Statistics and Machine Learning Methods for EHR Data: From Data Extraction to Data Analytics

Cover
Half Title
Series Page
Title Page
Copyright Page
Contents
Preface
About the Editors
Contributors
1. Introduction: Use of EHR Data for Scientific Discoveries--Challenges and Opportunities
	1.1. Real-World Data and Real-World Evidence: Big Data in Practice
	1.2. Use of EMR/EHR Database for Research and Scientific Discoveries: Procedure and Life Cycle
		1.2.1. Initiate a Project
		1.2.2. Data Queries and Data Extraction
		1.2.3. Data Cleaning
		1.2.4. Data Pre-Processing or Processing
		1.2.5. Data Preparation
		1.2.6. Data Analysis, Modeling and Prediction
		1.2.7. Result Validation
		1.2.8. Result Interpretation
		1.2.9. Publication and Dissemination
	1.3. Challenges and Opportunities
	References
2. EHR Project Management
	2.1. Introduction
		2.1.1. What is Project Management?
		2.1.2. Why We Need Project Management?
		2.1.3. Project Management Goals and Principles
	2.2. Project and Sub-Project in EHR Research
	2.3. Data, Code and Product Management
		2.3.1. Data Loss Prevention
		2.3.2. Naming Conventions
		2.3.3. Version Control
		2.3.4. Coding Convention
			Object-Oriented or Non-Object-Oriented Programming
		2.3.5. Document Management: Data Analysis Report, Papers and Read-Me Documents
	2.4. Team/People Management
		2.4.1. How to Form a Team: What Expertise is Needed for EHR Projects?
		2.4.2. How to Efficiently Manage a Multidisciplinary Team?
		2.4.3. Task Management
	2.5. Management Methods and Software Tools
	2.6. An Example of a Data Management Framework
		2.6.1. Folder Management
			Naming
			Structure
				Main Folders
					CBD_HS
					Public_Folder
					Admin
					Useful_Info
				Group Folders
				Project Folders
				Sub_Project Folders
		2.6.2. File Management
			Naming
			Structure
				File Submission
		2.6.3. User Management
			User Groups
		2.6.4. Data Management Framework
	2.7. Discussion and Summary
	2.8. Appendix--File Submission Form
	Note
	References
3. EHR Databases and Data Management: Data Query and Extraction
	3.1. Introduction
	3.2. EHR/EMR Database Availability and Access
	3.3. EHR/EMR Database Design and Structure: Database Queries
		3.3.1. Database Construction
		3.3.2. Traditional Relational Database System
		3.3.3. Distributed Database System
	3.4. Data Extraction
		3.4.1. Define Inclusion/Exclusion Criteria for Data Extraction
		3.4.2. Phenotyping: Cohort Identification
	3.5. Data Extraction Report
	3.6. Illustration Example: Subarachnoid Hemorrhage (SAH) Project
		3.6.1. EHR Database Design and Construction
		3.6.2. SAH Cohort Identification and Data Extraction
		3.6.3. Data Extraction Report
		3.6.4. Potential Data Extraction Pitfalls and Errors with Solutions
	References
4. EHR Data Cleaning
	4.1. Introduction
	4.2. Review of Current Data Cleaning Methods and Tools
		4.2.1. Data Wranglers
		4.2.2. Data Cleaning Tools for Specific EHR Datasets
		4.2.3. Data Quality Assessment
	4.3. Common EHR Data Errors and Fixing Methods
		4.3.1. List of Common Errors in an EHR Database
		4.3.2. Demographics Table
			Multiple Race and Gender
			Multiple Patient Keys for the Same Encounter ID
			Multiple Calculated Birth Date
		4.3.3. Lab Table
			Developing Conversion Map
				Conversion Map ID
				Convert To
				Conversion Equation
				The Lower Limit and Upper Limit
			Lab Date and Time
			User Input Form and Report Generator
			Output
		4.3.4. Clinical Event Table
			Variable Combining
			Information Recovery
				A Case Study
			Overlap of Different Tables
			Correction of Misinformation
		4.3.5. Diagnosis and Medication Table
		4.3.6. Procedure Table
			Introduction to the Procedure Code Data
			Procedure Table Data Cleaning
	4.4. Discussion
	Acknowledgments
	Notes
	References
5. EHR Data Pre-Processing and Preparation
	5.1. Introduction
		5.1.1. Definition of Data Pre-Processing/Processing
		5.1.2. Definition of Data Preparation
	5.2. Data Pre-Processing
		5.2.1. Tidy Data Principles
			Variable Encoding
		5.2.2. Feature Extraction: Derived Variables
		5.2.3. Dimension Reduction
			Variable Grouping or Clustering
			Principle Component Analysis (PCA)
			Embedding and Deep Learning
		5.2.4. Missing Data Imputation
	5.3. Data Preparation
		5.3.1. Define the Endpoint or Outcome
		5.3.2. Process Medical Record Timestamps
		5.3.3. Define the Encounter Time Interval
		5.3.4. Encounter Combination
		5.3.5. Define Comparison Groups
		5.3.6. Cohort Refining
		5.3.7. Leakage Detection
		5.3.8. Data Preparation for Different Analysis Purposes
	5.4. Data Processing/Preparation Errors and Pitfalls with Solutions
	5.5. Data Pre-Processing and Preparation Report
	5.6. Summary
	References
6. Missing Data Issues in EHR
	6.1. Introduction and Overview
	6.2. Missing Data Mechanisms
	6.3. Methods for Incomplete EHR Data
		6.3.1. Naïve Method
		6.3.2. Imputation Using Statistical Models
		6.3.3. Machine Learning and Deep Learning Models
		6.3.4. Choice of Best Method for EHR Data
	6.4. Case Study
		6.4.1. Missing Condition in EHR Data
		6.4.2. Missing Imputation in EHR Datasets
		6.4.3. Evaluating the Performance of Imputation Methods and Thresholds
	6.5. Discussion and Conclusion
	References
7. Causal Inference and Analysis for EHR Data
	7.1. Introduction
		7.1.1. Why Causal Inference
		7.1.2. Overview of Causal Inference Methods: Rubin Causal Model (RCM)
		7.1.3. Basic Framework in Causality: Potential Outcome Framework
			Average and Individual Treatment Effects
	7.2. Propensity Scoring
		7.2.1. Brief Introduction
		7.2.2. Propensity Scoring for Binary Treatments
		7.2.3. Propensity Scoring for Multiple Treatments
		7.2.4. Propensity Scoring for Ordinal Treatments
		7.2.5. Propensity Score Estimation for Complex Data Sets
		7.2.6. Illustration Example: Subarachnoid Hemorrhage (SAH) Project
	7.3. Mediation Analysis
		7.3.1. Introduction to Mediation Analysis
		7.3.2. The Product Method
		7.3.3. The Difference Method
		7.3.4. Other Considerations
	7.4. Instrumental Variables Networks for Treatment Effect Estimation in the Presence of Unmeasured Confounders
		7.4.1. Instrumental Variables Frameworks
		7.4.2. Two-Stage Least Square Methods with Linear Models
			Simple Linear Models
			Covariance Analysis
			Generalized Least Square Estimator
			Two-Stage Least Square Method
			Nonlinear Models for Two-Stage Least Squares Approach
	7.5. Learning Treatment Effect by Generative Adversarial Networks
		7.5.1. Introduction
		7.5.2. CGANs as a General Framework for Estimation of Individualized Treatment Effects
			The Architecture of CGANs for Generating Potential Outcomes
			CGANs for Estimating ITEs
			CGANs for Estimating ITEs in Survival Analysis
		7.5.3. Wasserstein GANs for Estimation of Individualized Treatment Effects
		7.5.4. MisCGANs for Estimation of Individualized Treatment Effects
			The General Process for Incompletely Observed Data
			MisGAN for Counterfactual Imputation
		7.5.5. Optimal Treatment Selection
			Sparse Techniques for Biomarker Identification
			Biomarker Identification for Optimal Treatment Selection
	7.6. Deconfounder in Estimation of Treatment Effects
		7.6.1. Introduction
		7.6.2. Causal Models with Latent Confounders
		7.6.3. Adversarial Learning Confounders
		7.6.4. Loss Function and Optimization for Estimating ITEs in the Presence of Confounders
	7.7. Targeted Maximum Likelihood Estimation
	7.8. Supplementary Note A
		7.8.1. Wasserstein GAN
			A1 Different Distances
				A1.1 Maximum Likelihood Estimation
				A1.2 Total Variation (TV) Distance
				A1.3 The Kullback-Leibler (KL) Divergence
				A1.4 The Jenson-Shannon (JS) Divergence
				A1.5 Earth Mover (EM) or Wasserstein Distance
			A2 Wasserstein GAN
			A3 Algorithm (WGAN)
	References
8. EHR Data Exploration, Analysis and Predictions: Statistical Models and Methods
	8.1. Introduction
		8.1.1. Statistical Challenges for EHR Data
		8.1.2. Overview of Existing Methods
	8.2. Data Exploration and Visualization
	8.3. Statistical Models for EHR Data
		8.3.1. Contingency Tables
		8.3.2. Chi-Square Test
		8.3.3. Hypergeometric Test
	8.4. GLM
	8.5. Survival Model
	8.6. Mixed-Effect Models
	8.7. Time Series Analysis
		8.7.1. AR, MA and ARMA Model
		8.7.2. Gaussian Process
	8.8. Variable Selection Methods
		8.8.1. Stepwise Variable Selection
		8.8.2. Purposeful Variable Selection
		8.8.3. SIS
		8.8.4. Penalty-Based Methods
	8.9. Divide-and-Conquer Method
	8.10. Validation
	8.11. Results and Examples
	8.12. Discussions and Conclusions
	References
9. Neural Network and Deep Learning Methods for EHR Data
	9.1. Introduction
	9.2. Deep Learning Methods for EHR Data
	9.3. Deep Learning Software Tools and Implementation
	9.4. Application Examples
		Case Study 1: Application of MLP for Mortality Prediction
		Case Study 2: Application of RNN for Heart Failure Prediction for Hypertension Patients
			Experimental Setting
			RNN Prediction Results
	9.5. Discussion
	References
10. EHR Data Analytics and Predictions: Machine Learning Methods
	10.1. Machine Learning Overview
	10.2. Machine Learning Methods
		Random Forest
		Extremely Randomized Tree
		Gradient Boosting
		XgBoost
		Support Vector Machine (SVM)
	10.3. Machine Learning Software Tools
		H2O
		Caret
		TPOT
		Auto-sklearn
	10.4. Application Example: SAH Project
		Prediction Scenarios
		Predictors and Outcome Data
		Model Training
		Result of Outcome Prediction
		Evaluation of Model Performance
	10.5. Conclusion and Recommendation
	References
11. Use of EHR Data for Research: Future
	11.1. Future EHR Research
	11.2. Post-Research Practice
	11.3. Summary
	References
Index