Atjaunināt sīkdatņu piekrišanu

E-grāmata: Data Driven Approaches for Healthcare: Machine learning for Identifying High Utilizers [Taylor & Francis e-book]

(University of Florida, Gainesville, USA), , (University of Kentucky, KY, USA), (University of Florida, FL. USA)
  • Formāts: 118 pages, 25 Halftones, black and white
  • Sērija : Chapman & Hall/CRC Big Data Series
  • Izdošanas datums: 07-Oct-2019
  • Izdevniecība: Chapman & Hall/CRC
  • ISBN-13: 9780429342769
Citas grāmatas par šo tēmu:
  • Taylor & Francis e-book
  • Cena: 231,23 €*
  • * this price gives unlimited concurrent access for unlimited time
  • Standarta cena: 330,33 €
  • Ietaupiet 30%
Data Driven Approaches for Healthcare: Machine learning for Identifying High UtilizersPalielināt
  • Formāts: 118 pages, 25 Halftones, black and white
  • Sērija : Chapman & Hall/CRC Big Data Series
  • Izdošanas datums: 07-Oct-2019
  • Izdevniecība: Chapman & Hall/CRC
  • ISBN-13: 9780429342769
Citas grāmatas par šo tēmu:
Taylor & Francis e-booksMore info about Taylor & Francis e-books
Rokasgrāmatas mājas lapa: https://www.taylorfrancis.com/books/9780429342769

Health care utilization routinely generates vast amounts of data from sources ranging from electronic medical records, insurance claims, vital signs, and patient-reported outcomes. Predicting health outcomes using data modeling approaches is an emerging field that can reveal important insights into disproportionate spending patterns. This book presents data driven methods, especially machine learning, for understanding and approaching the high utilizers problem, using the example of a large public insurance program. It describes important goals for data driven approaches from different aspects of the high utilizer problem, and identifies challenges uniquely posed by this problem.

Key Features:

  • Introduces basic elements of health care data, especially for administrative claims data, including disease code, procedure codes, and drug codes
  • Provides tailored supervised and unsupervised machine learning approaches for understanding and predicting the high utilizers
  • Presents descriptive data driven methods for the high utilizer population
  • Identifies a best-fitting linear and tree-based regression model to account for patients’ acute and chronic condition loads and demographic characteristics
    • Chapter 1 Introduction
    		1(6)
    1.1 Motivation
    		1(1)
    1.2 Goals of Data-Driven Approaches for High Utilizers
    		2(1)
    1.3 Challenges
    		3(1)
    1.4 Book Organization
    		4(3)
    Chapter 2 Overview of Health Care Data
    		7(8)
    2.1 Type of Health Care Data
    		7(2)
    2.2 Structure of Health Care Data
    		9(3)
    2.2.1 Structured Health Care Data
    		9(1)
    2.2.1.1 Diagnosis Codes
    		9(1)
    2.2.1.2 Procedure Codes
    		10(1)
    2.2.1.3 Pharmaceutical Codes
    		11(1)
    2.2.2 Unstructured Health Care Data
    		12(1)
    2.3 Common Data Sources for High Utilizers
    		12(3)
    2.3.1 Administrative Claims Data
    		13(1)
    2.3.2 PCORnet Common Data Model
    		13(2)
    Chapter 3 Machine Learning Modeling from Health Care Data
    		15(14)
    3.1 Supervised Models
    		15(4)
    3.1.1 Ordinary Least Squares Linear Regression (LR)
    		15(1)
    3.1.2 Regularized Regression (LASSO)
    		16(1)
    3.1.3 Gradient Boosting Machine (GBM)
    		16(1)
    3.1.4 Recurrent Neural Networks (RNN)
    		17(2)
    3.2 Interpreting Supervised Models
    		19(3)
    3.2.1 Global Interpretation: Understand Trained Model
    		19(1)
    3.2.2 Local Interpretation: Understand Each Prediction
    		20(1)
    3.2.3 Prediction Confidence
    		21(1)
    3.2.3.1 Voting and Consensus Rate
    		21(1)
    3.2.3.2 Providing Confidence Intervals
    		21(1)
    3.3 Unsupervised Models
    		22(3)
    3.3.1 Clinical Phenotyping
    		22(1)
    3.3.2 Behavioral Phenotyping: Clustering Inter-Arrival Time of Health Care Encounters
    		22(1)
    3.3.2.1 Histogram Representations of Asynchronous Time Series
    		23(1)
    3.3.2.2 Wasserstein Distance
    		23(2)
    3.3.2.3 Spectral Clustering
    		25(1)
    3.4 Discussion
    		25(4)
    Chapter 4 Descriptive Analysis of High Utilizers
    		29(18)
    4.1 Threshold-Based Methods for Frequent Emergency Department Users
    		29(10)
    4.1.1 Background
    		29(1)
    4.1.2 Methods
    		30(1)
    4.1.2.1 Approach
    		30(1)
    4.1.2.2 Study Population
    		30(1)
    4.1.2.3 Operational Definitions
    		30(1)
    4.1.2.4 Medical Expenditures
    		31(1)
    4.1.2.5 Enrollee Sociodemographics
    		31(1)
    4.1.2.6 Diagnostic History
    		32(1)
    4.1.2.7 New York University ED Profiling Algorithm
    		32(1)
    4.1.2.8 Frequent and Persistent Users
    		32(1)
    4.1.2.9 Annualized Visits
    		32(1)
    4.1.2.10 Statistical Analyses
    		32(1)
    4.1.3 Results
    		33(1)
    4.1.4 Characteristics of ED Users
    		33(3)
    4.1.4.1 Limitations
    		36(1)
    4.1.5 Discussion
    		37(1)
    4.1.5.1 Sociodemographics
    		37(1)
    4.1.5.2 Setting-Specific, High-Frequency Use
    		37(1)
    4.1.5.3 Cost Concentrations
    		37(1)
    4.1.5.4 Chronic, Comorbid Conditions, Mental Illness, and SUDs
    		38(1)
    4.1.5.5 Inappropriate and/or Avoidable Visits
    		38(1)
    4.1.5.6 Persistence
    		39(1)
    4.2 Temporal Consistency of High Utilizers
    		39(8)
    4.2.1 Background
    		39(1)
    4.2.2 Methods
    		39(1)
    4.2.2.1 Data
    		39(1)
    4.2.2.2 Experiment Setup
    		40(1)
    4.2.3 Results
    		40(1)
    4.2.3.1 Entire Adult Population
    		40(2)
    4.2.3.2 Temporal Correlation for the Top 10% Population
    		42(1)
    4.2.3.3 Chronic Conditions Cohorts
    		43(2)
    4.2.4 Discussion
    		45(2)
    Chapter 5 Residuals Analysis for Identifying High Utilizers
    		47(18)
    5.1 Background
    		47(1)
    5.2 Data and Methods
    		48(5)
    5.2.1 Study Population
    		48(1)
    5.2.2 Data Preprocessing
    		48(1)
    5.2.3 Model
    		48(1)
    5.2.3.1 Linear Regression
    		49(1)
    5.2.3.2 Tree-Based Model
    		49(1)
    5.2.4 Fitting the Model
    		50(1)
    5.2.4.1 Fitting Linear Regression
    		50(1)
    5.2.4.2 Fitting Tree-Based Model
    		51(1)
    5.2.5 Identifying the High Residuals Population
    		52(1)
    5.2.6 Breakdown Residuals
    		52(1)
    5.2.7 Stratified Model
    		53(1)
    5.3 Results
    		53(10)
    5.3.1 Compare Linear Regression and Tree-Based Model
    		53(1)
    5.3.2 Characterizing the High Utilizers
    		54(1)
    5.3.2.1 Demographics, Health Conditions, and Utilization
    		55(1)
    5.3.2.2 Temporal Consistency of Residuals
    		55(2)
    5.3.3 Breakdown Residuals to ICD-9-CM Codes
    		57(1)
    5.3.3.1 Essential Hypertension
    		58(1)
    5.3.3.2 Chronic Kidney Disease
    		59(1)
    5.3.4 Stratified Models by Service Settings
    		60(1)
    5.3.4.1 Residuals and Potentially Preventable Readmissions (PPR)
    		60(1)
    5.3.4.2 Residuals and Potentially Preventable Emergency Department Visits (PPV)
    		61(2)
    5.3.4.3 Residuals and Future Potentially Preventable Events
    		63(1)
    5.4 Discussion
    		63(2)
    Chapter 6 Machine Learning Results for High Utilizers
    		65(22)
    6.1 Predicting Hospital Readmissions
    		65(10)
    6.1.1 Background
    		65(1)
    6.1.2 Data and Methods
    		66(1)
    6.1.2.1 Dataset
    		66(1)
    6.1.2.2 Methods
    		67(1)
    6.1.2.3 Regularized Logistic Regression (LASSO)
    		68(1)
    6.1.2.4 Gradient Boosting Machine (GBM)
    		68(1)
    6.1.2.5 Deep Neural Networks (DNN)
    		68(1)
    6.1.3 Results
    		69(1)
    6.1.3.1 Prediction Accuracy
    		69(1)
    6.1.3.2 Interpret Models and Predictions
    		70(2)
    6.1.3.3 Prediction Confidence
    		72(2)
    6.1.4 Discussion
    		74(1)
    6.2 Predicting Health Care Expenditure
    		75(7)
    6.2.1 Background
    		75(1)
    6.2.2 Methods
    		75(1)
    6.2.2.1 Data
    		75(1)
    6.2.2.2 Objectives
    		75(1)
    6.2.2.3 Predictors
    		76(1)
    6.2.2.4 Predictive Models
    		76(1)
    6.2.2.5 Model Selection and Validation
    		76(1)
    6.2.3 Prediction Performance
    		77(1)
    6.2.3.1 Baseline
    		77(1)
    6.2.3.2 Choice of Period Length
    		77(1)
    6.2.3.3 Using Additional Information
    		78(1)
    6.2.3.4 Including Additional Prior Periods
    		79(1)
    6.2.4 Interpreting the Models
    		80(1)
    6.2.5 Choosing the Best Model
    		81(1)
    6.2.6 Discussion
    		82(1)
    6.3 Clustering Asynchronous Health Care Encounters Time Series
    		82(5)
    6.3.1 Emergency Department Visits Time Series
    		83(1)
    6.3.2 Inpatient Hospital Stays Time Series
    		83(1)
    6.3.3 Discussion
    		84(3)
    Chapter 7 Conclusions
    		87(2)
    Appendix A Acknowledgment 89(2)
    Bibliography 91(14)
    Index 105
    Chengliang Yang, Department of Computer Science, University of Florida Chris Delcher, Institute of Child Health Policy, University of Florida Elizabeth Shenkman, Institute of Child Health Policy, University of Florida Sanjay Ranka, Department of Computer Science, University of Florida.
    Permanent link: https://www.kriso.lv/db/9780429342769_pe.html
    Keywords: