predicting hospital readmissions from claims data deloitte ...predicting hospital readmissions from...
TRANSCRIPT
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
3 Predicting Hospital Readmissions from Claims Data Copyright © 2011 Deloitte Development LLC. All rights reserved.
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?
4 Predicting Hospital Readmissions from Claims Data Copyright © 2011 Deloitte Development LLC. All rights reserved.
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
5 Predicting Hospital Readmissions from Claims Data Copyright © 2011 Deloitte Development LLC. All rights reserved.
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
6 Predicting Hospital Readmissions from Claims Data Copyright © 2011 Deloitte Development LLC. All rights reserved.
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
7 Predicting Hospital Readmissions from Claims Data Copyright © 2011 Deloitte Development LLC. All rights reserved.
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%
8 Predicting Hospital Readmissions from Claims Data Copyright © 2011 Deloitte Development LLC. All rights reserved.
• 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
Copyright © 2011 Deloitte Development LLC. All rights reserved.9 Predicting Hospital Readmissions from Claims Data
• 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
10 Predicting Hospital Readmissions from Claims Data Copyright © 2011 Deloitte Development LLC. All rights reserved.
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%
11 Predicting Hospital Readmissions from Claims Data Copyright © 2011 Deloitte Development LLC. All rights reserved.
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
Copyright © 2011 Deloitte Development LLC. All rights reserved.12 Predicting Hospital Readmissions from Claims Data
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
13 Predicting Hospital Readmissions from Claims Data Copyright © 2011 Deloitte Development LLC. All rights reserved.
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
14 Predicting Hospital Readmissions from Claims Data Copyright © 2011 Deloitte Development LLC. All rights reserved.
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%
15 Predicting Hospital Readmissions from Claims Data Copyright © 2011 Deloitte Development LLC. All rights reserved.
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
16 Predicting Hospital Readmissions from Claims Data Copyright © 2011 Deloitte Development LLC. All rights reserved.
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
17 Predicting Hospital Readmissions from Claims Data Copyright © 2011 Deloitte Development LLC. All rights reserved.
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
19 Predicting Hospital Readmissions from Claims Data Copyright © 2011 Deloitte Development LLC. All rights reserved.
• 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