Simulation Models to Inform Health Policy: Colorectal Cancer Screening Carolyn Rutter, PhD Group Health Research Institute Supported by.

Slides:



Advertisements
Similar presentations
High Resolution studies
Advertisements

Evidence-based Dental Practice Developing guidelines or clinical recommendations Slide #1 This lecture follows the previous online lecture on evidence.
Helical CT Screening for Lung Cancer at Advanced Radiology Consultants
1 Bayesian CTS FDA/Industry Workshop September 18, 2003Copyright Pharsight Case Study in the Use of Bayesian Hierarchical Modeling and Simulation.
1 Eloise E. Kaizar The Ohio State University Combining Information From Randomized and Observational Data: A Simulation Study June 5, 2008 Joel Greenhouse.
1Regional policy responses to demographic challenges, Bruxelles, January 2007 EUROSTAT regional population projections Giampaolo LANZIERI Eurostat.
Using Cancer Registry Data for Comprehensive Cancer Control Christie Eheman, PhD, National Program of Cancer Registries, Division of Cancer Prevention.
Colorectal Cancer Awareness in TN: Risk Factors, Screening, Outreach Keith D. Gray, M.D. Assistant Professor of Surgery Division of Surgical Oncology The.
MCMC estimation in MlwiN
Evaluating Provider Reliability in Risk-aware Grid Brokering Iain Gourlay.
STATISTICAL INFERENCE ABOUT MEANS AND PROPORTIONS WITH TWO POPULATIONS
Chapter 7 Sampling and Sampling Distributions
Bayesian network for gene regulatory network construction
Phase II/III Design: Case Study
Lecture 3 Validity of screening and diagnostic tests
CT COLONOGRAPHY. CRC TRENDS  Incidence decreased by 7%  Mortality decreased by 20%  Five year survival rates increased by 12%
Increasing Patient Activation to Improve Health and Reduce Costs
1 General Iteration Algorithms by Luyang Fu, Ph. D., State Auto Insurance Company Cheng-sheng Peter Wu, FCAS, ASA, MAAA, Deloitte Consulting LLP 2007 CAS.
25 seconds left…...
Annual Industry Accounts Overview George Smith & Nicole Mayerhauser Current Industry Analysis Division Bureau of Economic Analysis Industry Accounts Users’
CANCER SCREENING 2011 DELAWARE CANCER EDUCATION ALLIANCE STEPHEN S. GRUBBS, M.D. HELEN F. GRAHAM CANCER CENTER DELAWARE CANCER CONSORTIUM OCTOBER 5, 2011.
January Structure of the book Section 1 (Ch 1 – 10) Basic concepts and techniques Section 2 (Ch 11 – 15): Inference for quantitative outcomes Section.
Institute for Public Health, Medical Decision Making and Health Technology Assessment 1 Results of the PanEuropean Hepatitis C Project 3 rd Paris Hepatitis.
Colorectal Cancer Screening and Surveillance FDA Advisory Committee March, 2002 David Lieberman MD Chief, Division of Gastroenterology Oregon Health Sciences.
†Source: U.S. Cancer Statistics Working Group. United States Cancer Statistics: 1999–2011 Incidence and Mortality Web-based Report. Atlanta (GA): Department.
Breast MR Imaging Workshop th September 2014 High-Risk Screening Evidence-based Clinical Indications for Breast MRI Dr. Muhamad Zabidi Ahmad, AMDI.
1 Parametric Sensitivity Analysis For Cancer Survival Models Using Large- Sample Normal Approximations To The Bayesian Posterior Distribution Gordon B.
HOW STANDING ORDERS HELPED US IMPROVE CANCER SCREENING: REPORT FROM A NEW PPRNet MEMBER JULIO A SAVINON, MD RIO GRANDE MEDICINE INC. HARLINGEN, TX.
Results 2 (cont’d) c) Long term observational data on the duration of effective response Observational data on n=50 has EVSI = £867 d) Collect data on.
Otis W. Brawley, M.D. Chief Medical and Scientific Officer Executive Vice President American Cancer Society Professor of Hematology, Medical Oncology,
Colorectal cancer: How do we approach health disparities? Marta L. Davila, MD, FASGE University of Texas MD Anderson Cancer Center.
Colorectal Cancer Screening John Pelzel MD Sleepy Eye Medical Center.
Economic Evaluation of Cancer Screening - Case of Colorectal Cancer – Cost-Effectiveness analysis of stool DNA to Screen for Colorectal Cancer October.
Update on Colorectal Cancer Screening Tests Source: Levin Bernard et al. Screening and Surveillance for the Early Detection of Colorectal Cancer and Adenomatous.
Sharp L, Tilson L, Whyte S, Ó Céilleachair A
Geriatric Health Maintenance: Cancer Screening Linda DeCherrie, MD Geriatric Fellow Mount Sinai Hospital.
AIMGP Seminar Series January 2004 Joo-Meng Soh Edited by Gloria Rambaldini CANCER SCREENING PART II.
Multiple Choice Questions for discussion
Decision Analysis of Colorectal Cancer Screening Tests by Age to Begin, Age to End, and Screening Intervals: Report to the United States Preventive Services.
Evidence Evaluation & Methods Workgroup: Developing a Decision Analysis Model Lisa A. Prosser, PhD, MS September 23, 2011.
Slides last updated: June 2015 CRC: CLINICAL FEATURES.
INCIDENCE AND SURVIVAL TRENDS OF COLORECTAL CANCER FROM 2002 TO 2011 BE Ansa; E Alema-Mensah; MD Claridy; JQ Sheats; B Fontenot, and SA Smith Georgia Regents.
EDRN Approaches to Biomarker Validation DMCC Statisticians Fred Hutchinson Cancer Research Center Margaret Pepe Ziding Feng, Mark Thornquist, Yingye Zheng,
CISNET and BCSC: Working Together To Model The Population Impact Breast Cancer Screening A Celebration of the Work of the Breast Cancer Surveillance Consortium.
HW215: Models of Health & Wellness Unit 7: Health and Wellness Models Geo-political Influences.
Joni Reynolds, RN-CNS, MSN Director of Public Health Programs Winnable Battles: Cancer in Colorado.
The effects of inadequate preparation quality for colonoscopy Eric Sherer and Michael Catlin August 20 th, 2010 HSR&D Work-in-Progress 1.
Latent Class Regression Model Graphical Diagnostics Using an MCMC Estimation Procedure Elizabeth S. Garrett Scott L. Zeger Johns Hopkins University
Modeling Efforts to Inform Countries’ Screening Decisions Ann Graham Zauber, Iris Vogelaar, Marjolein van Ballegooijen, Deb Schrag, Rob Boer, Dik Habbema,
Colorectal Cancer Screening Colorectal Cancer Screening VT SGNA Conference VT SGNA Conference October 24, 2015 October 24, 2015 Lynn Butterly, MD Lynn.
CT Colonography vs Colonoscopy for the Detection of Advanced Neoplasia David H. Kim, M.D., Perry J. Pickhardt, M.D., Andrew J. Taylor, M.D., Winifred K.
Screening – a discussion in clinical preventive medicine Galit M Sacajiu MD MPH.
CT Screening for Lung Cancer vs. Smoking Cessation: A Cost-Effectiveness Analysis Pamela M. McMahon, PhD; Chung Yin Kong, PhD; Bruce E. Johnson; Milton.
Decoding the USPSTF By: Dr Vikram Arora Heritage Valley Health System.
Towards Global Eminence K Y U N G H E E U N I V E R S I T Y Colonoscopy Surveillance After Colorectal Cancer Resection: Recommendations of the US Multi-Society.
Date of download: 5/31/2016 From: Tipping the Balance of Benefits and Harms to Favor Screening Mammography Starting at Age 40 Years: A Comparative Modeling.
Date of download: 7/1/2016 From: Evaluating Test Strategies for Colorectal Cancer Screening: A Decision Analysis for the U.S. Preventive Services Task.
Cancer prevention and early detection
Cancer Screening Guidelines
Colorectal Cancer Screening Guidelines
Mammograms and Breast Exams: When to start /stop mammograms
More Ontarians need to be screened for colorectal cancer (Sept. 2012)
Background & Objectives
From: Tipping the Balance of Benefits and Harms to Favor Screening Mammography Starting at Age 40 YearsA Comparative Modeling Study of Risk Ann Intern.
Simulation Models to Inform Health Policy: Colorectal Cancer Screening
Colorectal Natural History Model
Volume 154, Issue 3, Pages e18 (February 2018)
Volume 154, Issue 1, Pages e20 (January 2018)
Evidence Based Diagnosis
Presentation transcript:

Simulation Models to Inform Health Policy: Colorectal Cancer Screening Carolyn Rutter, PhD Group Health Research Institute Supported by NCI U01 CA

April, Models for Mammography On Nov 16, 2009 the United States Preventive Services Task Force (USPSTF) revised its recommendations for screening mammography, indicating that most women should start regular breast cancer screening at age 50, not 40, and women should test every other year instead of every year. The change was made in light of: New data, including a very large study of mammography initiated at 40 from England (~1.5 million) New focus on harms, including work-up following a false positive test and over diagnosis and treatment of cancer that would not otherwise have been detected in a woman’s lifetime. Results from disease models (CISNET) The models did not provide consistent results about over-diagnosis, with estimates that ranged from 6% to 50% of breast cancers. However, there was agreement that mammograms every two years give the nearly the same benefit as annual ones but confer half the risk of harms, and that there was little benefit to screening women in their 40’s. (Paraphrasing Don Berry) New York Times, 11/22/09: “Behind Cancer Guidelines, Quest for Data”, Gina Kolata

April, Models for Colorectal Cancer November 2008: The USPSTF used microsimulation models to guide recommendations for colorectal cancer screening. Both models supported a range of screening strategies, starting at age 50 up to age 75: colonoscopy every 10 years annual FOBT flexible sigmoidoscopy every 5 years with a mid-interval FOBT Zauber, Lansdorp-Vogelaar, Knudsen, et al 2008 January 2009: The Medicare Evidence & Discovery Coverage Advisory Committee considered coverage of CT colonography for colorectal cancer screening, based on: Operating characteristics & risks of different tests (systematic review) Expert Testimony Model Predictions Knudsen, Lansdorp-Vogelaar, Rutter, et al 2010 Recommendation: Insufficient evidence to support CTC for screening

April, Microsimulation Models Microsimulation models simulate outcomes in a population of interest by simulating individual event history trajectories. A natural history model refers to the mathematical formulae that describe these individual histories. The natural history model is combined with a screening or treatment model to project population-level effects of treatment. Existing models describe: Cancers: prostate, breast, lung, colorectal, esophogeal, cervical Cardiovascular disease, diabetes

April, Adenoma to Carcinoma Pathway Normal Epithelium Colorectal Cancer Small Adenoma Advanced Adenoma Thanks to Ann Ann Zauber!

April, Example: CRC-SPIN model for colorectal cancer Time-Dependent Poisson Process with intensity i ( t) i ( t ) a function of age gender patient-specific risk adenoma growth model + P(invasive) =  (size) no lesion adenoma invasive disease clinical disease t1t1 t2t2 t3t3 Lognormal sojourn time model) ColoRectal Cancer Simulated Population model for Incidence and Natural history Rutter & Savarino, Cancer Epidemiology Biomarkers & Prevention, 2010 DEATH

April, CRC-SPIN model for colorectal cancer Model Summary 1.Adenoma Risk Model (7 parameters): 2.Adenoma Growth Model: Time to 10mm (4 parameters) 3.Adenoma Transition Model (8 parameters) 4.Sojourn Time Model (4 parameters) 23 parameters

April, CRC-SPIN model for colorectal cancer Adenoma Location: Multinomial distribution, informed by 9 autopsy studies (and similar to a recent colonoscopy study) : 0 parameters P(cecum)=0.08 P(ascending colon)=0.23 P(transverse colon)=0.24 P(descending colon)=0.12 P(sigmoid colon)=0.24 P(rectum)=0.09 CRC Survival: Assign CRC survival using survival curves estimated SEER survival data from 1975 to 1979, stratified by location (colon or rectum) and AJCC stage with age and sex included as covariates. Other Cause Survival: Assign other-cause mortality using product-limit estimates for age and birth-year cohorts from the National Center for Health Statistics Databases.

Calibrating Models in Economic Evaluation: A Seven-Step Approach Vanni, Karnon, Madan, et al. Pharmacoepidemiology, Which parameters should be varied in the calibration process? 2. Which calibration targets should be used? 3. What measure of goodness of fit (GOF) should be used? 4. What parameter search strategy should be used? 5. What determines acceptable GOF parameter sets (convergence criteria)? 6. What determines the termination of the calibration process (stopping rule)? 7. How should the model calibration results and economic parameters be integrated? April,

10 Example: Calibration of CRC-SPIN no lesion adenoma invasive disease clinical disease t1t1 t2t2 t3t3 1 study of Adenoma Prevalence (3 age groups) 2 studies of cross- sectional size information (3 size categories) 23 Parameters 1 study of preclinical cancer incidence SEER cancer incidence rates by age, sex, & location (1978) + Expert opinion about t1, t2, and t3 and prior information, about adenoma prevalence 29 data points

April, Microsimulation Model Calibration: Published Approaches One-at-a-time parameter perturbation, with subjective judgment One-at-a-time parameter perturbation, with chi-square or deviance statistics Grid Search using objective functions (e.g., likelihood) Undirected: evaluate objective function at each node or at a random set of parameter values. Not computationally feasible for highly parameterized models, dense grids can miss maxima. Directed: move through the parameter space toward improved goodness of fit to calibration data. Fewer evaluations of the likelihood, but can converge to local maxima. A key challenge is numeric approximation of derivatives. Likelihood approaches Bayesian estimation (MCMC, other approaches are possible) Simplify the likelihood to allow usual estimation approaches. The simplified model needs to be flexibly enough to be useful for prediction Active area of research

April, Microsimulation Model Calibration Bayesian calibration approach: Place priors on model parameters (allows explicit incorporation of expert opinion) Use simulation-based estimation (Markov Chain Monte Carlo or Sampling Importance Resampling)

JSM Vancouver, Microsimulation Model Calibration Bayesian model: data distribution: Y |    (y;  ) prior distribution:  ~  (·) posterior distribution  | y ~ h(  ;y) =  (y;  )  (·)/  (y) Microsimulation model: data distribution: Y |    (y;g(  )) prior distribution:  ~  (·) posterior distribution  | y ~ h(  ;y) =  (y;g(  ))  (·)/  (y) Bayesian Estimation: Closed form posterior Simulation-based estimation: * Markov Chain Monte Carlo (MCMC) * Importance sampling Likelihood is not closed form The data distribution is parameterized by g(  ), an unknown function of model parameters. Data sets do not have a 1-1 correspondence with model parameters. Solution: Simulate data given θ, and use this to simulate g(θ)

April, Microsimulation Model Calibration Advantages of Bayesian Calibration (simulated draws from the posterior distribution): Interval estimation Posterior predicted values for calibration data (Goodness of Fit) Can compare prior and posterior distributions to gain insight about parameter identifiability (Garrett & Zeger, Biometrics 2000)

JSM Vancouver, Bayesian MSM Calibration Overlap Statistic Compare prior and posterior distributions Garret & Zeger `Latent Class Model Diagnostics', Biometrics,

April, Microsimulation Model Calibration A 2009 review by Stout and colleagues (Pharmacoeconomics) found: Most modeling analyses did not describe calibration approaches Those that did generally used either undirected or directed grid searches. Big problem: lack of information about precision. Grid searches generally provide point estimates, with no measures of uncertainty for either estimated parameters or resulting model predictions. A few ad hoc approaches have been proposed, for example: ‘uncertainty intervals’ based on sampling parameters over a specified range Provide a range of predictions based on parameters that provide equally good fit to observed data.

April, Example of a Modeling Analysis Table 8C. Undiscounted costs by type, number of life-years gained, and number of cases of CRC per year-olds, by screening scenario – SIMCRC Scenario Screening Costs Total Costs LYG Symptomatic CRC Screen-Det CRC No Screening0$3.5M0600 SENSA$122K$2.8M FIT$248K$2.6M FSIG+SENSA$444K$2.7M FSIG+FIT$638K$2.8M CSPY$783K$2.7M CTC(1)$1.1M$3.3M CTC(2)$1.2M$3.3M160712

April,

April, Simpler models, similar answers Comparative Effectiveness Studies of CT colonography (Rutter, Knudsen, Pandharipande, Aca Radiol, 2011) Literature review: 1decision tree model 6 cohort models 3microsimulation models Similar overall disease processes (adenoma-carcinoma), similar ‘calibration’ data, but different levels of detail. Microsimulation models: multiple adenomas, of different sizes and in different (& detailed) locations in the colorectum Cohort / decision tree models: most with one adenoma, one with two in either the proximal or distal colon.

April, Simpler models, similar answers First Author YearModel TypeMost Effective Least Costly Most Cost-Effective Heitman 2005Decision TreeCSPY Sonnenberg 1999CohortCSPYCTCCSPY Ladabaum 2004CohortCSPY Vijan 2007CohortCSPY Hassan 2007CohortCSPYCTC Lee 2010CohortCSPYCTCCTC & CSPY Telford 2010CohortCSPY Knudsen 2010MicrosimCSPY

April, colonoscopy CT C Costs: CTC vs colonoscopy with and without polypectomy Hassan et al Lee et al

Accuracy of CTC v Colonoscopy April, Assumptions made by Hassan et al Assumptions made by Lee et al Small lesions not modelled

April, What next?  Simulation models are increasingly being used to inform health policy  Useful tool, with some shortcomings oMany are complex (when do simper models make sense?) oFew methods for estimating the precision of model predictions oFew (no?) models are ‘public use’  Potential for other applications oStudy design