Evaluating the “One-Model Fits All” Approach for Modeling Clinical Trial Adverse events Stephanie Pan, MS Icahn School of Medicine at Mount Sinai Hospital,

Slides:



Advertisements
Similar presentations
7. Models for Count Data, Inflation Models. Models for Count Data.
Advertisements

Exploring uncertainty in cost effectiveness analysis NICE International and HITAP copyright © 2013 Francis Ruiz NICE International (acknowledgements to:
RELATIVE RISK ESTIMATION IN RANDOMISED CONTROLLED TRIALS: A COMPARISON OF METHODS FOR INDEPENDENT OBSERVATIONS Lisa N Yelland, Amy B Salter, Philip Ryan.
ODAC May 3, Subgroup Analyses in Clinical Trials Stephen L George, PhD Department of Biostatistics and Bioinformatics Duke University Medical Center.
Statistical Issues in Contraceptive Trials
= == Critical Value = 1.64 X = 177  = 170 S = 16 N = 25 Z =
Decision Analysis as a Basis for Estimating Cost- Effectiveness: The Experience of the National Institute for Health and Clinical Excellence in the UK.
Estimation of the number of people with undiagnosed HIV infection in a country Andrew Phillips, UCL, London HIV in Europe Meeting 2 November 2009, Stockholm.
Dr. Kari Lock Morgan Department of Statistics Penn State University Teaching the Common Core: Making Inferences and Justifying Conclusions ASA Webinar.
Measurement Error.
Frequency and type of adverse events associated with treating women with trauma in community substance abuse treatment programs T. KIlleen 1, C. Brown.
Empirical Methods for Microeconomic Applications University of Lugano, Switzerland May 27-31, 2013 William Greene Department of Economics Stern School.
Risk Assessment and Comparative Effectiveness of Left Ventricular Assist Device and Medical Management in Ambulatory Heart Failure Patients Assessment.
What do we know about overall trends in patient safety in the USA? Patrick S. Romano, MD MPH Professor of Medicine and Pediatrics University of California,
EBC course 10 April 2003 Critical Appraisal of the Clinical Literature: The Big Picture Cynthia R. Long, PhD Associate Professor Palmer Center for Chiropractic.
Statistical estimation, confidence intervals
1 An Interim Monitoring Approach for a Small Sample Size Incidence Density Problem By: Shane Rosanbalm Co-author: Dennis Wallace.
Patricia Guyot1,2, Nicky J Welton1, AE Ades1
August 20, 2003FDA Antiviral Drugs Advisory Committee Meeting 1 Statistical Considerations for Topical Microbicide Phase 2 and 3 Trial Designs: A Regulatory.
Bivariate Poisson regression models for automobile insurance pricing Lluís Bermúdez i Morata Universitat de Barcelona IME 2007 Piraeus, July.
1 Study Design Issues and Considerations in HUS Trials Yan Wang, Ph.D. Statistical Reviewer Division of Biometrics IV OB/OTS/CDER/FDA April 12, 2007.
Simulation Study for Longitudinal Data with Nonignorable Missing Data Rong Liu, PhD Candidate Dr. Ramakrishnan, Advisor Department of Biostatistics Virginia.
IMPORTANCE OF STATISTICS MR.CHITHRAVEL.V ASST.PROFESSOR ACN.
Confidence Interval Estimation For statistical inference in decision making: Chapter 9.
Discrete Choice Modeling William Greene Stern School of Business New York University.
Changes in Quality of Life and Disease- Related Symptoms in Patients with Polycythemia Vera Receiving Ruxolitinib or Best Available Therapy: RESPONSE Trial.
European Patients’ Academy on Therapeutic Innovation The Purpose and Fundamentals of Statistics in Clinical Trials.
Rosuvastatin 10 mg n=2514 Placebo n= to 4 weeks Randomization 6weeks3 monthly Closing date 20 May 2007 Eligibility Optimal HF treatment instituted.
Long-Term Tolerability of Ticagrelor for Secondary Prevention: Insights from PEGASUS-TIMI 54 Trial Marc P. Bonaca, MD, MPH on behalf of the PEGASUS-TIMI.
Long-Term Tolerability of Ticagrelor for Secondary Prevention: Insights from PEGASUS-TIMI 54 Trial Marc P. Bonaca, MD, MPH on behalf of the PEGASUS-TIMI.
Anemia in CKD The TREAT Trial Reference Pfeiffer MA. A trial of Darbepoetin alpha in type II diabetes and chronic kidney disease. N Engl J Med. 2009;361:2019–2032.
1 Basics of Inferential Statistics Mark A. Weaver, PhD Family Health International Office of AIDS Research, NIH ICSSC, FHI Lucknow, India, March 2010.
Methods of Presenting and Interpreting Information Class 9.
Microeconometric Modeling
Is High Placebo Response Really a Problem in Clinical Trials?
Harvard T.H. Chan School of Public Health
Microeconometric Modeling
Chapter 12 Analysis of count data.
OHDSI Method Evaluation
The Importance of Adequately Powered Studies
Adaptive non-inferiority margins under observable non-constancy
Alcohol, Other Drugs, and Health: Current Evidence
Impact of State Reporting Laws on Central Line– Associated Bloodstream Infection Rates in U.S. Adult Intensive Care Units Hangsheng Liu, Carolyn T. A.
Martha Carvour, MD, PhD March 2, 2017
Sample Size Estimation
Discrete Choice Modeling
Generalized Linear Models
Strategies for Implementing Flexible Clinical Trials Jerald S. Schindler, Dr.P.H. Cytel Pharmaceutical Research Services 2006 FDA/Industry Statistics Workshop.
Discrete Choice Modeling
Analysis of count data 1.
Aiying Chen, Scott Patterson, Fabrice Bailleux and Ehab Bassily
Microeconometric Modeling
Statistical considerations for the Nipah virus treatment study
SIGNIFY Trial design: Participants with stable coronary artery disease without clinical heart failure and resting heart rate >70 bpm were randomized to.
DOSE SPACING IN EARLY DOSE RESPONSE CLINICAL TRIAL DESIGNS
Why use marginal model when I can use a multi-level model?
Section 3: Estimating p in a binomial distribution
Interpreting Basic Statistics
Discrete Choice Modeling
Count Models 2 Sociology 8811 Lecture 13
CHAMPION Trial design: Patients with recent hospitalization for heart failure were implanted with a pulmonary artery pressure monitor and randomized so.
Introduction to Probability Distributions
Handling Missing Not at Random Data for Safety Endpoint in the Multiple Dose Titration Clinical Pharmacology Trial Li Fan*, Tian Zhao, Patrick Larson Merck.
Introduction to Probability Distributions
Empirical Methods for Microeconomic Applications University of Lugano, Switzerland May 27-31, 2019 William Greene Department of Economics Stern School.
Type I and Type II Errors
Biomarkers as Endpoints
Use of Piecewise Weighted Log-Rank Test for Trials with Delayed Effect
Phase III randomized study of the proposed adalimumab biosimilar GP2017 in psoriasis: impact of multiple switches A. Blauvelt,1 J.-P. Lacour,2 J. F. Fowler.
Yu Du, PhD Research Scientist Eli Lilly and Company
Presentation transcript:

Evaluating the “One-Model Fits All” Approach for Modeling Clinical Trial Adverse events Stephanie Pan, MS Icahn School of Medicine at Mount Sinai Hospital, New York, NY

Background In an RCT, AEs are collected to monitor patient safety. These data are typically collapsed into counts per patient and summarized as rates over patient time at risk. Although the distribution of events per patient will vary widely depending on AE type, treatment differences are typically estimated assuming the same underlying distributional assumption for all AEs (typically Poisson). However, what happens if the distributional assumptions are not met and how robust are these count models to misspecifications?

Simulation Method Negative Binomial (NB) Zero-Inflated Negative Binomial (ZINB) Poisson Inverse Gaussian (PIG) Poisson Hurdle (PH) Zero-Inflated Poisson (ZIP) Negative Binomial Hurdle (NBH) We evaluated the robustness of the typical Poisson model versus extensions and alternatives under various assumptions using simulation.  Alternatives explored include: Generated 1000 repeated samples of AE counts from 500 patients assuming AEs arise from true Poisson, over-dispersed NB, zero-inflated, and zero-count with positive-count process distributions Assumed 20% structural zeroes in simulating zero-inflated and zero-count process Fit various models and compared on mean estimates, mean squared error (MSE), and coverage probability (how often does the 95% confidence interval include the true estimate) We then fit these models to AE data from a recent RCT to assess performance and fit. 

True Distribution (β=1.60) Simulation Results Mean estimate and coverage probabilities are reported in the table below: True Distribution (β=1.60) Poisson NB* NB** ZIP ZINB* 1.61 (0.95) 1.60 (0.82) 1.61 (0.55) 1.62 (0.94) 1.60 (0.83) NB 1.60 (0.96) 1.61 (0.94) 1.62 (0.95) 1.60 (0.95) PIG --- 1.57 (0.96) 1.33 (0.92) 1.55 (0.96) 1.49 (0.96) 1.32 (0.73) 1.13 (0.42) 1.56 (0.88) 1.35 (0.75) ZINB 1.57 (0.94) 1.58 (0.92) 1.55 (0.90) 1.60 (0.91) Poisson Hurdle 2.09 (0.97) 2.35 (0.97) 1.36 (0.75) NB Hurdle 2.11 (0.97) 1.64 (0.82) 1.57 (0.93) 2.38 (0.97) 1.67 (0.94) Poisson mean estimate is consistent across the varying distributions However, coverage probability for Poisson model in the presence of overdispersion is reduced Poisson model had consistently produced estimated standard errors that were too small PIG model is comparable to NB but failed to converge for simulated Poisson process distributions *Low dispersion (k=0.25) **High dispersion (k=0.05)

Zero-Count, Zero Truncated Poisson Zero-Count, Zero Truncated NB* Simulation Results Mean estimate and coverage probabilities are reported in the table below: Important to consider if zeroes are from a single “structural” source and if more flexible models are needed to handle zero counts If the true underlying distribution follows a zero-count, zero truncated Poisson we find that the Poisson and NB hurdle models are comparable in estimation and coverage If the true underlying distribution follows a zero-count, zero truncated NB we find that the Poisson hurdle model performs poorly in estimation True Distribution (β=1.60) Zero-Count, Zero Truncated Poisson Zero-Count, Zero Truncated NB* Poisson 0.30 (0) 0.66 (0) NB PIG --- 0.61 (0) ZIP 0.79 (0) ZINB 0.68 (0) Poisson Hurdle 1.63 (0.95) 1.27 (0.27) NB Hurdle 1.64 (0.96) 1.60 (0.94) Important to consider if zeroes are from a single “structural” source and if more flexible models are needed to handle zero counts *Low dispersion (k=0.25)

Application To AN RCT Cardiothoracic Surgical Network (CTSN) Trial on Intramyocardial Injection of Mesenchymal Precursor Cells (MPCs) among Left Ventricular Assist Device (LVAD) recipients with advanced heart failure Patients (N=159) were randomized to receive injections of MPCs or placebo and adverse events were reported at 6 months Population of patients were more likely to experience certain type of AEs such as bleeding, infections, cardiac arrhythmias Model estimates, fit, and performance were assessed using 1000 repeated sub-sampling and cross-validation Poisson AIC was consistently high in the presence of specific AEs with overdispersion In some instances, robust or sandwich estimators can be used to adjust SEs Poisson Distribution of MSE for Overall Bleeding Poisson Distribution of AIC for Overall Bleeding

Conclusion and Ongoing Research Explored extended count models such as PIG, zero-Inflated, and hurdle with applications to an RCT Prior studies have shown that the Poisson models, including ZIP and Poisson hurdle, have poorer coverage probability in the presence of underlying overdispersion Given the potential cost of misspecification, careful consideration is needed for a “One-Model Fits All” approach and the functional form of the count responses Further research is needed on incorporating a composite effect size across multiple AE models versus modeling the total number of AEs Acknowledgements Jessica Overbey, MS and Emilia Bagiella, PhD for their thoughtful insights and helpful review of the abstract and presentation.

References Hilbe, J. (2014). Modeling Count Data. Cambridge: Cambridge University Press. doi:10.1017/CBO9781139236065 Yau TM, Pagani FD, Mancini DM, et al. Intramyocardial Injection of Mesenchymal Precursor Cells and Successful Temporary Weaning From Left Ventricular Assist Device Support in Patients With Advanced Heart Failure: A Randomized Clinical Trial. JAMA. 2019;321(12):1176–1186. doi:10.1001/jama.2019.2341 Horton NJ, Kim E, Saitz R. A cautionary note regarding count models of alcohol consumption in randomized controlled trials. BMC Med Res Methodol. 2007;7:9. Published 2007 Feb 15. doi:10.1186/1471-2288-7-9 Hu MC, Pavlicova M, Nunes EV. Zero-inflated and hurdle models of count data with extra zeros: examples from an HIV-risk reduction intervention trial. Am J Drug Alcohol Abuse. 2011;37(5):367–375. doi:10.3109/00952990.2011.597280 Cameron AC, Trivedi PK. Regression analysis of count data. Cambridge, UK: Cambridge University Press; 1998.