predicting hospital readmissions from claims data deloitte ... · pdf filepredicting hospital...

Predicting Hospital Readmissions from Claims DataDeloitte Analytics

Nazmul KhanAditya SaneDavid Steier, Ph.D.

Business Intelligence & Analytics for Healthcare Conference12 July 2011, San Diego, CA

1 Predicting Hospital Readmissions from Claims Data Copyright © 2011 Deloitte Development LLC. All rights reserved.

Problem Statement

• Nearly 20% of in-patient admissions result in a readmission

• 90% of readmissions are preventable

• Unplanned readmissions cost approx $16,000 per instance• $42 billion / year

nationally

Solution Benefits

• Lower direct medical expense and care management costs

• Improved patients’ quality of care and satisfaction

• Improved quality metrics due to lower readmission rates

MotivationIn-patient Readmissions

Copyright © 2011 Deloitte Development LLC. All rights reserved.2 Predicting Hospital Readmissions from Claims Data

• Readmissions analytics: From Hindsight to Insight to Foresight

• Predictive Model and Results

• Experience with a Managed Care Application

Overview


Hindsight Insight Foresight

Analytical SolutionsUnderstanding In-patient Readmissions

Broad historical reporting on key performance indicators.

Macro analysis of process

What happened?

Statistical analyses (e.g. profiling and segmentation) help

organizations understand historical performance.

Macro analysis of populations

Why did it happen?

Advanced analysis, machine learning and modeling predict

future performance.

Micro analysis of individuals

What could happen?


Charts and ReportsHindsight

0%

5%

10%

15%

20%

25%

30%

35%

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27

Rea

dmis

sion

Rat

e

Week

Readmission Rate by Provider ID

Alpha

Bravo

Charlie

Delta


0%10%20%30%40%50%60%70%

Rea

dmis

sion

Rat

e Rx History

0%10%20%30%40%50%60%70%

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29

Rea

dmis

sion

Rat

e

Length of Stay

Length of Stay

0%

5%

10%

15%

20%

25%

30%

1 4 7 10131619222528313437404346495255586164

Rea

dmis

sion

Rat

e

Age

Age

Dashboard with FactorsInsight

0% 10% 20% 30% 40% 50%

Heart Failure & ShockPsychoses

Esophagitis and GastroentritisPTCA

Joint ReplacementChest Pain

Back & Neck ProcSpinal Fusion

Readmission Rate

DRG


Patient ID: X12345Age: 29 Sex: MalePrimary DX: 996.12(Mechanical Complication of vascular device / implant)

History: Anemia, Congestive heart failure, HypertensionRx History: G.I. Drugs, Beta blockers, Diuretics, Antihypertensives, Nitrates, Anticoagulants, HypnoticsService History: Excess Transport, Durable medical equipment (DME)

Readmission PredictionForesight


Readmission PredictionForesight

0%

20%

40%

60%

80%

100%

0 1 2 3 4 5 6 7 8 9 10 11 ≥12

Rea

dmis

sion

Rat

e

Number of Claims in Past 90 days

Transport Claims

0%

20%

40%

60%

80%

100%

0 1 2 3 4 5 6 ≥ 7

Rea

dmis

sion

Rat

e

Number of Claims in Past 90 days

CHF

Patient ID: X12345Age: 29 Sex: MalePrimary DX: 996.12(Mechanical Complication of vascular device / implant)

ReadmissionPropensity

84%

History: Anemia, CHF, HypertensionRx History: G.I. Drugs, Beta blockers, Diuretics, Antihypertensives, Nitrates, Anticoagulants, Hypnotics

180 day horizon

DRG 144 – Other Circulatory System Diagnosis with CC

Readmission Rate48%


• 237,129 commercial claims from 212,955 members• Set is a 5% de-identified national sample• Claims from 30% of members held back for cross validation

Dataset

• Claims with Transfer as discharge status• Claims with Mortality as discharge status• Claims associated with pregnancy and childbirth

Exclusions

Sourced from Thompson-Reuters MarketScan (Redbook)Data for Creating the Prediction Model

Timeline

Prediction HorizonClaimsHistory01 Jan 2006 01 Apr 2006 01 Jul 2006 31 Dec 2006


• Demographics• Admission Status• Type of Admission• DRG / Primary Dx• Secondary Dx• Discharge Status• Service / Revenue Codes

In-patient Data

• Demographics• Procedure Code• Diagnosis Code• Service / Revenue Codes

Out-patient Data

• NDC / Therapeutic Class• Quantity dispensed

Pharmacy Data

Data Sources and Model VariablesReadmission Model

Model Variables

• Age• Sex• DRG on present claim• Type of admission• Discharge status• Clinical history

• Diabetes, Hypertension, Depression, etc

• Prescription history• Nitrates, Beta blockers,

Lipid regulators, etc• Service history

• Transport, Physiotherapy, Laboratory test, etc

ExtractTransform

Load

+

Feature Derivation

Readmission model has 50+ variables


Age and GenderData Characteristics

0

500

1000

1500

2000

2500

3000

3500

4000

0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 56 58 60 62 64

Num

ber o

f Adm

issi

ons

Age

Male

Female

Number of claims = 237,129Number of members = 212,955Observed readmission rate = 19%


Time between admissionsData Characteristics

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

0 20 40 60 80 100 120 140 160 180

Cum

ulat

ive

Popu

latio

n

Days between readmission

Too early for intervention

80% of readmissions are after 15 daysThere is sufficient time for intervention and possible avoidance of a readmission


C5.0 Decision TreePrediction Model

• A Decision Tree is a series of closely linked questions that can be sequentially answered to arrive at a conclusion

• Decision trees can be automatically generated using statistical methods

• We use a consolidated result from twenty decision trees to improve accuracy


Important VariablesPrediction Model Characteristics

0.0134 0.0136 0.0138 0.014 0.0142 0.0144 0.0146 0.0148 0.015 0.0152

Rx Antipsychotics

Hx CHF

Hx Electrolytes

Rx Analgesics

Discharge Status

Rx Antibiotics

Hx Anemia

Hx Metastatic Cancer

Hx Solid Tumor

Excess Transport

Hx Psychoses

DRG Code

Average Information Gain

Rx – Excessive prescription history Hx – Clinical history


Precision and Sensitivity Prediction Model Performance

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%

Sens

itivi

ty

Precision

TestingTrainingRandom

Specificity = 99.44%

Precision = True Positives / (True Positives + False Positives)Sensitivity = True Positives / (True Positives + False Negatives)Specificity = True Negatives / (True Negatives + False Positives)

Typical Capacity BoundsSpecificity = 98.25%


Receiver Operating Characteristic (ROC)Prediction Model Performance

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%

True

Pos

itive

Rat

e(S

ensi

tivity

)

False Positive Rate(1 – Specificity)

TrainingTestingRandom

Training Area Under Curve = 0.8612Testing Area Under Curve = 0.8127


Client Experience

• Client used the model to streamline and optimize member selection for managed care

• Client increased the member pool size being actively managed to leverage early detection of complex chronic conditions

• Nurses use the model predictions in a ranked list to assess care management and coordination needs

• Members received telephonic intervention (health coaching, referrals, care coordination, etc)

• Length of care management is 3-4 months

Benefits

• Better visibility of factors that drive utilization and program participation

• Model provided a boost of 50x in selection rates – the selection rate went from 1:100 to 1:2

• Lowered direct medical cost $12,000 per member on average

• Lowered effort in identification of appropriate members for managed care

Member selection for managed careManaged Care Application


Providers

• Readmission can be predicted with present claim data

• Improved pay for performance can be achieved with increased quality metrics

• Timely intervention improves patient satisfaction

Payers

• Avoidable readmissions can be predicted with over 80% precision with claims history

• Improved accuracy in member selection for care management

• Reduced medical costs and care management costs

Summary

About DeloitteDeloitte refers to one or more of Deloitte Touche Tohmatsu Limited, a UK private company limited by guarantee, and its network of member firms, each of which is a legally separate and independent entity. Please see www.deloitte.com/about for a detailed description of the legal structure of Deloitte Touche Tohmatsu Limited and its member firms. Please see www.deloitte.com/us/about for a detailed description of the legal structure of Deloitte LLP and its subsidiaries. Certain services may not be available to attest clients under the rules and regulations of public accounting.

Copyright © 2011 Deloitte Development LLC. All rights reserved.Member of Deloitte Touche Tohmatsu Limited

www.deloitte.com/aboutfor

www.deloitte.com/us/aboutfor


• Published Literature– O Hasan, DO Meltzer, SA Shaykevich, CM Bell, PJ Kaboli, AD Auerbach, TB Wetterneck, VM

Arora, J Zhang and JL Schnipper. Hospital readmission in general medicine patients: a prediction model. Journal of General Internal Medicine. 2010. 25(3):211-219.

– C van Walraven, IA Dhalla, CM Bell, E Etchells, IG Stiell, K Zarnke, PC Austin and AJ Forster. Derivation and validation of an index to predict early death or unplanned readmission after discharge from hospital to the community. Canadian Medical Association Journal. 2010. 182(6):551-557.

– TL Whitlock, A Tignor, EM Webster, K Repas, D Conwell, PA Banks and BU Wu. A Scoring System to Predict Readmission of Patients With Acute Pancreatitis to the Hospital Within Thirty Days of Discharge. Clinical Gastroenterology and Hepatology. 2011. 9(2):175-180.

– PT Donnan, DWT Dorward, B Mutch, and AD Morris. Development and Validation of a Model for Predicting Emergency Admissions over the next Year (PEONY). Archives of Internal Medicine. 2008. 168(13):1416-1422

– S Howell, M Coory, J Martin, and S Duckett. Using routine inpatient data to identify patients at high risk of hospital readmission. BMC Health Services Research. 2009. 9(96).

– GM Hackbarth. Reforming America’s Healthcare Delivery System. Medicare Payment Advisory Commission Statement before US Senate Finance Committee. April 21, 2009.

• Acknowledgements – Jason Chiu, Carter (Todd) Shock, Kevin Hua, Stephen Bay

Related Efforts

predicting hospital readmissions from claims data deloitte ... · pdf filepredicting hospital...

Documents