expanding the reach of predictive models: using clinical ... · pdf fileexpanding the reach of...

27
CONFIDENTIAL & PROPRIETARY Expanding the Reach of Predictive Models: Using Clinical, HRA, and Consumer Data Dan Dunn, PhD, Senior VP of R&D, Ingenix The National Predictive Modeling Summit December 13, 2007 Washington, DC

Upload: dothu

Post on 07-Mar-2018

218 views

Category:

Documents


2 download

TRANSCRIPT

Page 1: Expanding the Reach of Predictive Models: Using Clinical ... · PDF fileExpanding the Reach of Predictive Models: Using Clinical, HRA, and Consumer ... risk and cost with the addition

CONFIDENTIAL & PROPRIETARY

Expanding the Reach of Predictive Models: Using Clinical, HRA, and Consumer Data

Dan Dunn, PhD, Senior VP of R&D, Ingenix

The National Predictive Modeling Summit December 13, 2007 ●

Washington, DC

Page 2: Expanding the Reach of Predictive Models: Using Clinical ... · PDF fileExpanding the Reach of Predictive Models: Using Clinical, HRA, and Consumer ... risk and cost with the addition

© Ingenix, Inc. 2CONFIDENTIAL & PROPRIETARY

Agenda

Context for InnovationNew Sources of Data and Changing the Focus of Measurement – a Conceptual ModelUsing Alternative Data Sources in Risk Modeling

Page 3: Expanding the Reach of Predictive Models: Using Clinical ... · PDF fileExpanding the Reach of Predictive Models: Using Clinical, HRA, and Consumer ... risk and cost with the addition

© Ingenix, Inc. 3CONFIDENTIAL & PROPRIETARY

Context for Innovation

Information tools to support care and health management – current state:

Primary focus is on disease populations or individuals of moderate to higher riskClinical information and concepts supported by administrative medical and pharmacy claims, some use clinical dataOutputs include measures of risk, some add gaps in careMany tools add reporting and some cohort modeling capabilitiesLimited use of alternative sources of data

Page 4: Expanding the Reach of Predictive Models: Using Clinical ... · PDF fileExpanding the Reach of Predictive Models: Using Clinical, HRA, and Consumer ... risk and cost with the addition

© Ingenix, Inc. 4CONFIDENTIAL & PROPRIETARY

Context for Innovation

Increasing interest in focusing on healthier members in a population, or members of emerging risk

Extend interventions to the lower end of the risk spectrumImprove wellness, healthy behaviors and lifestyleImprove attitudes on healthIntervene “upstream” in a more pro-active way, e.g., pre-diabetes, and “pre-pre”-diabetes

Interest in creating a personal health record (PHR)Integrates information from a number of data sources to provide a multi-dimensional profile of an individual’s health

Support interventions in a more complete way – from “end-to-end”

Page 5: Expanding the Reach of Predictive Models: Using Clinical ... · PDF fileExpanding the Reach of Predictive Models: Using Clinical, HRA, and Consumer ... risk and cost with the addition

© Ingenix, Inc. 5CONFIDENTIAL & PROPRIETARY

Changing focus for information solutions

0

Increasing demand for information solutions that support interventions for relatively healthier members or those of emerging risk

Members without medical or pharmacy claims

Members of emerging clinical risk

-

pre-diabetic-

onset of chronic condition

Higher PM Risk- higher cost conditions- multiple co-morbidities

- recent acute events

Moderate PM Risk- chronic conditions

- some co-morbidities- recent history, stable

Lower PM Risk- smokers,

-

sleep problems,-

obese, inactive“Sweet spot” for current state of predictive modeling (PM) is patients of moderate to higher risk – supporting more traditional disease and care management

Page 6: Expanding the Reach of Predictive Models: Using Clinical ... · PDF fileExpanding the Reach of Predictive Models: Using Clinical, HRA, and Consumer ... risk and cost with the addition

© Ingenix, Inc. 6CONFIDENTIAL & PROPRIETARY

Identification and StratificationIdentification and StratificationIdentification and Stratification

Intervention and Management

Intervention and Management

Medical ClaimsRx Claims

DemographicsClinical Data

HRAsConsumer Data

Medical ClaimsRx Claims

DemographicsClinical Data

HRAsConsumer Data

Support “end-to-end” intervention solutions

SegmentationSegmentationSegmentation

ActivationActivationActivation

Risk PredictionClinical Profile

Health Behaviors

Match Patients to Programs

Support for Engagement and Intervention

Page 7: Expanding the Reach of Predictive Models: Using Clinical ... · PDF fileExpanding the Reach of Predictive Models: Using Clinical, HRA, and Consumer ... risk and cost with the addition

© Ingenix, Inc. 7CONFIDENTIAL & PROPRIETARY

Change of Focus and Requirements

Support analysis of healthier populations and emerging patientsLeverage existing and new sources of data, including HRA/self report and consumer informationIntegrate these different sources of data in innovative ways:

Improve on existing concepts, e.g., measures of future riskSupport new domains of measurement, including behaviors, attitudes, and social context

Accommodate different data scenarios – consistent data availability unlikely across and within populationsCreate a useful context for analysis

We are pulling together even a larger number of concepts and variablesAdd value by developing a context – organize information for analysis, presentation, and operations – in a flexible way

Page 8: Expanding the Reach of Predictive Models: Using Clinical ... · PDF fileExpanding the Reach of Predictive Models: Using Clinical, HRA, and Consumer ... risk and cost with the addition

© Ingenix, Inc. 8CONFIDENTIAL & PROPRIETARY

New Information and Domains – Opportunities

Address disease and lifestyle riskWhole-person approach to health management – across the full continuum of health and risk

Complement and expand opportunities to address further domains of health that they may not be concentrating onExpand models of clinical, risk and cost with the addition of new dimensions and sources of data

Prediction based on a set of new conceptsBring behavior and attitudes to the equationBring social and consumer variables to bear on risk

Tailor interventions based on a central repository of data that has key variables associated with outreach, intervention and outcomeSupport a Personal Health Record – informed by multiple sources of data, describing key dimensions of health

Page 9: Expanding the Reach of Predictive Models: Using Clinical ... · PDF fileExpanding the Reach of Predictive Models: Using Clinical, HRA, and Consumer ... risk and cost with the addition

© Ingenix, Inc. 9CONFIDENTIAL & PROPRIETARY

Using New Sources of Data and Changing the Focus in Measurement:

A Conceptual Design

Page 10: Expanding the Reach of Predictive Models: Using Clinical ... · PDF fileExpanding the Reach of Predictive Models: Using Clinical, HRA, and Consumer ... risk and cost with the addition

© Ingenix, Inc. 10CONFIDENTIAL & PROPRIETARY

What model of health can be used to structure a more complete approach? Wilson Cleary model (1995) of HRQOL is helpful because it represents a full picture of health

A model of health

Page 11: Expanding the Reach of Predictive Models: Using Clinical ... · PDF fileExpanding the Reach of Predictive Models: Using Clinical, HRA, and Consumer ... risk and cost with the addition

© Ingenix, Inc. 11CONFIDENTIAL & PROPRIETARY

Operational Model of Health: Concepts and Domains

A more complete approach requires methods and outputs to measure individuals along the different domains that describe healthDomains that support identification/stratification,segmentation, andactivationIntervention Groups – a context for integrating the five domainsNote – prediction and “risk”are only one component

Risk and Severity

Intervention Groups

Clinical

Health Behaviors

Social Context

Health Attitudes

Page 12: Expanding the Reach of Predictive Models: Using Clinical ... · PDF fileExpanding the Reach of Predictive Models: Using Clinical, HRA, and Consumer ... risk and cost with the addition

© Ingenix, Inc. 12CONFIDENTIAL & PROPRIETARY

Health Model Concepts and Domains

Information/domains to support identification and stratification: Clinical

A clinical description of an individual, based on diagnostic and

procedural concepts –

from claims, clinical results and self report–

Examples

diabetes, pre-diabetes, CHF, depression, sleep disorder, obesity, propensity for a clinical condition

Risk or Severity–

Predictive model risk, condition severity, self-report health status–

Examples

relative risk, condition episode severity, health statusBehavior (Healthy behaviors)

HRA and claims-based measures of behavior, behaviors inferred from consumer data

Examples

smoking, physical activity, compliance with chronic and preventive quality rules (gaps in care), prescription adherence

Page 13: Expanding the Reach of Predictive Models: Using Clinical ... · PDF fileExpanding the Reach of Predictive Models: Using Clinical, HRA, and Consumer ... risk and cost with the addition

© Ingenix, Inc. 13CONFIDENTIAL & PROPRIETARY

Health Model Concepts and Domains

Further segmentation and activation can be supported by: Attitudes about Health

Readiness to change, activation and perceived social supportSocial Context (Social Score)

Ascribed and achieved status, plus consumer-oriented variables–

Examples

Age, gender, race ethnicity, education, income, SES

Page 14: Expanding the Reach of Predictive Models: Using Clinical ... · PDF fileExpanding the Reach of Predictive Models: Using Clinical, HRA, and Consumer ... risk and cost with the addition

© Ingenix, Inc. 14CONFIDENTIAL & PROPRIETARY

What will Intervention Groups do?

Provide a context to organize and focus information – in a way that is consistent from both a clinical perspective and also from an operational perspectiveDescribe both clinical and wellness concepts – e.g., diabetes, smoking, sleep disorderHave defined levels – that map to potential cohorts for intervention – e.g., level of acuity; categories of smoking status; level of physical activityHave rules and algorithms that assign an individual to an Intervention Group – and further to a levelIncorporate methods to accommodate different data availability scenarios for each individual

Page 15: Expanding the Reach of Predictive Models: Using Clinical ... · PDF fileExpanding the Reach of Predictive Models: Using Clinical, HRA, and Consumer ... risk and cost with the addition

© Ingenix, Inc. 15CONFIDENTIAL & PROPRIETARY

Examples of Intervention Groups

Disease Management Wellness

Asthma/COPD Smoking/Tobacco

CAD Physical Activity

CHF Nutrition

Diabetes Safety

Back, Joint, Surgical Option ProblemsStress

Mental Health (Depression) Safety

Obesity Alcohol Abuse

Sleep Problems Sexual Risk Activity

Pain Syndromes

Page 16: Expanding the Reach of Predictive Models: Using Clinical ... · PDF fileExpanding the Reach of Predictive Models: Using Clinical, HRA, and Consumer ... risk and cost with the addition

© Ingenix, Inc. 16CONFIDENTIAL & PROPRIETARY

Diabetes Intervention Group Levels

Severe DiabetesModerate DiabetesMild DiabetesPre-Diabetes“Pre-Pre-Diabetes”No Diabetes

Information used to identify and stratifyMedical and Rx: diagnoses, drug therapiesPredictive model riskHRA and consumer: self-report, obesity, behaviors consistent with propensity for diabetesMap relevant clinical and family history to further define levels

Ask ourselves: If I run a diabetes management program, what would I want to understand about my members?

Severity of diabetes, propensityAssociated health behaviors, co-morbid conditions and attitudesWhat factors are associated with engaging members?

Page 17: Expanding the Reach of Predictive Models: Using Clinical ... · PDF fileExpanding the Reach of Predictive Models: Using Clinical, HRA, and Consumer ... risk and cost with the addition

© Ingenix, Inc. 17CONFIDENTIAL & PROPRIETARY

Sleep Problems

Intervention Group LevelsSevere sleep problemsModerate sleep problemsMild sleep problemsNo sleep problems

Information used to identify and stratifyMedical and pharmacy: diagnoses, drug therapies for treatment, diagnostic testsPredictive model riskHRA: self-report, sleep problem questions, medication self report

Ask ourselves: If I run a program for sleep problems what would I want to understand about my members?

Severity of sleep problemOther behaviors, conditions and attitudes associatedWhat factors are associated with engaging members?

Page 18: Expanding the Reach of Predictive Models: Using Clinical ... · PDF fileExpanding the Reach of Predictive Models: Using Clinical, HRA, and Consumer ... risk and cost with the addition

© Ingenix, Inc. 18CONFIDENTIAL & PROPRIETARY

Outputs

Summary and detail results for an individual along each of the domainsInformation, reports and views centered around the concept of an Intervention Group

with links between a patient, a group, their leveldetailed information supporting:

Intervention Group assignment–

appropriate segmentation–

activation for intervention–

the intervention itself

Risk scores and other summary measures

Page 19: Expanding the Reach of Predictive Models: Using Clinical ... · PDF fileExpanding the Reach of Predictive Models: Using Clinical, HRA, and Consumer ... risk and cost with the addition

© Ingenix, Inc. 19CONFIDENTIAL & PROPRIETARY

New Data Sources and Domains: Challenges

Consistency in the availability of information across individuals –

most will have claimssome will have HRAs and/or consumer dataclinical lab results may be availabletimeliness of the information

Opportunities for risk models – leveraging different types of informationCreating a flexible context for using this information – it translates in different ways depending on the appropriate focus for a patient and the domains

Page 20: Expanding the Reach of Predictive Models: Using Clinical ... · PDF fileExpanding the Reach of Predictive Models: Using Clinical, HRA, and Consumer ... risk and cost with the addition

© Ingenix, Inc. 20CONFIDENTIAL & PROPRIETARY

Using New Sources of Data in Risk Modeling

Page 21: Expanding the Reach of Predictive Models: Using Clinical ... · PDF fileExpanding the Reach of Predictive Models: Using Clinical, HRA, and Consumer ... risk and cost with the addition

© Ingenix, Inc. 21CONFIDENTIAL & PROPRIETARY

Measuring Health Risk – Overview

Markers of Risk

Translating Markers into Risk Measures

Data Inputs

Page 22: Expanding the Reach of Predictive Models: Using Clinical ... · PDF fileExpanding the Reach of Predictive Models: Using Clinical, HRA, and Consumer ... risk and cost with the addition

© Ingenix, Inc. 22CONFIDENTIAL & PROPRIETARY

New sources of data in risk modeling* (*in addition to administrative claims and enrollment)

HRA surveysWhat it adds

Clinical indicators –

e.g., self report of a condition not observed in claims

Overall assessment of health status–

Behaviors that indicate propensity for a higher risk clinical condition

Modeling approach–

New indications for disease risk markers–

Propensity-based markers of risk –

e.g., likelihood of diabetes–

Behaviors, other –

smoking, obesity–

Estimate risk weights for new markers –

use to adjust risk scoreChallenges

Data availability and timeliness–

Reconciling conflicting information

Page 23: Expanding the Reach of Predictive Models: Using Clinical ... · PDF fileExpanding the Reach of Predictive Models: Using Clinical, HRA, and Consumer ... risk and cost with the addition

© Ingenix, Inc. 23CONFIDENTIAL & PROPRIETARY

New sources of data in risk modeling* (*in addition to administrative claims and enrollment)

Clinical lab resultsWhat it adds

Condition severity –

e.g., organ function tests and cancer tumor/stage diagnostics

Trends in levelsModeling approach

Add lab-result based risk markers to a model–

Estimate risk weights for new markers –

use to adjust risk scoreChallenges

Data availability–

Timing–

Benefits a relatively small percentage of population –

although impact can be significant for these patients

Page 24: Expanding the Reach of Predictive Models: Using Clinical ... · PDF fileExpanding the Reach of Predictive Models: Using Clinical, HRA, and Consumer ... risk and cost with the addition

© Ingenix, Inc. 24CONFIDENTIAL & PROPRIETARY

Lab Results and PredictionAdded risk indicated by lab result markedly outside of normal range

Lab performed in last 90 days. Comparison of predicted (Impact Pro without Lab Model) and actual PMPM and relationship of prediction error with lab results ranges (“Difference”). Only most extreme lab result findings included on slide.

-2000

200400600800

1,0001,2001,4001,6001,800

Pred

ictio

n D

iffer

ence

($

PM

PM)

Albumin ALP CRP Chol Ratio CA-125 HbA1c

Using lab results in risk modeling

Page 25: Expanding the Reach of Predictive Models: Using Clinical ... · PDF fileExpanding the Reach of Predictive Models: Using Clinical, HRA, and Consumer ... risk and cost with the addition

© Ingenix, Inc. 25CONFIDENTIAL & PROPRIETARY

New sources of data in risk modeling* (*in addition to administrative claims and enrollment)

Consumer dataWhat it adds

Social Context –

income, education–

Consumer habits –

purchases, auto registration–

Categories –

groupings of individuals to Modeling approach

Categories and derived variables–

Test risk weights for new markers –

use to adjust risk score?Challenges

Data availability–

Timing–

TBD on general contribution to predictive accuracy on top of claims –

likely most helpful for lower risk

Page 26: Expanding the Reach of Predictive Models: Using Clinical ... · PDF fileExpanding the Reach of Predictive Models: Using Clinical, HRA, and Consumer ... risk and cost with the addition

© Ingenix, Inc. 26CONFIDENTIAL & PROPRIETARY

Summary

Information tools to support care and health management –current state:

Primary focus: disease populations, moderate to higher riskLimited use of alternative sources of dataMostly support ID & stratification

Use of alternative data sources both provides new opportunities and requires a new conceptual idea about “predictive modeling”

More complete view of the patientSupporting the full cycle of care and health management, including segmentation, activation and the intervention itselfFocus on healthier individuals and wellness programs is not bestsupported by a risk “score” – but by a multi-domain description of that individual

Challenges – consistent availability of data and creation of a context that supports operational realities

Page 27: Expanding the Reach of Predictive Models: Using Clinical ... · PDF fileExpanding the Reach of Predictive Models: Using Clinical, HRA, and Consumer ... risk and cost with the addition

© Ingenix, Inc. 27CONFIDENTIAL & PROPRIETARY

Questions/Comments