Individual patient data meta-analysis of continuous diagnostic markers



Similar documents
Knowledge Discovery and Data Mining

Biomarker Discovery and Data Visualization Tool for Ovarian Cancer Screening

Methods for Meta-analysis in Medical Research

Strategies for Identifying Students at Risk for USMLE Step 1 Failure

Evaluation of Diagnostic Tests

Despite its emphasis on credit-scoring/rating model validation,

THE RISK DISTRIBUTION CURVE AND ITS DERIVATIVES. Ralph Stern Cardiovascular Medicine University of Michigan Ann Arbor, Michigan.

Organizing Your Approach to a Data Analysis

Original Article Performance of osteopontin in the diagnosis of malignant pleural mesothelioma: a meta-analysis

Measures of diagnostic accuracy: basic definitions

Non-invasive diagnosis of pleural malignancies: The role of tumour markers

Data Mining Techniques for Prognosis in Pancreatic Cancer

Combined CA125 and Mesothelin Levels for the Diagnosis of Malignant Mesothelioma*

Update on Mesothelioma

STATISTICA Formula Guide: Logistic Regression. Table of Contents

Credit Risk Models. August 24 26, 2010

SOME STATISTICAL ISSUES IN THE DESIGN AND ANALYSIS OF VACCINE CLINICAL TRIALS IN CANCER PATIENTS

Scatter Plots with Error Bars

RESEARCH ARTICLE. Fibulin-3 as a Diagnostic Biomarker in Patients with Malignant Mesothelioma

Health Care and Life Sciences

L Lang-Lazdunski, A Bille, S Marshall, R Lal, D Landau, J Spicer

Journal of Information

DIAPHRAGM. DIAgnostic and Prognostic biomarkers in the Rational Assessment of Mesothelioma

Rulex s Logic Learning Machines successfully meet biomedical challenges.

Systematic Reviews and Meta-analyses

ABSTRACT INTRODUCTION

Bayesian Penalized Methods for High Dimensional Data

S The Difference Between Predictive Modeling and Regression Patricia B. Cerrito, University of Louisville, Louisville, KY

Rare Thoracic Tumours

The interpretation of diagnostic tests

Chapter 4 - Lecture 1 Probability Density Functions and Cumul. Distribution Functions

NHS Diabetes Prevention Programme (NHS DPP) Non-diabetic hyperglycaemia. Produced by: National Cardiovascular Intelligence Network (NCVIN)

Malignant Pleural Diseases Advances Clinicians Should Know F Gleeson

Beware that Low Urine Creatinine! by Vera F. Dolan MSPH FALU, Michael Fulks MD, Robert L. Stout PhD

Big Data and Oncology Care Quality Improvement in the United States

MedicalBiostatistics.com

Occupational respiratory diseases due to Asbestos. Dirk Dahmann, IGF, Bochum

Early Detection Research Network: Validation Infrastructure. Jo Ann Rinaudo, Ph.D. Division of Cancer Prevention

Data Mining. Nonlinear Classification

Identifying SPAM with Predictive Models

National Coverage Determination (NCD) for Tumor Antigen by Immunoassay - CA 125 (190.28)

7. Prostate cancer in PSA relapse

Na#onal Asbestos Forum 2013: Advance in Medical Research on Asbestos- Related Diseases

Retrospective analysis of large scale research screening of construction workers for the early diagnosis of mesothelioma

A Bayesian hierarchical surrogate outcome model for multiple sclerosis

R 2 -type Curves for Dynamic Predictions from Joint Longitudinal-Survival Models

Development and Validation of a Screening Questionnaire for Psoriatic Arthritis

CHILDHOOD CANCER SURVIVOR STUDY Analysis Concept Proposal

How can you unlock the value in real-world data? A novel approach to predictive analytics could make the difference.

Bayes Theorem & Diagnostic Tests Screening Tests

Asbestos-Related Cancer Research and Prevention

Title: A prospective trial evaluating the role of mesothelin in undiagnosed pleural effusions. Authors

Adequacy of Biomath. Models. Empirical Modeling Tools. Bayesian Modeling. Model Uncertainty / Selection

Predicting The Risk Of Rheumatoid Arthritis

How to set the main menu of STATA to default factory settings standards

Social inequalities impacts of care management and survival in patients with non-hodgkin lymphomas (ISO-LYMPH)

Combining diagnostic test results to increase accuracy

Medicare Coverage for Individuals Exposed to Asbestos in Lincoln County, Montana.

Package ROC632. February 19, 2015

How To Use An Osteopontin Prognostic Marker In A Cancer

Screening for psychological distress in inoperable lung cancer patients: an evaluation of the Brief Distress Thermometer (BDT)

A Cross-Sectional Study of Asbestos- Related Morbidity and Mortality in Vermonters Residing Near an Asbestos Mine November 3, 2008

Classification models for predicting a patient clinical status: discussion on the case of the chemotherapy induced neutropenia duration

CGS X X X X X

Non-invasive evaluation of liver fibrosis: current clinical use and next perspectives (chronic hepatitis C)

The 4Kscore blood test for risk of aggressive prostate cancer

Leveraging electronic health records for predictive modeling of surgical complications. Grant Weller

Unmet Needs for Parkinson s Disease Therapeutics

Research Article Pleural Fluid Mesothelin as an Adjunct to the Diagnosis of Pleural Malignant Mesothelioma

RESEARCH COMMUNICATION. HE4 and Mesothelin: Novel Biomarkers of Ovarian Carcinoma in Patients with Pelvic Masses

EVALUATION/PRIORITIZATION CRITERIA: C, L, R, S *to meet requirement 1

Cancer Medicine. Newly established ELISA for N-ERC/mesothelin improves diagnostic accuracy in patients with suspected pleural mesothelioma

Actuarial projections for mesothelioma: an epidemiological perspective Mark Clements, Geoffrey Berry and Jill Shi

Online estimation of the risk of blood contamination

Fingerprinting the Datacenter: Automated Classification of Performance Crises

Basic research methods. Basic research methods. Question: BRM.2. Question: BRM.1

Hologic Selenia Dimensions C-View Software Module. October 24, 2012

Independent Validation of the Prognostic Gene Expression Ratio Test in Formalin Fixed, Paraffin Embedded (FFPE) Mesothelioma Tumor Tissue Specimens

Advances in Diagnostic and Molecular Testing in Prostate Cancer

AFFORDABLE CARE ACT AUTHORIZING LEGISLATION

How to Get More Value from Your Survey Data

Fibulin-3 as a Blood and Effusion Biomarker for Pleural Mesothelioma

Fibulin-3 as a Blood and Effusion Biomarker for Pleural Mesothelioma

Developing Risk Adjustment Techniques Using the System for Assessing Health Care Quality in the

CEA, AFP, CA125, CA153 and CA199 in Malignant Pleural Effusions Predict the Cause

1 page Overview. CONCURRENT 1D, 1E, 1F Biology & Pathogenesis Multi-Modality Immunology 1

J of Evidence Based Med & Hlthcare, pissn , eissn / Vol. 2/Issue 33/Aug. 17, 2015 Page 5063

The risks of mesothelioma and lung cancer in relation to relatively lowlevel exposures to different forms of asbestos

MesoPDT. Photodynamic Therapy for malignant pleural mesothelioma ONCO-THAI. Image Assisted Laser Therapy for Oncology

Effect of Risk and Prognosis Factors on Breast Cancer Survival: Study of a Large Dataset with a Long Term Follow-up

Presenter:Zheng-kun Shi Supervisor:Si-yuan Tang Monica Parry

The diagnostic usefulness of tumour markers CEA and CA-125 in pleural effusion

Appendix G STATISTICAL METHODS INFECTIOUS METHODS STATISTICAL ROADMAP. Prepared in Support of: CDC/NCEH Cross Sectional Assessment Study.

Screening, early referral and treatment for asbestos related cancer

Critical Appraisal of Article on Therapy

An Empirical Evaluation of Guidelines on Prostate-specific Antigen Velocity in Prostate Cancer Detection

PROTOCOL OF THE RITA DATA QUALITY STUDY

Research Article Frequency of Surgery in Black Patients with Malignant Pleural Mesothelioma

FINDING SUBGROUPS OF ENHANCED TREATMENT EFFECT. Jeremy M G Taylor Jared Foster University of Michigan Steve Ruberg Eli Lilly

How To Predict Diabetes In A Cost Bucket

Transcription:

Individual patient data meta-analysis of continuous diagnostic markers J.B. Reitsma Julius Center for Health Sciences and Primary Care UMC Utrecht / www.juliuscenter.nl

Outline IPD benefits Meta-analytical challenges ROC regression based on standardized marker values of cases Example study: mesothelin for diagnosing pleural mesothelioma

What is IPD? Traditional meta-analysis uses published summary (aggregate) data from primary studies Dataset: one row per included study with effect measure(s), their precision and study features Individual Patient Data meta-analysis combines the original (raw, crude) data from individual patients of different studies Dataset: each row is a individual patient with outcomes, patient and study characteristics

IPD benefits: general & diagnostic General IPD benefits: quality control & use of additional data & improve comparability Examining patient-level subgroups: more flexibility & more statistical power Continuous marker: handling of differences in cutoffs & better estimation of the summary ROC curve Statistical approach not straightforward

IPD Challenges Need to adjust for study when combining curves: Example overall shift in marker values Covariate adjusted ROC curve = weighted average of the study-specific curves Covariates affecting the ROC curve: Covariates have an impact on the overall accuracy: affect the separation between cases and controls) Covariate-specific ROC curves

Meta-analytical Challenges

ROC curves: need for adjustment Distribution of marker values of control groups differs between studies Mesothelin example: different kits by different studies

2 ROC curves: Covariate-specific (red) + Pooled ROC curve (black) Prevalence or P(D=1 Z=0) is 25% Prevalence or P(D=1 Z=1) is 75%

2 ROC curves: Covariate-specific (red) + Pooled ROC curve (black) Prevalence in Z=0 is 50% Prevalence in Z=1 is 50%

Cumulative distribution of Standardized Marker Approach

Standardized case marker approach Based on the approach of Janes & Pepe to adjust ROC curves for covariates in a single study Modified steps of their approach: 1. Estimate the cumulative distribution of the marker among controls as a function of study 2. Calculate the percentile values of the cases based on this distribution conditional on study 3. The cumulative distribution of these percentile values of cases is then the weighted ROC curve Janes & Pepe. Adjusting for covariates in studies of diagnostic, screening, or prognostic markers: an old concept in a new setting. Am J Epidmiol 2008.

Step 1 & 2: Percentile values of cases based on control distribution Specify how study (Z) affects the distribution of Y among controls: = β + β Z + ε Y 0 1 Calculate percentile values of cases using the control distribution standardized on Z as reference: pv DZ = Fˆ ( Y ( ˆ β + β 0 1Z )) Interpretation percentile values (pv): If a case has value Y and 20% of the corresponding controls have a value higher than Y, the pv =0.2 Also known as placement value

Calculate ROC curve using PV of cases ROC curve is the cumulative pv distribution of cases: ROC( f ) = P(1 pv f DZ ) To fit ROC, select a set (p) of discrete False Positive Rate values (f) spanning the interval (0,1) For each case create, n p records where the outcome is either U=1 or 0 A binary regression model with link g, outcome U and covariate g -1 (f) will generate the binormal (g=standard normal CDF) or bilogistic ROC curve Bootstrap to obtain standard errors or CI bands

Case Study: Mesothelin for Mesothelioma

Mesothelin for malignant mesothelioma Mesothelioma is asbestos-related malignancy with a poor prognosis in general Early diagnosis improves outcome, but diagnosis typically occurs in advanced stage due to non-specific complaints & slow onset Biomarkers may lead to an earlier diagnosis Problems identified in traditional reviews: - use of different cut-offs - non-standardized measurements - impact clinical subgroups Hollevoet et al. J Clin Oncol 2012

Hollevoet et al. Journal of Clinical Oncology 2012

Mesothelin example Hollevoet et al. J Clin Oncol 2012

Mesothelin example

Mesothelin example AUC = average of percentile values of cases Hollevoet et al. J Clin Oncol 2012

Discussion & Conclusion IPD data generate much more insight in the distribution of marker values across studies Better estimation of summary ROC curve and comparison of ROC curves between subgroups IPD no cure for poorly designed studies Approach of placement values appears attractive: flexible and use of existing techniques Comparison with alternative approaches needed