The Comparative Effectiveness Large Dataset Analysis Core: A Resource for Accelerating Research with Large, Public Datasets Janet Coffman, PhD Associate.

The Comparative Effectiveness Large Dataset Analysis Core: A Resource for Accelerating Research with Large, Public

Datasets

Janet Coffman, PhD

Associate Adjunct Professor

Philip R. Lee Institute for Health Policy Studies

University of California, San Francisco

October 5, 2012

Outline

• Examples of major types of large, public datasets

• Overview of Comparative Effectiveness Large Dataset Analysis Core (CELDAC)

• CELDAC datasets• CELDAC accomplishments • K Scholar success story• Discussion

2

Major Types of Large Datasets Used in

Health Services Research

3

Major Sources of Data

Type of Data Set Description Examples

Survey Collects information from individuals, families, or organizations

• Medical Expenditure Panel Survey• National Health and Nutrition

Examination Survey

Administrative claims

Information from records of health professionals and health care facilities, usually from billing records

• HCUP National Inpatient Sample• Medicare Research Identifiable

Files

Registries Information from datasets that incorporate all persons with a particular condition(s)

• California Cancer Registry• San Francisco Mammography

Registry

4

Major Units of Observation

5

Unit of Observation ExamplesIndividual • National Health and Nutrition Examination Survey

• National Survey of Children’s HealthHousehold • Medical Expenditure Panel Survey

• National Health Interview SurveyVisit or discharge • HCUP Kid’s Inpatient Databases

• National Ambulatory Medical Care SurveyPhysician • American Medical Association Masterfile

• HSC Health Tracking Physician SurveyFacility (e.g., hospital, clinic) •American Hospital Association Annual Survey

•California OSHPD Hospital Annual Financial DataGeographic area (e.g., county, state)

•US Census•Area Resource File

Major Types of Survey DesignsType of Survey

Description Examples

Cross-sectional

Data collected from a single sample at a single point in time

•National Health Interview Survey

•National Survey of Children’s Health

•California Health Interview Survey

Panel Data collected from a single sample at multiple points in time

•Medical Expenditure Panel Survey

•Health and Retirement Survey•National Longitudinal Study of

Adolescent Health

6

National Health and Nutrition Examination Survey

• Nationally representative sample of 5,000 persons per year

• Data collected in 15 counties per year• Two major components

– Interviews: demographic characteristics, socioeconomic status, diet, health behaviors

– Physical examinations: medical, dental, physiological, lab tests

http://www.cdc.gov/nchs/nhanes.htm7

Examples of UCSF Faculty Publications Using NHANES

• Seligman H.K. Food insecurity is associated with diabetes mellitus: results from the National Health Examination and Nutrition Examination Survey (NHANES) 1999-2002. Journal of General Internal Medicine. 2007 Jul;22(7):1018-23.

• Woodruff T, Zota A, Schwartz J. Environmental chemicals in pregnant women in the United States: NHANES 2003-2004. Environmental Health Perspectives. 2011 Jun;119(6):878-85. 2007 Jul;22(7):1018-23.

8

Medical Expenditure Panel Survey• Nationally representative sample of 22,000 to

37,000 persons• Overlapping panel design• 2 years of data collected through 5 rounds of

interviews• Three major components

• Household survey• Data on cost and utilization from providers caring for

household survey participants• Survey of employers regarding employer-sponsored

health insurance benefits

http://www.meps.ahrq.gov/mepsweb/9

Examples of UCSF Faculty Publications Using MEPS

• Newacheck P, Kim S. A national profile of health care utilization and expenditures for children with special health care need. Archives of Pediatric and Adolescent Medicine. 2005 Jan;159(1):10-7.

• Yelin E., et al. Medical care expenditures and earnings losses among persons with arthritis and other rheumatic conditions in 2003, and comparisons with 1997. Arthritis and Rheumatism. 2007 May;56(5):1397-407.

10

Overview of CELDAC

11

CELDAC Partners

CELDAC is a partnership at UCSF among the – Philip R Lee Institute for Health Policy Studies– Clinical and Translational Science Institute– Academic Research Systems

Funding– Administrative supplement to the NCRR grant for

UCSF’s Clinical & Translational Science Institute– California HealthCare Foundation

12

CELDAC Mission

The mission of CELDAC is to enhance UCSF's capacity for analysis of large local, state, and national health datasets to conduct comparative effectiveness research and other types of health services and health policy research.

13

CELDAC Goals• Accelerate access to and use of local, state, and national

health datasets, as a model for other CTSAs and health research organizations.

• Enhance UCSF researchers’ ability to compete for funding to use large data sets to conduct research.

• Develop procedures and infrastructure by conducting pilot studies.

• Support additional studies using large, public datasets.

• Provide consultation to researchers currently working with or interested in working with large datasets.

14

CELDAC’S Main Components, 2013

• Online, searchable inventory of datasets

• Consultation

• Repository of select datasets available through MyResearch

15

Find Large Datasetshttp://ctsi.ucsf.edu/research/celdac

A guided search tool to find the best datasets for a project. Builds on previous efforts by Nancy Adler, Andy Bindman, Claire Brindis, Charlie Irwin and others.

16

Search Results –Search for administrative data on infants’ use of health care services

http://ctsi.ucsf.edu/research/celdac

17

Dataset Description and Links

18

Provide Consultation

• Study design/conceptualization • Identification of relevant datasets• Assistance with dataset acquisition• Cohort selection• Data cleaning• Linking datasets• Strategies to deal with common methodological

issues in analysis of observational data• Programming support for preliminary analyses

19

Provide Consultation

• CELDAC provides some services on its own

• Links researchers with other CTSI Consultation Service units as needed– Data management– Biostatistics– Other

20

CELDAC Datasets

21

Analyze Large Datasets• CELDAC has created a repository of select large,

public datasets that are available to UCSF faculty at no cost.

• These data sets include– American Hospital Association Annual Survey– Area Resource File– HCUP Kids Inpatient Database – HCUP National Emergency Department Sample– HCUP National Inpatient Sample– HCUP State Emergency Department and Inpatient

Databases (select states)

22

National Inpatient Sample

• Largest publicly available all-payer inpatient database

• 20% stratified sample of admissions to community hospitals

• 8 million discharges• Data available from 1988 to 2010• Number of participating states has

increased over time from 8 to 45

23

Kid’s Inpatient Sample

• Only all-payer inpatient care database on children

• 3 million discharges of children and adolescents ≤ 20 years old

• Data available for 1997, 2000, 2003, 2006, and 2009

• Number of participating states has increased over time from 22 to 44

24

State Inpatient Databases

• Universe of inpatient discharge abstracts from community hospitals

• 46 states currently participate: > 90% of community hospital discharges

• Some states provide variables for tracking readmissions

• Data available from 1990 onward• UCSF has data from 2006 to 2010 for

states with readmissions variables25

National Emergency Department Sample

• 20% stratified sample of visits to community hospital EDs

• 25 to 30 million unweighted records• Data available from 2006 to 2009• 29 states currently participate• Includes ED visits that resulted in

– Treat-and-release– Transfer to another hospital– Admission to the same hospital

26

27

State Emergency Department Databases

• Universe of ED visits that did not result in a hospital admission from community hospitals in participating states

• 27 states currently participate• Some states provide variables for tracking

revisits• Data available from 1999 onward• UCSF has data from 2006 to 2010 for

states with revisits variables

CELDAC Accomplishments

28

CELDAC Clients

• CELDAC has assisted over 70 faculty, staff, and trainees at UCSF– 22 using datasets in CELDAC’s repository– 18 consultations– 11 linkages with other UCSF resources– 9 presentations to faculty, staff, and trainees

29

CELDAC Clients• CELDAC serves a wide range of departments

30

School of Medicine• Anesthesia• Dermatology• Emergency Medicine• Family & Community

Medicine• Medicine• Neurological Surgery• Neurology• Obstetrics & Gynecology• Orthopedic Surgery• Psychiatry• Pediatrics• Radiology• Surgery• Urology

School of Nursing• Community Health

Systems Nursing• Family Health Care

Nursing

School of Dentistry• Preventive and

Restorative Dentistry

School of Pharmacy• Clinical Pharmacy

CELDAC Extensions• UCSF Library

– Cross reference available datasets

• UCSF CTSI– Assist Community Engagement with

consultations that concern data analysis– Collaborate with Data Management

consultation unit on project to identify UCSF staff with data management expertise

31

CELDAC Extensions

• California HealthCare Foundation– Assessment of state policymakers’ needs

for health care data

32

K Scholar Success Story

33

Naomi Bardach, MD, MAS

• Assistant Professor of Pediatrics

• Former KL2 Scholar

• Current K23 Scholar (NICHD)

34

Initial Study

35

• Objective– To describe variation in hospital-level

pediatric asthma readmission rates in community hospitals and hospital and patient characteristics associated with readmissions

Initial Study

• Design/Methods– HCUP State Inpatient Databases for states with revisit

linkages (AZ, CA, FL, NC, UT)– Readmissions of patients age 2-21 years to non-

federal hospitals within 30-days of asthma related admission

– Outliers = hospitals with readmission rates that did not overlap with estimate of group mean

– Random effects logistic model to assess hospital and patient characteristics associated with readmissions

36

Initial Study

• Results– 1.9% of admissions were readmissions within

30 days (391 of 20,323)– Readmissions ranged from 0% to 7.3%– Only 2 hospitals had readmission higher rates

the group mean and none had lower rates– Patient age, race, payor, and immunological

complex chronic condition associated with odds of readmission

37

Subsequent Studies• Used HCUP State Inpatient and

Emergency Department databases to assess readmission and revisit rates across diseases and conditions

• Used the HCUP Kids Inpatient Database and a database from freestanding children’s hospitals to analyze pediatric ED visits and hospitalizations for mental health conditions

38

How CELDAC Helped• Initial K23 proposal not funded in part

due to concerns about data sources• Revised proposal to incorporate analysis

of HCUP state databases• Revised proposal funded• Platform presentation at Pediatric

Academic Societies Annual Meeting• Manuscript revised and resubmitted to Pediatrics

39

40

Questions for Discussion

• How could CELDAC better serve K Scholars?• What are the biggest barriers to

research with large, public datasets at UCSF?

• What services relating to analysis of large, public datasets would be most helpful?

Contact CELDAC

• Janet Coffman, PhD, Principal Investigator: [email protected]/415-476-2435

• Claire Will, PhD, Project Coordinator: [email protected]/415-476-6009

• http://ctsi.ucsf.edu/research/large-datasets

41