
  
• In some settings population models can add
  additional insights to C/E issues especially
  when
  
• Capacity is an issue
• Costs depend on volume