Signal Detection for Medical Scientists: Likelihood Ratio Test-based Methodology

Description

Signal Detection for Medical Scientists: Likelihood Ratio Based Test-Based Methodology presents the data mining techniques with focus on likelihood ratio test (LRT) based methods for signal detection. It emphasizes computational aspect of LRT methodology and is pertinent for first-time researchers and graduate students venturing into this interesting field. The book is written as a reference book for professionals in pharmaceutical industry, manufactures of medical devices, and regulatory agencies. The book deals with the signal detection in drug/device evaluation, which is important in the post-market evaluation of medical products, and in the pre-market signal detection during clinical trials for monitoring procedures. It should also appeal to academic researchers, and faculty members in mathematics, statistics, biostatistics, data science, pharmacology, engineering, epidemiology, and public health. Therefore, this book is well suited for both research and teaching. Cover Half Title Title Page Copyright Page Dedication Contents Preface I. LRT Methodology 1. Introduction 1.1 Background of Data Mining for Safety Signals 1.2 Post-market Safety Databases 1.3 openFDA for FAERS Database 2. Data Mining Methods for Signal Detection 2.1 Review of Common Frequentist Methods 2.1.1 Reporting odds ratio (ROR) 2.1.2 Proportional reporting ratio 2.1.3 Information component 2.1.4 Chi-squared test 2.2 Review of Common Bayes and Empirical Bayes 2.2.1 Bayesian confidence propagation neural network (BCPNN) 17 2.2.2 Multi-item gamma poisson shrinker (MGPS) 2.2.3 New IC 2.2.4 Simplified Bayes 2.2.5 Poisson-DP method 2.3 Notes and Discussion 2.4 Appendix 2.4.1 Relationship between new IC and sB methods 2.4.2 Polya-urn representation of the Dirichlet process prior—Conditional distributions 3. Basic LRT Method 3.1 Method Development 3.1.1 Derivation of likelihood ratio test statistic 3.1.2 MLR incorporating covariate information 3.1.3 Hypothesis testing 3.2 Simulation 3.2.1 Data simulation 3.2.2 Performance characteristics 3.2.3 Simulation results 3.3 Applications to FAERS Data 3.3.1 Montelukast analysis 3.3.2 Heparin analysis 3.4 Notes and Discussion 3.5 Further Discussion Points 3.6 LRT Tool in openFDA 3.7 Tables and Figures 4. LRT Methods for Drug Classes 4.1 Signals for Drug Class 4.1.1 Drug signals in FAERS data 4.1.2 Signals of single drug-AE combination 4.1.3 Performance evaluation using simulation 4.2 Ext-LRT Signal Detection using Double Maximum 4.2.1 Ext-LRT statistic and statistical inference 4.2.2 Application of Ext-LRT method 4.3 LRT for Identification of Signals Using Weight Matrix 4.3.1 Modi ed LRT statistic and statistical inference 4.3.2 Applications in detection of signals—Collection of drugs 4.4 Notes and Discussion 4.5 Appendix: Tree-based Scan Statistic Method for AE Signals Evaluation 4.5.1 Summary of the method 4.5.2 Relationship to the LRT method 4.6 Tables and Figures 5. ZIP-LRT Method for Modeling Extra-zeros 5.1 ZIP-LRT Model 5.2 EM Algorithm for ZIP Model 5.3 Likelihood Ratio Test Statistic 5.4 Strati ed ZIP LRT 5.5 Hypothesis Testing 5.6 Simulation Study 5.6.1 Data simulation 5.6.2 Simulation results 5.7 Application to 2006-2011 FAERS Data 5.7.1 Estimated percentage of true zeros 5.7.2 Application to selected drugs 5.8 Notes and Discussion 5.9 Appendix: Proof of Theorem 5.5.1 5.10 Tables and Figures II. Extensions 6. LRT Method for Active Safety Surveillance with Exposure Information 6.1 Medical Background Based on the Example Data 6.1.1 Data structure 6.1.2 De nitions of drug exposure 6.1.3 De ning multiple looks 6.2 Longitudinal LRT Method for Active Safety Surveil Surveillance 6.2.1 LongLRT for comparing multiple events using event-time 6.2.2 SeqLRT for comparing two drugs and one AE with single occurrence using person-time 6.2.3 LongLRT for comparing multiple drugs and one AE with recurrence using exposure-time 6.2.4 Assumption of independence 6.3 Statistical Inference with Multiple Looks 6.4 Applications 6.4.1 Safety signals (among multiple AEs) by drug 6.4.2 Safety signals for the rst occurrence of a composite AE and two drugs using SeqLRT 6.4.3 Safety signals for multiple occurrences of a composite AE and two drugs using LongLRT 6.4.4 Safety signals for a composite AE with recurrence from multiple drugs using LongLRT 6.5 Simulation Study for Longitudinal LRT Methods 6.5.1 Data simulation 6.5.2 Performance characteristics 6.5.3 Simulation results 6.6 Discussion 6.7 Appendix 6.7.1 De nition of the composite AE 6.7.2 Data structure for the I x J matrix 6.7.3 Independence of the parameters 6.8 Tables and Figures 7. LRT-based Methodologies for Analysis of Multiple Studies 7.1 Background and Motivation 7.2 Methods 7.2.1 Summary of basic LRT with and without exposure information 7.2.2 LRT analysis approaches for signal detection from multiple studies 7.3 Applications 7.3.1 Analysis of PPI data with two drugs and a composite AE 7.3.2 Analysis of lipiodol data with one drug and multiple AEs 7.3.3 Summary of the two examples 7.4 Simulation 7.4.1 Simulation setup 7.4.2 Results 7.5 Discussion 7.6 Tables and Figures III. Additional Frameworks 8. LRT Methods in Medical Device Safety Evaluation 8.1 Background 8.2 LRT Methods for Device Data from MDR 8.2.1 MDR data for LVADs 8.2.2 Exploration of the device data 8.2.3 Remarks 8.3 Safety Evaluation in Treatment vs. Control Group for Medical Device 8.3.1 Data source 8.3.2 Statistical models 8.3.3 Conventional Z-test with P-value adjustment 8.3.4 Modified LRT 8.3.5 Results 8.3.6 Performance evaluation using simulation 8.3.7 Remarks 8.3.8 Appendix: LRT and tree-based scan statistic method for comparing treatment vs. control 8.3.8.1 Summary of tree-based scan method 8.3.8.2 Relationship between LRT method and tree-based scan statistic method 8.4 Spatial-Cluster Signal Detection in Medical Devices using LRT 8.4.1 Background 8.4.2 Spatial-LRT with exposure information 8.4.3 Medical device safety database considerations 8.4.4 Illustrations 8.4.5 Remarks 8.5 Weighted LRT Method to Device Data From Multiple Sites 8.5.1 Background 8.5.2 Data structure for data with multiple studies 8.5.3 Weighted LRT method 8.5.4 Results 8.5.5 Remarks 8.6 Tables and Figures 9. LRT Method for Multiple-Site Device Data with Continuous Outcomes 9.1 Background 9.2 Data Structure and Problem Formulation 9.2.1 Data structure 9.2.2 Problem formulation 9.3 Normal-LRT method 9.3.1 MLRs of parameters uij ; vij , and σ2j 9.3.2 Test statistic MLLR 9.3.3 Permutation-based empirical distribution 9.4 Application 9.5 Simulation 9.5.1 Data simulation 9.5.2 Type-I error 9.5.3 Power, sensitivity, and false discovery rate 9.6 Notes and Discussion 9.7 Appendix 9.7.1 Data generation process for hypothetical case study 9.7.2 Simulation for data including regions without data 9.8 Tables and Figures 10. Use of LRT in Site Selection 10.1 Background 10.2 Data and P-value Matrix 10.3 Statistical Approaches for Site Ranking Generation from Pvalue Matrix 10.3.1 Fisher combination approach 10.3.2 LRT approach 10.4 Application to a Case Study 10.5 Comprehensive Simulation Study 10.5.1 Performance characteristics 10.5.2 Results 10.6 Notes and Discussion 10.7 Appendix 10.7.1 Statistical tests for individual variables 10.7.2 Simulation of P-values for site signals 10.7.3 Simulation study for correlation P-values 10.8 Tables and Figures Bibliography Subject Index