Comparisons of Modeling Methods on Longitudinal and Survival Data: Identifying Use of Repeat Biomarker Measurements to Predict Time-to-Event Outcome in.

Slides:



Advertisements
Similar presentations
Allison Dunning, M.S. Research Biostatistician
Advertisements

Grading the Strength of a Body of Evidence on Diagnostic Tests Prepared for: The Agency for Healthcare Research and Quality (AHRQ) Training Modules for.
Study Design 121 Relapsing-remitting MS patients randomized to –Stress Management Therapy MS active treatment* 16 individual sessions conducted over 24.
Breakout Session 4: Personalized Medicine and Subgroup Selection Christopher Jennison, University of Bath Robert A. Beckman, Daiichi Sankyo Pharmaceutical.
Continuous versus Intermittent Androgen Deprivation Therapy for Prostate Cancer Robert Dreicer, M.D., M.S., FACP, FASCO Chair Dept of Solid Tumor Oncology.
Staff Oncologist, Mayo Clinic Arizona
Introduction of Cancer Molecular Epidemiology Zuo-Feng Zhang, MD, PhD University of California Los Angeles.
OncoTracker James Berenson, MD President and CEO November 2014.
Analysis of Complex Survey Data
Asking Questions Robert M. Rowell, DC, MS.
Marshall University School of Medicine Department of Biochemistry and Microbiology BMS 617 Lecture 12: Multiple and Logistic Regression Marshall University.
Dr. Abdulaziz BinSaeed & Dr. Hayfaa A. Wahabi Department of Family & Community medicine  Case-Control Studies.
Essentials of survival analysis How to practice evidence based oncology European School of Oncology July 2004 Antwerp, Belgium Dr. Iztok Hozo Professor.
Dr Laura Bonnett Department of Biostatistics. UNDERSTANDING SURVIVAL ANALYSIS.
9/17/ Access WHI: Proposing Analyses and Ancillary Studies Andrea Z. LaCroix, PhD Professor of Epidemiology and Co-PI WHI Clinical Coordinating Center.
Director of Scientific Affairs
A role for lipids and statins in breast cancer risk and prevention? Dr. Mieke Van Hemelrijck Senior Lecturer in Cancer Epidemiology 3 August 2015.
Screening and Diagnostic Testing Sue Lindsay, Ph.D., MSW, MPH Division of Epidemiology and Biostatistics Institute for Public Health San Diego State University.
HSRP 734: Advanced Statistical Methods July 17, 2008.
HSRP 734: Advanced Statistical Methods July 31, 2008.
RevMan for Registrars Paul Glue, Psychological Medicine What is EBM? What is EBM? Different approaches/tools Different approaches/tools Systematic reviews.
Poster Title ABSTRACT #59 Cell cycle progression genes differentiate indolent from aggressive prostate cancer. Steven Stone 1 Jack Cuzick 2, Julia Reid.
1 Lecture 6: Descriptive follow-up studies Natural history of disease and prognosis Survival analysis: Kaplan-Meier survival curves Cox proportional hazards.
SPRINT What Remains Unanswered? Where Do We Go From Here? Embargoed Until 2 p.m. ET, Monday, Nov. 9, 2015.
Armando Teixeira-Pinto AcademyHealth, Orlando ‘07 Analysis of Non-commensurate Outcomes.
BC Cancer Agency CARE & RESEARCH Breast Cancer Mortality After Screening Mammography in British Columbia Women Andrew J. Coldman, Ph.D. Norm Phillips,
Promoting Patient Involvement in Medication Decisions David H. Hickam, MD, MPH Professor, Dept. of Medicine Oregon Health & Science University Portland,
EBM --- Journal Reading Presenter :黃美琴 Date : 2005/10/27.
© 2010 Jones and Bartlett Publishers, LLC. Chapter 12 Clinical Epidemiology.
Marshall University School of Medicine Department of Biochemistry and Microbiology BMS 617 Lecture 13: Multiple, Logistic and Proportional Hazards Regression.
R4 Jae Joon Han.
RTI International RTI International is a trade name of Research Triangle Institute. Selecting Endpoints for Clinical Trials.
Importance of Exercise with Diabetes
Population-based study on the impact of familial form of Waldenström’s macroglobulinemia on overall survival Vilhjálmur Steingrímsson1, Sigrún Helga.
Advance Care Planning in dementia Dr Karen Harrison Dening Head of Research & Evaluation Dementia UK GSF 2016.
Results from the International, Randomized Phase 3 Study of Ibrutinib versus Chlorambucil in Patients 65 Years and Older with Treatment-Naïve CLL/SLL (RESONATE-2TM)1.
Sofija Zagarins1, PhD, Garry Welch1, PhD, Jane Garb2, MS
Minimal Residual Disease (MRD) in Multiple Myeloma
Descriptive study design
Anastasiia Raievska (Veramed)
Interpretation of effect estimates in competing risks survival models: A simulated analysis of organ-specific progression-free survival in a randomised.
IFM/DFCI 2009 Trial: Autologous Stem Cell Transplantation (ASCT) for Multiple Myeloma (MM) in the Era of New Drugs Phase III study of lenalidomide/bortezomib/dexamethasone.
Director Department of Pediatric Hematology & Oncology Delhi, INDIA.
Evidence-Based Medicine
Meta-analysis of joint longitudinal and event-time outcomes
Critical Reading of Clinical Study Results
Knowledge l Action l Impact
CDC Guidelines for Use of QuantiFERON®-TB Gold Test
Towards UK poSt Arthroplasty Follow-up rEcommendations: UK SAFE
Jeffrey E. Korte, PhD BMTRY 747: Foundations of Epidemiology II
Dose-finding designs incorporating toxicity data from multiple treatment cycles and continuous efficacy outcome Sumithra J. Mandrekar Mayo Clinic Invited.
11/20/2018 Study Types.
Sampling Studies for Longitudinal Functional Data
Testosterone Deficiency
Chen S, Dong Y, Kiuchi MG, et al
Day 2 Applications of Growth Curve Models June 28 & 29, 2018
Fig. 1. scRNA-seq applications in cancer medicine.
Prepared by staff in Prevention and Cancer Control.
Statistical Considerations for Using Multiple Databases to Build a Biomarker Probability Tool Shijia Bian MS1; Wenting Wang PhD1; Nancy Maserejian.
Role for XRT in treatment of early stage Follicular lymphoma?
Clinical Implications
Level of Evidence Lecture 4.
THE LANCET Oncology Volume 19, No. 1, p27–39, January 2018
Use of Piecewise Weighted Log-Rank Test for Trials with Delayed Effect
Design and Analysis of Survival Trials with Treatment Crossover, delayed treatment effect and treatment dilution Presenter: Xiaodong Luo– R&D-SANOFI US.
T. Tzellos1,2; H. Yang3; F. Mu3; B. Calimlim4; J. Signorovitch3
Björn Bornkamp, Georgina Bermann
Evidence Based Diagnosis
Logical Inference on Treatment Efficacy When Subgroups Exist
Planning and analyzing clinical trials with competing risks: Recommendations for choosing appropriate statistical methodology Presented by Misun Yu.
Presentation transcript:

Comparisons of Modeling Methods on Longitudinal and Survival Data: Identifying Use of Repeat Biomarker Measurements to Predict Time-to-Event Outcome in Cancer Research Meng Ru, MS Erin Moshier, MS Vernon Wu, MD Ajai Chari, MD Madhu Mazumdar, PhD Icahn School of Medicine at Mount Sinai Tisch Cancer Institute July 30, 2018

Time-to-Event Outcome: MOTIVATION Disease: Smoldering Multiple Myeloma (SMM) Data: Biomarkers (Hemoglobin, etc) measured longitudinally during monitoring window) Progression Objective: Modeling Methods Longitudinal Data: Biomarker Levels Link Time-to-Event Outcome: Progression This project was inspired by a clinical collaborative study with the multiple myeloma group at Mount Sinai. As a cancer biostatistician, we are constantly reminded by our collaborators of the importance of biomarkers. It’s everywhere and it tells so much not only about patients’ current stage but also the likelihood of a future event which might help clinical decision making. Our study here was designed to identify longitudinal profiles of various biomarkers, collected from smoldering multiple myeloma patients, that were associated with an increased risk of progression to active myeloma.. We hope to find out the patients with early time to progression who may benefit more from early interventional therapies. There are several modeling available to associate longitudinal profiles and time to event outcomes. of The retrospective study recently published in Blood Advances sets out to identify longitudinal profiles, of various biomarkers collected during routine monitoring of smoldering multiple myeloma patients (such as hemoglobin, free light chain ratios, bone marrow plasma cells and m-protein), that were associated with an increased risk of progression to active myeloma. The clinical importance of this research question is that patients with biomarker profiles associated with early time to progression may benefit from early interventional therapies. Historically, many risk prediction models only accounted for the biomarker levels at diagnosis, but current research in many cancer disease groups including multiple myeloma, lymphoma and prostate cancer have focused on not only the static value of the biomarker at a particular point in time but the rate of change or trajectory of the biomarker over time which can be clinically important. The purpose of this project is to review and compare the various methods available for associating longitudinal biomarker profiles and time to event outcomes. Prediction

Longitudinal Analysis METHODS Longitudinal Analysis Link Survival Analysis Time-dependent Cox Model (TDCM): TDCM with time-varying covariate Mixed Model(MM): Joint Model (JMM) vs. Two-Stage Approach (TSMM) Longitudinal: Mixed model on (current value/current value + current slope) Survival: Piecewise PH model (JMM); TDCM (TSMM) Latent Class Model (LCM): Joint Model (JLCM) vs. Two-Stage Approach (TSLCM) Longitudinal: Latent class model (JLCM); Group-based trajectory model (TSLCM) Survival: Piecewise PH model (JLCM); Cox PH (TSLCM) The go-to method for many statisticians is the time-dependent cox model with the repeated measures of biomarker entered as a time-varying covariate, however this method does not specify a particular model for the longitudinal profile component of the data. Two more sophisticated methods in the literature now are the joint and two stage modeling approaches. As the names suggest the joint model simultaneously models the longitudinal and survival components of the data while the two stage approach separately models the longitudinal and survival components of the data in two different stages. Both longitudinal and survival components can be specified with different submodels. the longitudinal submodel is usually estimated with either a mixed or latent class model. Latent class modeling is not currently as popular as mixed modeling, it assumes the biomarker profile is better described by multiple distinct trajectories rather than the single trajectory assumed by a mixed model. We do find that it has a much more clinically meaningful interpretation in many settings and lends itself well to risk stratification studies. We applied all methods to our dataset, which we will share in more detail in our poster, and compared the estimates obtained. When using a mixed model you assume that the biomarker profile can be appropriately described with a single trajectory whereas a latent class model assumes that the biomarker profile is better described by multiple distinct trajectories. Latent class modeling is not currently as popular as mixed modeling, however we do find that it has a much more clinically meaningful interpretation in many settings and lends itself well to risk stratification studies. With the mixed model method you can use various specifications of the longitudinal profile (current value, current slope, cumulative slope, etc.) to test for association each with answering a different research question. The joint modeling approach is the gold standard however the two-stage approach remains popular due to its convenience to implement and interpret.

RESULTS TDCM vs. Joint Model/ Two-stage: TDCM produced larger estimates of HRs and SEs than Joint Model and Two-stage approaches for current value specification. Joint Model vs. Two-stage: Two-stage yielded smaller estimates of HRs and SEs than Joint Model with both MM and LCM approaches. (simulation: Sweeting et. al (2011)) *Forest plot of HR in Poster. We found that the time dependent cox tends to produce larger estimates of the hazard ratio than joint and two-stage modeling approaches using mixed modeling to estimate the longitudinal component. The two-stage approach produced smaller and more conservative estimates of the hazard ratio than the joint modeling approach with both mixed models and latent class analysis methods used to model the longitudinal profile. This finding also corresponds to another simulation study by Sweeting et al. We present all hazard ratios from all methods in forest plot in our poster. The time dependent cox model can only be compared to the current value specification of the longitudinal profile estimated with the mixed modeling approach; it’s not comparable to results from latent class modeling of the longitudinal profile.

RECOMMENDATIONS A* C,A A S D Scenarios TDCM JM TS JMM JLCM TSMM TSLCM Events concurrent to (C) or after (A) longitudinal exposure window A* C,A A Single (S) or Distinct (D) trajectory best captures longitudinal process S D To summarize, we describe some scenarios where the methods discussed may be most suitable. If the outcome event can only occur after the longitudinal exposure window, then all methods are appropriate to varying degrees of accuracy of estimates and assuming if missingness in the longitudinal measures is not a problem for time dependent cox. However, if the outcome event can occur concurrent to the longitudinal exposure window then time dependent cox and two-stage approach (without landmark). With either the joint or two stage approaches we recommend using the latent class model for the longitudinal component if your research goal is to identify heterogeneous risk groups. One thing that should be noted is that our example here is a cancer study, but the application for joint models is not limited to cancer research, people in the field of chronic disease, psychology and social science. with larger estimates of the association likely with the two stage approach and TDCM only appropriate. One thing that should be noted is that our example here is a cancer study, but the application for joint models is not limited to cancer research, people in the field of chronic disease, psychology and social science have been using this method for quite a while to do research like, using maternal depression symptoms during pregnancy and the first 12 months postpartum to predict children’s behaviors like hyperactivity/inattention and physical aggression at a young age.

THANKS, QUESTIONS? Poster Number #9 Session #215942 Time: 7/30/2018 2:00 PM – 2:45 PM Location: Vancouver Convention Centre, West Hall B. Presenter Contact Info: Meng Ru, MS meng.ru@mountsinai.org