data mining techniques as healthcare · pdf file•goal –improve the quality and...

17
Company Confidential - For Internal Use Only Copyright © 2012, SAS Institute Inc. All rights reserved. DATA MINING TECHNIQUES AS HEALTHCARE ANALYTICS FOR LARGE DATA: PRACTITIONER’S GUIDE ISF 2014 24 JUNE 2014 Tae Yoon Lee Analytical Consultant [email protected]

Upload: lykiet

Post on 28-Mar-2018

219 views

Category:

Documents


4 download

TRANSCRIPT

Company Confidential - For Internal Use Only

Copyright © 2012, SAS Insti tute Inc. Al l r ights reserved.

DATA MINING TECHNIQUES AS

HEALTHCARE ANALYTICS FOR LARGE DATA:

PRACTITIONER’S GUIDE

ISF 2014 – 24 JUNE 2014

Tae Yoon Lee

Analytical Consultant

[email protected]

Company Confidential - For Internal Use Only

Copyright © 2012, SAS Insti tute Inc. Al l r ights reserved.

HEALTHCARE

ANALYTICS

DATA MINING

introduction

Healthcare predictive analytics today requires the processing of big data relating to hospital

and patients administrative data, clinical and non-clinical data including patient

demographics, disease diagnoses and procedures, patient charges, medical health

records, discharge status. This big data needs to be processed and analyzed to extract

knowledge for decision-making and cost-saving. Data mining techniques provide a set of

tools that can be applied to detect patterns, classifications, hospital transfers, and

mortality. In this session we demonstrate data mining techniques including Decision

Trees, Logistic Regression, Neural Networks, and Survival Data Mining using an example

with hospital data, to identify at which medical state a patient should be transferred to

tertiary hospital, and to predict the probability of mortality, and when the next event will

occur.

2

Company Confidential - For Internal Use Only

Copyright © 2012, SAS Insti tute Inc. Al l r ights reserved.

HEALTHCARE

ANALYTICS

DATA MINING

CONTENTS

This presentation is a Practitioner’s Guide on Data Mining Techniques as

Healthcare Analytics for Big Healthcare data. We explain:

• How can we implement Healthcare Analytics Data Mining Techniques?

• Example Applications

3

Company Confidential - For Internal Use Only

Copyright © 2012, SAS Insti tute Inc. Al l r ights reserved.

HEALTHCARE

ANALYTICS

DATA MINING

What is Healthcare Analytics Data Mining?

• What is Healthcare data? – Large data Management

• Huge amount of data generated by healthcare transactions (too complex and

voluminous to be processed and analysed by traditional methods)

• New methods needed – Data Mining provides the methodology and technology to

transform these mounds of data into useful information for decision making.

• Goal – improve the quality and cost of healthcare

• Healthcare insurers detect fraud and abuse

• Healthcare organizations make customer relationship management decisions

• Physicians identify effective treatments and best practices

• Patients receive better and more affordable healthcare services.

4

Company Confidential - For Internal Use Only

Copyright © 2012, SAS Insti tute Inc. Al l r ights reserved.

IMPACTS…

Lower CostsBetter Health

Outcomes

Improved Quality of Life

Through Prediction and Prevention

Company Confidential - For Internal Use Only

Copyright © 2012, SAS Insti tute Inc. Al l r ights reserved.

Step 1:

Data Checks, Segmentation Suites

Description:

Cleaned data for Modeling

Data checks, Feasibility Analysis

EM Input data, StatExplore,

Multiplot , Impute nodes

Output Clean data for modeling

Step 2:

Variable Reduction

Identify Patient risk drivers

Fine/Coarse Classing, Variable

Clustering, Correlation, VIF

Analysis

EM Interactive Grouping, Variable

Clustering, SAS code nodes

Candidate variables and Shortlist

variabales

Step 3:

Segmentation Variable Selection

Discrete & Continuous

Segmentation variables

Decision Tree, Segment Profiling

Analysis

EM Decision Tree, Clustering, Logistic

Regression, Segment Profiling

nodes

Optimum Segmentation split on segmentation

variables

Step 4:

Model Build

Build both a whole patient level and

segment level models

Fine/Coarse Classing,

Correlation, VIF, Logistic Regression

EM Interactive Grouping,

Scorecard, SAS code, Variable

Clustering, Segment Profiling nodes

Model Estimates, Parent Gini & System

Gini’s

Step 5:

Model Comparison

Compare the best System Gini with the

Parent Gini

Compare Gini’s, consider Benefits of

segmentation

EM SAS code, Model Comparison nodes

Identify if the Segmentation is

effective and required

Company Confidential - For Internal Use Only

Copyright © 2012, SAS Insti tute Inc. Al l r ights reserved.

COMPREHENSIVE

APPROACH

Company Confidential - For Internal Use Only

Copyright © 2012, SAS Insti tute Inc. Al l r ights reserved.

Discovery Modeling Deployment

Inp

ut

s

Automated

In-Database

Real Time

Embedded in Applications

Operationalize

2. Test/Stage

5. Retire

1. Development

3. Deploy

4. Track

ANALYTICS TOOLS

Company Confidential - For Internal Use Only

Copyright © 2012, SAS Insti tute Inc. Al l r ights reserved.

SAS®

ENTERPRISE

MINER™

6.2

MODEL DEVELOPMENT PROCESS

Sample Explore Modify Model Assess

Utility

Company Confidential - For Internal Use Only

Copyright © 2012, SAS Insti tute Inc. Al l r ights reserved.

HEALTHCARE

ANALYTICS

DATA MINING

Example using SAS Enterprise Minter

• Identify Transfer to Tertiary Hospital

• Predicting Probability of Mortality

10

Company Confidential - For Internal Use Only

Copyright © 2012, SAS Insti tute Inc. Al l r ights reserved.

Demo

11

Company Confidential - For Internal Use Only

Copyright © 2012, SAS Insti tute Inc. Al l r ights reserved.

Demo

12

Company Confidential - For Internal Use Only

Copyright © 2012, SAS Insti tute Inc. Al l r ights reserved.

Demo

13

Company Confidential - For Internal Use Only

Copyright © 2012, SAS Insti tute Inc. Al l r ights reserved.

Demo

14

Company Confidential - For Internal Use Only

Copyright © 2012, SAS Insti tute Inc. Al l r ights reserved.

Demo

15

Company Confidential - For Internal Use Only

Copyright © 2012, SAS Insti tute Inc. Al l r ights reserved.

16

Model Data Target FALSE TRUE FALSE TRUE

Model Description Role Target Negative Negative Positive Positive

Scorecard2 Scorecard TRAIN transfer 310 9568 41 79

Scorecard2 Scorecard VALIDATE transfer 263 7167 41 29

Scorecard Scorecard TRAIN transfer 338 9579 30 51

Scorecard Scorecard VALIDATE transfer 273 7176 32 19

Tree Decision TRAIN transfer 354 9609 . 35

Tree Decision VALIDATE transfer 256 7204 4 36

Reg Regression TRAIN transfer 357 9607 2 32

Reg Regression VALIDATE transfer 265 7198 10 27

Neural Neural TRAIN transfer 389 9609 0 0

Neural Neural VALIDATE transfer 292 7208 0 0

Model Event Classification Table

Company Confidential - For Internal Use Only

Copyright © 2012, SAS Insti tute Inc. Al l r ights reserved.

THANK YOU!

ISF 2014 – 24 JUNE 2014

Tae Yoon Lee

Analytical Consultant

[email protected]