Leveraging Clinical Data for Risk Adjusting Bundled Payments

April 13, 2015

Suma Thomas, MD, MBA, FACC Soyal Momin, MS, MBA

DISCLAIMER: The views and opinions expressed in this presentation are those of the author and do not necessarily represent official policy or position of HIMSS.

Conflict of Interest

Suma Thomas, MD, MBA, FACC Soyal Momin, MS, MBA Has no real or apparent conflicts of interest to report The study was sponsored by AstraZeneca Pharmaceuticals LP

© HIMSS 2015

Learning Objectives 1. Discuss the need for risk adjustment of bundled

payments using both payer claims and provider clinical data

2. Demonstrate how to integrate payer claims data with provider clinical data for the development of the Longitudinal Patient Record (LPR)

3. Demonstrate the application of the LPR for risk adjusting bundled payments

4. Debate the need for risk adjustment of bundled payments using both payer claims and provider clinical data

The Value of Health IT • Episode based treatment can lead to an

increase in care coordination and patient satisfaction

Satisfaction

• Bundled payments promote healthcare system transformation by reducing variations in care

Treatment/Clinical

• LPR efforts provide a better understanding of how patient variables affect outcomes

Electronic Information/Data

• Episode based treatment can promote patient engagement

Prevention/ Patient Education

• Use of integrated data for the PCI research study shows applications for payment innovation

Savings

http://www.himss.org/ValueSuite

Background

• Healthcare costs account for 17.2% of the GDP* • Traditional payment methods are not associated with

better patient outcomes** • Alternative payment models are being explored that

promote quality, efficiency, and effectiveness in health care delivery

* The Commonwealth Fund, Issues in International Health Policy, May 2012

** HealthAffairs, Health Policy Briefs, October 2012

Bundled Payments

• Bundled payments employ a single payment for all services related to a treatment or condition

• Goals: • Target a common procedure with variation in costs and outcomes • Cover essential clinical items and services • Provide a financial incentive to reduce inappropriate care • Yield a “win-win-win” outcome for payers, providers, and patients • Be simple enough to administer effectively • Maintain or improve quality of care • Appear seamless to policy holders • Be scalable and sustainable

HealthAffairs (Millwood). 2009;28(5):1418–28; 10.1377/hlthaff.28.5.1418

The Problem • Episodes of care should be risk adjusted* • Traditional risk adjustment approaches rely on healthcare claims

data** • Primary Objectives:

• Integrate payer claims data with provider clinical data for development of the LPR

• Demonstrate the application of the LPR for risk adjusting bundled payments

• Secondary Objectives: • Determine pre-procedural risk factors for high episode costs • Examine the procedures and services in the PCI episodes that

lead to high episode costs * http://aspe.hhs.gov/health/reports/09/mcperform/report.pdf

**http://www.hci3.org/sites/default/files/files/Severity%20Adjustment%20Fact%20Sheet.pdf

Data Environment

Database

Transactional Systems

• Medical evidence development is a major repurposing of administrative data for health services research

• Minor repurposing of administrative data for health plan operations

• Core claims processing and adjudication

What we see is not always what we get!

Limitations of Claims Data

Limitations of Claims Data

Deep Data Required to Answer Complex Questions

Stakeholders

• Partnership: • Cleveland Clinic Foundation (CCF) and HealthCore/Anthem

(HC/ANTM) • Executed Non-Disclosure Agreement (NDA), Business

Associate Agreement (BAA), and Data Transfer and Collaboration Agreement (DTCA) in January 2013

• Percutaneous Coronary Intervention (PCI) was selected for the research study

• All parties agreed on study objectives and protocols • Guiding Principles:

• Protect patient privacy • Protect business interests of data sources • Protect integrity of research

Data Sources

Data Sources: Following minimum Personal Health Information (PHI) exchange criteria, this study included:

• A CCF PCI patient list developed by HC/ANTM and CCF • HC/ANTM healthcare database administrative payer data

including claims, eligibility, lab results, and provider information

• CCF patient episode data • CCF PCI registry and Electronic Medical Record (EMR)

data The LPR was developed by HC/ANTM

Project Management - Phases

Define Plan Project Approval Launch Manage Close

Problem & High-Level Solution Definition

Project Approval

Requirements Definition

System Design

System Build, Testing & Deployment

SDLC Phases

Patient Reconciliation

Claims/Eligibility Extracts Provided for CCF Grouper Input

PCI Registry, EMR and HC/ANTM Core Files Provided to CDI Vendor

LPR Data Researchable

LPR Output

Prelim Results

Researchable PCI Registry Data

Grouper Results Returned from CCF

Project Management - Project Timeline

Patients Reconciliation

* Before Final Filtering for Age, Insurance Product, and Eligibility Periods

Year of PCI Event

Total Unique Patients

Matched CCF & HCI

Percent Matched

Yet To Be Matched Cleveland Clinic

(CCF)

Yet To Be Matched HealthCore

(HCI)

2006 238 0% 238

2007 204 0% 204

2008 205 7 3% 184 14

2009 286 118 41% 75 93

2010 273 111 41% 89 73

2011 284 112 39% 98 74

2012 283 78 28% 145 60

Totals 1773 426 24% 1033 314

Bundle Grouper

• CCF grouped claims into PCI treatment episodes capturing utilization 30 days pre and 180 days post PCI surgery

• HC/ANTM received grouper output and replaced imputed costs with actual allowed amounts for use in research study

Historical Data*

Bundle Definitions

Bundle Engine

Reports: •Summary •Exclusions

•Complications

Bundle Editor

Step 1: Identify and Select Bundle

Step 2: Identify Anchor Records

Step 3: Gather Additional Claim

Records to a Bundle

Step 4: Finalize Bundle

1. Identify the trigger claim - Use procedure and/or diagnosis code information

2. Determine services during the time window

- Time windows defined before and after trigger service - Definition considers the relevance of timing of a patient’s services

3. Identify unexpected outcomes

- Identifies inpatient admission for medical or surgical complication

Bundle Grouper Process

LPR Development Process

Data Sources & Integration Efforts

Pseudo Coding PCI Registry/PCI Clinical

Pseudo Coding PCI Registry/PCI Lesions

Pseudo Coding PCI Registry/PCI Medications

Pseudo Coding EMR/Medication Administered

Longitudinal Patient Record (LPR) Longitudinal Patient Record (LPR) Sample data and does not reflect

an actual individual

Longitudinal Patient Record (LPR) Sample data and does not reflect

an actual individual

Sample data and does not reflect an actual individual

Longitudinal Patient Record (LPR) Sample data and does not reflect

an actual individual

Longitudinal Patient Record (LPR)

Sample data and does not reflect an actual individual

Research Study Data Assets Utilized

• Variables Unique to the Registry

• Demographics: Race

• Clinical Characteristics: Elective vs. emergent PCI, BMI

• Recent Medication Use: Beta blockers, Long acting nitrates

• Angina and Heart Failure Classifications

• Stress/Imaging Study Results

CCF PCI ACC Registry + EMR Data

HealthCore Integrated Research

Environment (HIRE) (Anthem

Members)

Enriched Data for the Research Study

PCI Bundle Grouper

Data

Study Design and Population

• Retrospective observational study design • Patients receiving PCI procedures at Cleveland Clinic Main Campus • Patients ages 18 – 75 years old at time of PCI • Study period: 1/1/2006 through 12/31/2012

Up to 30-day prior

PCI performed at CCF Main Campus

PCI episode identified via PCI episode grouper

1/1/2006 PCI Registry data + HIRE data + Grouper output 12/31/2012

Up to 180-day post

Study Population Characteristics

Characteristic Number of PCI bundles, n 388 Emergent, % 47% Elective, % 53% Demographics Age, mean (SD) 55.9 (8.0) Male, % 81% Race, % Caucasian 89% Black, not of Hispanic origin 5% Other 6% BMI, mean (SD) 30.9 (6.0)

Characteristic Other risk factors, % Current smoker 26% Prior MI 30% Prior PCI 33% Prior CABG 20% Current on dialysis 1% Peripheral vascular disease 9% Diabetes 29% Clinical evaluations, % CAD presentation of STEMI 15% CAD presentation of NSTEMI 7% CAD presentation of unstable angina 24%

Anginal classification within 2 weeks-CCS IV 17%

NYHA Class within 2 weeks-Class IV 10% Cardiomyopathy or LV systolic dysfunction 4% Stress/imaging study - positive results 55%

Statistical Methodology

Modeling

• Variables associated with costs and deemed clinically relevant were considered

• A backward selection algorithm was used to choose the best fit model

• Cross validation was performed to avoid overfitting the data

Cost as a Continuous Outcome

• Generalized linear models with a gamma distribution were used to predict costs

Cost as a Dichotomous Outcome

• Modeled using logistic regression to predict a patient having costs in the top 25 percentile

Predictive Model – Continuous Outcome (n=388)

Cos

ts

Observation

Actual Value

Predicted Value

Variables included Effect direction

Main effectsAgeElective PCI (vs. Emergent PCI)Prior MIPrior PCICAD presentation of STEMICAD presentation of NSTEMIStress/imaging study - positive resultsInteraction TermsAge^2BMI^2Age * BMIAge^2 * BMIAge^2 * BMI^2Age^2 * CAD presentation of NSTEMIPrior MI * Stress/imaging study - positive resultsPrior PCI * CAD presentation of STEMI

Predictive Model – Dichotomous Outcome (n=388)

Model predicting “high costs” as the outcome (top quartile of costs)

Variables Included Effect direction

Independent VariablesAgeBMIElective PCI (vs. Emergent PCI)CAD presentation of STEMICAD presentation of unstable anginaStress/imaging study - positive resultsInteraction TermsAge * BMICAD presentation of unstable angina * Stress/imaging study - positive results

Results Stratified by Elective vs. Non-elective

Emergent PCI (n = 183)

Elective PCI (n = 205)

AUC = 0.71

AUC = 0.72

Variables included in fitted model Effect direction

Independent VariablesAgeBMIStress/imaging study - positive resultsInteraction TermsAge * BMIAge * BMI^2

BMI^3

Variables included Effect direction

Independent VariablesBMIFamily history of premature CADPrior MIBeta blocker use within past 2 weeksInteraction

BMI * Prior MI

Limitations

• Small sample size and limited power in the integrated data, especially with stratifying by PCI status and treating cost as a dichotomous outcome

• Possible incomplete capture of Medicare costs in individuals 65 years and older

• Potential limitations in application to non-commercial and non-Cleveland Clinic patients

• Limited auditing of registry data sources

Challenges - Legal/Compliance

• Non Disclosure Agreement (NDA) • Development of Data Transfer and Collaboration

Agreement (DTCA) • Agreement on specific legal clauses for protecting

patient privacy and business interests for data sources

• Business Associate Agreement (BAA) • Protection of non-CCF provider fee schedules through

the use of imputed costs • Institutional Review Board (IRB) approval

Challenges - Business

Challenges - Technical

Findings

• Clinical data and administrative healthcare data are limited by collection methods and clinical accuracy

• HC/ANTM’s in-depth data quality assessment was valuable when utilizing clinical data for research purposes

• Integrating multiple data sources is complex • Expectation management for timelines and budgets is

essential • Interoperability issues should be considered early in the

project

The Value of Health IT • Episode based treatment can lead to an

increase in care coordination and patient satisfaction

Satisfaction

• Bundled payments promote healthcare system transformation by reducing variations in care

Treatment/Clinical

• LPR efforts provide a better understanding of how patient variables affect outcomes

Electronic Information/Data

• Episode based treatment can promote patient engagement

Prevention/ Patient Education

• Use of integrated data for the PCI research study shows applications for payment innovation

Savings

http://www.himss.org/ValueSuite

Questions? Suma Thomas, MD, MBA, FACC Vice-Chairman, Operations and Strategy Cleveland Clinic , Cardiovascular Medicine Phone: (216) 445-0473 Soyal Momin, MS, MBA Director, Research Environment/Informatics HealthCore/Anthem Inc. Phone: (404) 384-4526