cer hub an informatics platform for conducting compartive effectiveness with emr hazlehurst

© 2012, KAISER PERMANENTE CENTER FOR HEALTH RESEARCH

The CER Hub: An Informatics Platform for Conducting Comparative Effectiveness Research with Comprehensive Electronic Medical Record Data

Brian Hazlehurst, PhD

Kaiser Permanente Northwest

Center for Health Research


Outline

Why do we need the CER Hub? The CER Hub extends and makes available an

automated medical record classifier (MediClass) The development of projects using the CER Hub The current CER Hub members and projects

under way


CER requires LOTS of data

Diverse populations, many topic areas Increasing adoption of EMR systems provides an

emerging opportunity for developing large databases KP covers ~9M lives @ 4 encounters/yr, roughly

100,000 encounters per day captured in the EMR A vast amount of this data is captured in unstructured (non-coded)

text


Example clinical encounter record segments addressing family and personal hx for cancer

Med Hx:Asthma-Azmacort, Ventolin, rarely prednisone Surg Hx:negFamily Hx:Fa-aodm, pgf colon ca, mgm bone marrow ca ------------------------------------------------------------------------Last Mammogram: 1 yr ago. Previous Paps have been normal There is a strong family hx of breast cancer.(M, MGM,Aunt) ------------------------------------------------------------------------RN noted S OB comma asked that I see pt. She has invasive ductal CA of the breast, and is getting chemo. Has today become more acutely SOB. ------------------------------------------------------------------------ROS: neg for exertional chest pain or pressure, shortness breath, changes inbowel habits.Fam Hx: + early MI, colon cancer-- sister in her 50s

None

------------------------------------------None

------------------------------------------174.9 CA FEMALE BREAST, INFILTRATING DUCTAL

------------------------------------------V16.0 FAMILY HX MALIGNANCY GI TRACT

Clinical note segment written Relevant ICD9 dx code applied

1

2

3

4


Coverage of the RAND QA measures by standardized CODED data

The remainder necessary for comprehensive quality assessment is found in either the templated- or free-text clinical notes of the EMR!

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Data Sources

% o

f Q

A m

easu

res Claims

Claims+Lab

Claims+Proced

Claims+Vitals

Claims+S/S

Claims+All


The CER HUB

A web-based platform for collaborative development of study-specific, standardized, processors of comprehensive electronic medical record data. Site data is extracted locally in industry standard form (HL7 CDA) Centrally developed processor of entire medical record creates a

standardized and reusable/extensible resource for CER Hub users Sensitive source clinical data (e.g., text progress notes) remain under local

control and is extracted on demand for specific projects Standardized (study-specific) datasets that are generated by applying the processor locally are pooled to answer targeted study questions centrally and these remain under study-team’s control


Why do we need the CER Hub?

EMR adoption promises LOTS of data, but the data are heterogeneous both across and within institutions EMR’s are variable (diverse representation of events) Clinical practices are variable (diverse priorities and capture of events) Patients are variable (diverse conditions and needs)

Need scalable informatics solutions allowing assignment of consistent (and specific) meanings to highly heterogeneous data want to remove spurious variation to highlight the “real” variation specific to

a study question


Outline



under way


MediClass (Medical Classifier)

Utilize a standard representation for electronic medical record data (HL7 Clinical Document Architecture, CDA-CCD) potential to process records of any EMR.

Process both text and coded data in the EMR potential to process any type of data captured in the EMR.

Allow for modular definition of measures or study variables (classifications determined by plug-in “knowledge modules”) potential to apply any specific measure.

Capable of local installation and operation potential to create shareable, standardized research data

MediClass System

EMR System(data

warehouse)

Event ClassifierClinical Event Classification

Rulebase

Results Repository

CDA w/Free Text Concepts& Structured

Data Concepts & Event Classification

CDA Medical Record(XML)

CDA Parser

CDA Medical Record( Java object model)

Coded Data Concept Mapper

Unified Medical

Language System (UMLS)

CDA w/Free Text Concepts

CDA w/Free Text Concepts& Structured

Data Concepts

Concept Identifier

ConceptIdentification

EMR Integration

Classification

JAMIA Sep-Oct, 2005

KnowledgeModule


Summary of MediClass study resultsProject Title Funder Description Key Results

Automated Assessment of Asthma Incidence and Prevalence (manuscript in preparation)

CDCUse of electronic medical records for surveillance of

asthma in an HMO

Sensitivity, 62-95%;Specificity, 90-100%

Automating Assessment of Asthma Care Quality (in press, AJMC)

AHRQ

Development of comprehensive automated assessment of outpatient

asthma care quality

Sensitivity(1), 62-92%;Specificity(1), 75-93%.Sensitivity(2), 35-69%Specificity(2), 69-95%

Vaccination Safety Datalink: Adverse Vaccine Event Detection (Hazlehurst et al, 2009 -- Vaccine)

CDC

Detection of possible vaccine-related adverse

events in large-linked databases

Sensitivity, 75-81%;Specificity, 97-98%;

Identifying Family and Personal History of Cancer in the EMR (Hazlehurst et al, 2005 HMORN research conf poster)

NCI

Identification of breast/ovarian and other

cancer family and personal history in

progress notes


HMO Interventions in Tobacco (Hazlehurst et al, 2005 – AJPM)

NCIAssessing compliance

with the 5A’s guideline in four HMO’s


(four HMO’s)


Outline



under way


The CER HUB

A web-based platform for collaborative development of study-specific, standardized, processors of electronic clinical datasets.

A web site with functions related to building, testing, sharing, study-specific processors of heterogeneous clinical data.


The CER HUB: Building

A set of tools for collaborative development of study-specific data processors. Operationalizing study variables in terms of concepts

identified in clinical records These variables may involve concepts identified in text and/or

structured data elements of clinical records. Eg., “persistent asthma” can be operationalized in terms of

sequences of asthma medication fills, exacerbation visits, and clinician assessment in the progress note.


The CER HUB: Testing

Developed processors can be run on the HUB against (de-identified) test datasets to evaluate the data processor.

Allows for rapid development of knowledge modules through iterative test-refine cycles.

Creates validation metrics that provide a “profile” about the data processor that is retained in the library.


The CER HUB: Sharing

A web site hosting virtual communities of researchers with shared interests (i.e., organized around a shared study topic).

A library of study-specific data processors are available for download as applications addressing a range of research questions.

Researchers who join the consortium build out the library over time through their activities using the HUB for their research.

The CER HUB workflow1) Developa study protocol

(define study measuresand populations of interest)

2) Develop and validatea standardized data processor

(operationalizestudy measures basedon concepts in data)

3) Configure the processor for your site

(defines site-specific parameters for the processor)

4) Apply processor to local data

5) Pool standardized,sharable data for analysis

1a) Data extraction

(extract data in a standard format)

De-identifiedsamples

Downloadprocessor

Pop def


1a) Data Extraction (EMR Integration)E

MR

Da

taW

are

ho

use

(RD

BM

S)

Study pop and data elements

CD

A

Da

tase

tX

ML

EMR-to-CDA mapping

EMR Adapter

(Runtime Engine)

Site

Sp

ecific

E

MR

Da

tase

t (C

SV

, X

ML

)

Site Specific schema

(CSV or XML)

Dataset Publish

HL7CDA

Schema

EMR Warehouse

Schema

MediClassApplication

Study Protocol

Site specific schema mapping

EMRAdapter(Schema Mapper)

CRD Schema

Clinical Research Data

Dictionary

De-identify and upload

to Hub

© 2012, KAISER PERMANENTE CENTER FOR HEALTH RESEARCHProcessor Validation Tools

Aggregate manual

coding using Gold

Standard Maker

Satisfactory Performance?

Define/refine

concepts and rules

MediclassProcesses

Job

GoldStd

Yes

Job

No

Processor building tools

Chart abstraction

using Manual

Coder tool

Validation Data

Done.Processor is

ready for download

Development data

From Step 1 and 1a

Study Protocol w/operationalized

measures

Uploaded de-identified data

· Direct Inspection of classification results

· Comparison to Manual Coding (development)

· Comparison to Gold Standard (validation)

2) Data Processor Development and Validation

EMRAdapter

Data WarehouseCDA

(XML)

MediClass Application

CDAw/ MediClass Classifications

(XML)

Post-Processor

EventsDataset(Flat file)

EventsDatasetProcessor

StudyMeasures

Data Extraction Event Identification Study Analyses

Study variables operationalized in terms of temporally located events

Application specific extraction filter for study specific (and sharable) events

Knowledge module and Configuration for specific application

Population and data element selection for encounter-based extraction

Study Protocol

Local Site Local Site DCC

CER HUB


Outline


automated medical record classifier (MediClass) The development of projects using the CER Hub The CER Hub project: members and studies

under way


The CER HUB Project

A consortium of researchers from 6 health systems KPNW, KPGA, KPHI, VA PugetSound, Baylor HealthCare System, OCHIN (consortium of FQHCs mostly on west coast)

Developing and using the CER HUB to address effectiveness questions in asthma control therapy and smoking cessation services


CER HUB Project Specific Aims1. Develop, operate, and evaluate a centralized CER service on the Internet

that provides automated tools, methods, and support for generating standardized datasets to answer CER questions.

2. Utilize the CER HUB to develop and validate an EMR-based measure of “asthma control" in accord with established national guidelines, and evaluate effectiveness of treatment intensification options on asthma control.

3. Utilize the CER HUB to assess implementation of the US Preventive Services Task Force evidence-based tobacco treatment guideline (the 5A’s) in the six participating organizations, and evaluate the comparative effectiveness of smoking cessation services on quitting in whole populations of patients in the course of real-world, routine clinical care.


Asthma Control

Risk Asthma-related steroid use –

orders or dispensings consistent with 2 or more courses in past 12 months;

ED visits or hospitalizations – 2 or more in past 12 months;

Progressive loss of pulmonary function over time (by spirometry or peak flow testing);

Medication side effects, such as dysphonia, thrush, osteoporosis (for inhaled corticosteroids), nervousness and tachydysrhythmia (for beta-2-agonists)

Impairment Asthma symptoms (wheezing, chest

cough, chest tightness, or shortness of breath) > two days per week;

Experiencing night-time awakening one or more per week;

Using reliever medications more than two days per week;

Symptoms interfere with normal activity; Reporting unacceptable control; Low asthma questionnaire score (e.g., ACT

score < 19); FEV1< 80% predicted and/or PEFR <80%

best;


Compare effectiveness of step-up therapies for asthma control

Main Asthma study CER question:

For patients on low-dose inhaled corticosteroid therapy whose asthma is not well controlled (i.e., failed EPR-3 Step 2 therapy), we will investigate the comparative effectiveness of the following step-up therapies (i.e., EPR-3 recommended Step 3 therapies)

(1) addition of a leukotriene modifier (2) addition of a long-acting beta-agonist (3) increase to medium-dose inhaled corticosteroids

On the basis of efficacy studies, options 2 and 3 are considered ‘first-line’ options.


CER Hub Study Protocol - Asthma Care

Possible Asthma Definition“Possible Asthmatic" is defined by patient having received at least one ICD-9-CM diagnosis code at any visit during study period of 493.xx

Study Inclusion Protocol We will include all patients, 12 years and older on 1/1/2006, identified as possible asthmatic (see definition above) during observation period (1/1/2006-12/31/2010) and also assess outcomes (2011-2012)

Persistent Asthma DefinitionWe will focus on patients whose asthma is “persistent” using the developed data processor that will consider medication usage (orders and dispenses), visits (inpatient, outpatient, and ED), and clinical judgement (clinician assessment that the patient has persistent asthma as documented in the progress note).


CER Hub Possible Asthma Population

Study Sites 2006 2007 2008 2009 2010

TOTAL DISTINCT

PATIENTS

Baylor 3166 6138 9836 6504 4850 30494

KPSE 9858 10224 10357 10917 5266 26756

KPHI 12637 12182 12227 12756 12324 33349

KPNW 21342 22495 23677 24741 24731 64764

OCHIN 1997 3870 6204 11260 15306 26922

VA-PS 1377 1668 1880 1972 2235 4667

TOTALDISTINCTPATIENTS 50377 56577 64181 68150 64712 186952


Asthma Investigators

Rich Mularski, MD (KPNW) Michael Schatz, MD (KPSC) Jerry Krishnan, MD, PhD (U of Chicago) David Au, MD (VAPS) Mark Millard, MD (Baylor) Bob Davis, MD (KPGA)


The 5 A’s of Smoking Cessation

5A step Operationaldefinition

Example in free-textsection of EMR

Ask Identify tobacco userstatus at every visit

“Patient smokes 1 ppd”

Advise Advise all tobaccousers to quit

“It is important for youto quit smoking now”

Assess Determine patient’swillingness tomake a quitattempt

“Patient not interestedin quitting smoking”

Assist Aid the patient inquitting

“Started patient onZyban”

Arrange Schedule follow-upcontact, in personor via telephone

“Follow-up in 2 weeksfor quit progress”


CER Hub Study Protocol-Smoking Cessation

Smoker Definition"Current smoker" is defined on an annual basis as having received at least one of the following during a calendar year:

1) ICD-9-CM diagnosis code indicating “tobacco abuse” at any visit2) An update of their social history to indicate "current smoker“

Additional measures are defined for “Quitter” (someone who recently quit) and “Former Smoker” (someone who has stayed quit).

Study Inclusion ProtocolUnique patients, 12 years and older, identified as:

1) current smoker and 2) having received primary care (one or more primary care visits)

All such patients will be included in the study and will be flagged as to their status according to these measures in each of the study years (2006 – 2010) and outcomes assessed (2011 – 2012).


CER Hub Smoker Population

Study Sites 2006 2007 2008 2009 2010

TOTAL DISTINCT

PATIENTS

Baylor 2707 12338 22741 22162 24025 58616

KPSE 17385 16067 15188 13722 6534 37868

KPHI 19160 17849 17164 18406 20896 51847

KPNW 47202 47786 46375 50944 50630 120328

OCHIN 8489 14726 23769 39946 56340 78736

VA-PS 10944 12334 13570 15052 15860 30535

TOTAL DISTINCT PATIENTS 105887 121100 138807 160232 174285 377930


Smoking Cessation Investigators

Victor Stevens, PhD (KPNW) Rebecca Williams, PhD (KPHI) Nancy Rigotti, MD (Harvard) Leif Solberg, MD (Health Partners) Andrew Williams, PhD (KPHI) Andrew Massica, MD (Baylor)


Informatics Investigators

Brian Hazlehurst, PhD (KPNW) Yan Xiao, PhD (Baylor) Jon Puro (OCHIN) Paul Nichol, MD (VAPS) MaryAnn McBurnie, PhD (KPNW)


www.cerhub.org

cer hub an informatics platform for conducting compartive effectiveness with emr hazlehurst

Documents

health research

standardized research

kaiser permanente center

heterogeneous data

data sources

type of data

site data

current cer hub members