Clinical Applications of Artificial Intelligence in Real-World Data

Contents
Data Processing, Storage, Regulations
Biomedical Big Data: Opportunities and Challenges (Overview)
	Abstract
	References
Quality Control, Data Cleaning, Imputation
	Abstract
	1	Introduction
		1.1	Quality Control
		1.2	Data Preparation
	2	Missing Data
		2.1	Types of Missing Data Mechanisms
		2.2	Types of Missing Data Patterns
		2.3	A Bird’s Eye View on Missing Data Methods
		2.4	Introduction of Case Study Data (MIMIC-III)
	3	Single Imputation Methods
	4	Multiple Imputation Methods
		4.1	Joint Modelling Imputation
		4.2	Conditional Modelling Imputation
		4.3	Machine Learning Imputation
			4.3.1 Nearest Neighbor Methods
			4.3.2 Matrix Completion Methods
			4.3.3 Tree-Based Ensembles
			4.3.4 Support Vector Machines
			4.3.5 Neural Networks
		4.4	Analyzing and Combining the Imputed Datasets
	5	Non-imputation Methods
		5.1	Complete Case Analysis
		5.2	Likelihood-Based Methods
		5.3	Pattern Submodels
		5.4	Surrogate Splits
		5.5	Missing Indicator
	6	Imputation of Real-World Data
		6.1	Informative Missingness
		6.2	Longitudinal and Sequence Data
		6.3	Choosing an Appropriate Imputation Method
	7	Summary
	References
Data Standards and Terminology Including Biomedical Ontologies
	Abstract
	1	Introduction
		1.1	The Need for Standards and Their Application
	2	Controlled Clinical Terminologies and Clinical Classifications Systems
		2.1	SNOMED-CT
		2.2	International Classification of Disease (ICD)
	3	Defining Diseases in Electronic Health Records
		3.1	The Need for Aggregated Code Representations
		3.2	Bridging Molecules to Phenotypes
	4	Application of Standards to Aid Machine Learning
	5	Future directions
	References
Data Integration and Harmonisation
	Abstract
	1	Introduction to Common Data Models
		1.1	Introduction
		1.2	Common Data Models
		1.3	Common Data Models in the Biomedical Domain
		1.4	Benefits of Harmonisation to a CDM
	2	The OMOP CDM
		2.1	History
		2.2	The OMOP CDM
		2.3	The OMOP Standardised Vocabularies
		2.4	Use Cases from the OHDSI Community
	3	General Pipeline of the Data Source Transformation to OMOP CDM Process
		3.1	Source Preparation
		3.2	Environment Setup
		3.3	Data Profiling
		3.4	Syntactic Mapping
		3.5	Semantic Mapping
		3.6	Validation
	4	Challenges of Harmonisation
		4.1	Data and Information Loss
			4.1.1 Data Loss
			4.1.2 Information Loss
		4.2	Data Privacy and Sensitivity
	References
Natural Language Processing and Text Mining (Turning Unstructured Data into Structured)
	1 Introduction
	2 What Is Natural Language Processing
		2.1 Text Preprocessing
		2.2 Text Classification
		2.3 Text Clustering and Topic Modeling
		2.4 Information Extraction
		2.5 Text Representations
	3 Pre-trained Language Models
		3.1 Transformers and BERT
		3.2 Transfer Learning: Pre-training and Fine-Tuning
		3.3 BERT Models in the Health Domain
	4 NLP Tasks and Challenges in Healthcare
		4.1 Data Privacy
		4.2 Biomedical Data Sources and Their Challenges
		4.3 Tasks and Applications
	5 Bias and Fairness
		5.1 Bias in Clinical NLP
		5.2 Bias Measurement
		5.3 Bias Mitigation
	6 Explainability
		6.1 Methods for Explainability
		6.2 Evaluation of Explainability
		6.3 Explainability in Clinical NLP Tasks
	7 Summary and Recommendations
		7.1 Clinical Natural Language Processing
		7.2 Transfer Learning in Health
		7.3 Bias and Fairness
		7.4 Explainability
	References
Analytics
Statistical Analysis—Measurement Error
	Abstract
	1	Introduction
	2	Types of Measurement Error
	3	Consequences of Measurement Error
		3.1	Goal of the Analysis
		3.2	The Impact of Measurement Error in Explanatory Modelling
		3.3	The Impact of Measurement Error in Predictive Modelling
	4	Correction of Measurement Error
		4.1	Validation Studies
		4.2	Correction Methods
	References
Causal Inference and Non-randomized Experiments
	1 Causal Effects and Potential Outcomes
	2 Necessary Conditions for Causality
		2.1 Randomized Studies with Perfect Compliance
		2.2 Observational Studies
	3 Randomized Controlled Trials and Estimands of Treatment Effect
		3.1 Why Association Does Not Imply Causation
		3.2 Treatment Estimands in Trials
	4 Non-randomized Experiments of Time-Fixed Exposure and Confounders
		4.1 Analytical Methods to Estimate the Effect of Time-Fixed Exposures
	5 Non-randomized Experiments of Time-Dependent Exposure and Confounders
		5.1 Why Standard Methods May Fail
		5.2 Use of G-Methods to Overcome the Problem
	References
Statistical Analysis—Meta-Analysis/Reproducibility
	Abstract
	1	Introduction
	2	Data Sharing Arrangements in Multi-database Studies
		2.1	Centralized Data
		2.2	Distributed Data
	3	Regression Methods for Multi-database Studies
		3.1	Pooled Regression
		3.2	Meta-Analysis of Database-Specific Regression Coefficients
		3.3	Contemporary Distributed Regression Methods: Homogeneous Data
		3.4	Contemporary Distributed Regression Methods: Heterogeneous Data
	4	Discussion
	References
Machine Learning—Basic Unsupervised Methods (Cluster Analysis Methods, t-SNE)
	1 Introduction
	2 Related Work
		2.1 Explaining Training
		2.2 Explaining the Model
		2.3 Explaining the Inference
	3 IBIX Method
		3.1 Definitions
		3.2 Forward Mapping
		3.3 Backward Mapping
	4 Explainer Applications
		4.1 Explaining Autoencoders
		4.2 Explaining MRI-to-CT Generators
	5 Discussion
	6 Conclusion
	References
Machine Learning—Automated Machine Learning (AutoML) for Disease Prediction
	Abstract
	1	Introduction
		1.1	Cleaning Data
		1.2	Feature Selection
		1.3	Engineering New Features
		1.4	Choosing Classification and Regression Algorithms
		1.5	Assessing the Quality of a Model
		1.6	Explaining a Model
	2	Automated Machine Learning
		2.1	Auto-WEKA
		2.2	Auto-sklearn
		2.3	Tree-Based Pipeline Optimization Tool
	3	The Tree-Based Pipeline Optimization Tool (TPOT) Algorithm
		3.1	Representing TPOT Pipelines
		3.2	Initializing TPOT Pipelines
		3.3	Generating TPOT Pipeline Variation
		3.4	Evaluating TPOT Pipelines
		3.5	Selecting TPOT Pipelines
		3.6	Picking the Final TPOT Pipeline
	4	Scaling TPOT to Big Data
	5	Using TPOT to Automate Neural Networks
	6	Biomedical Applications of TPOT
	7	Future Directions
	References
Machine Learning—Evaluation (Cross-validation, Metrics, Importance Scores...)
	1 Background
	2 Train-Test Split
		2.1 Random Split
		2.2 Split with Stratification
		2.3 Cross-validation
		2.4 Bootstrap
	3 Evaluation Measures
		3.1 Evaluating the Supervised Models
		3.2 Evaluating the Unsupervised Models
	4 Conclusion
	References
Deep Learning—Prediction
	Abstract
	1	Introduction
	2	Deep Learning: A Theoretical Overview
	3	Deep Learning: Model Architectures
	4	Prediction Tasks for Deep Learning in Healthcare
	5	Applications: Medical Imaging
	6	Applications: Electronic Health Records Data
	7	Applications: Genomics
	8	Shortcomings and Challenges of Deep Learning in Healthcare
	References
Deep Learning—Autoencoders
	Abstract
	1	Introduction
	2	The Intuition Behind Auto-encoders
	3	Principal Component Analysis
	4	Methodology Behind Auto-encoders
	5	Auto-encoders for Denoising and Anomaly Detection
	6	Auto-encoders for Latent Vector and Feature Learning
	7	Variational Auto-encoders
	8	Disentanglement and Posterior Collapse
	9	Use Cases for VAEs and Latent Traversals
	10	Auto-encoders Versus Variational Auto-encoders (Summary)
	11	Designing an Auto-encoder and Common Pitfalls
	12	Examples Using the MNIST Dataset
	13	Demonstrator Use Case of an VAE for the Electrocardiogram: The FactorECG
	References
Artificial Intelligence
	Abstract
	1	The Anatomy and Physiology of the Generic Expert System
	2	Population Health Surveillance
	3	Summary
	References
Machine Learning in Practice—Clinical Decision Support, Risk Prediction, Diagnosis
	Abstract
	1	Introduction
	2	The Nature of Machine Learning
		2.1	Inputs and Outputs
		2.2	Criteria
	3	Reasoning with Medical Data
		3.1	Deduction
		3.2	Induction
		3.3	Inference
		3.4	Synthesis
	4	The Nature of Clinical Tasks
	5	Model Requirements
	6	The Optimal Model
	7	Clinical Applications
		7.1	Disease Stratification and Prognosis
		7.2	Interventional Inference and Prescription
		7.3	Clinical Pathways
	8	Ethical Aspects
	9	Quantifying Model Equity
	10	Conclusion
	References
Machine Learning in Practice—Evaluation of Clinical Value, Guidelines
	Abstract
	1	Introduction
	2	The Need for Frameworks in Ml-Based Clinical Research
	3	Evaluation and Monitoring in Clinical Ml Research
	4	The Issue of Interpretabiliy
	5	Reporting Statements, Checklists and Position Papers in Ml Research
		5.1	CONSORT-AI
		5.2	SPIRIT-AI
		5.3	STARD-AI
		5.4	TRIPOD-ML and PROBAST-AI
		5.5	DECIDE-AI
		5.6	PRIME-Checklist
		5.7	BIAS
		5.8	MI-Claim
	6	Guidelines
	7	Regulatory Aspects
		7.1	Data Quality and Algorithmic Bias
		7.2	Continuous Learning and Post-market Risk Assessment
		7.3	Transparency
	8	Conclusions
	References
Challenges of Machine Learning and AI (What Is Next?), Responsible and Ethical AI
	Abstract
	1	Trustworthy and Responsible AI
		1.1	Risks in Medical AI
			1.1.1 Patient Safety Issues Due to AI Errors
			1.1.2 Misuse and Abuse of Medical AI Tools
			1.1.3 Risk of Bias in Medical AI and Perpetuation of Inequities
			1.1.4 Lack of Transparency
			1.1.5 Privacy and Security Issues
			1.1.6 Gaps in AI Accountability
			1.1.7 Barriers to Implementation in Real-World Healthcare
		1.2	Approaches Towards Trustworthy and Responsible AI
			1.2.1 Guidelines for Developing Trustworthy Medical AI
			1.2.2 Evaluation of Medical AI Technologies
			1.2.3 Regulatory Aspects
		1.3	Summary and Discussion
	References