Using Health Related Data Sets for Research

1
Using Health Related Data Sets for Research

• Working Group Report

Jon Mark Hirshon, MD, MPH Committee Chair
2
Presentation Outline

• Introduction
• Emergency Medicine and Public Health
• Definitions
• What are Health Related Data Sets (HRDSs)?
• Examples of HRDSs
• HRDSs Research Potential
• NHAMCS by Dr. Margaret Warner
• Future Directions and Challenges for Research Use
  of HRDSs

3
Working Group

• Daniel Andersen, PhD
• Jon Mark Hirshon, MD, MPH
• Charlene Irvin, MD
• Linda McCaig, MPH
• Richard Niska, MD, MPH
• Gordon S. Smith, MBChB, MPH
• Margaret Warner, PhD

4
Why Emergency Medicine and Public Health?

• Institute of Medicine has stated
• Strengthening the relationship between public
  health and medicine is critical to address the
  health needs of the public
• Emergency departments (EDs) play a vital role in
  our health care system.
• In part due to key location at the interface
  between the populace and healthcare

5
EDs Data Collection

• Emergency Departments are well positioned to
  collect data concerning major public health,
  medical and social problems
• Improved technology and governmental programs
  have dramatically increased information
  availability
• Can be viewed as a window on the health and
  social needs of the community because EDs are a
  key healthcare access entry point

6
Research Based on ED Data

• Emergency physicians, researchers and other
  health care professionals must understand and use
  the data collected from EDs and related systems
  to
• Better understand patterns of health resource
  utilization
• Recognize worsening public health problems
• Help in the development of
• More focused research activities
• More thoughtful and efficacious interventions

7
What are Health Related Data Sets (HRDSs)?

• No standard definition found within the
  literature
• Working Definition
• Any data that is created as result of patient
  interactions with the health care system or
  relates to interactions with the health care
  system
• Includes routine physician visits, hospital
  admissions, prescription purchases, but also
  vital statistics and law enforcement records

8
Health Services Data Sets

• Specific Subset of HRDSs are Health Services Data
  Sets
• Large amount of literature related to Health
  Services Data Sets
• Does not include all HRDSs
• E.g. law enforcement records, longitudinal
  surveys
• Defined as
• Extant data that can serve as a proxy measure of
  the health of populations

9
Broad Definition of HRDSs

• Any systematic collection of any information
  related to health care, including
• related costs
• services (pharmacy, physician, ambulance or
  transport, hospital, clinic, school, prison,
  other institution)
• health related surveys or studies
• information from health insurance providers.

10
Importance of HRDSs

• Opportunity to conduct epidemiological studies on
  population health
• Evaluate the determinants and distribution of
  diseases
• Potential to study issues related to health
  disparities and health care for the underserved
• Conduct public health surveillance
• Potential to link data from multiple sources

11
Limitations of HRDSs

• Frequently based upon existing data
• Usually appropriate for retrospective studies
• Can be costly to create and maintain
• Many HRDSs already exist and are readily
  available, though may cost a nominal fee
• Important to maintain patient confidentiality
• Only a limited data set may be publically
  available

12
Barriers to Use of HRDSs

• Concerns related to
• Completeness, accuracy, and timeliness of data
• Generalizability
• May require specific analytical skills to use
• IRB approval
• Generally not an issue since most data sets use
  de-identified data
• Each HRDS may have unique limitations or barriers
  to use

13
Types of Health Related Data Sets

• Complete Enumeration
• Population-based Sample Surveys
• Non-population Based Registries
• Longitudinal Surveys
• Linked Data

14
Complete Enumeration
15
Population Based Sample Surveys
16
Non-Population Based Registries
17
Longitudinal Surveys
18
Linked Data
19
What are the research potential for HRDSs?

• Explore a specific example
• National Hospital Ambulatory Medical Care Survey
  (aka NHAMCS)
• Margaret Warner, PhD
• National Center for Health Statistics
• Centers for Disease Control and Prevention

20
Overview of the NHAMCS

• Margaret Warner PhD
• National Center for Health Statistics
• Consensus Conference
• May 13, 2009

21
Overview
NHAMCS is a national probability sample survey of
visits to emergency departments (EDs) of
non-Federal, short-stay, and general hospitals in
the United States.

• How are the NHAMCS data used?
• NHAMCS Survey Methodology
• Data user considerations
• Accessing the data

22
How are NHAMCS data used?

• Changes in utilization and practice
• diagnoses, tests/procedures, prescribing
• Quality of care
• Impact of performance measures and educational
  campaigns
• Healthy People 2010 objectives
• Health disparities
• Adoption/Diffusion of new technologies

23
(No Transcript)
24
Data users

• Medical associations
• Government agencies
• Institute of Medicine
• Health services researchers
• University and medical schools
• Broadcast and print media

25
(No Transcript)
26
The IOM report
27
Trends in emergency department visits, number of
hospitals, and number of emergency departments in
the United States, 1994-2004

• Kellermann A. N Engl J Med 20063551300-1303

28
Percentage of ED visits at which an opioid was
prescribed by pain severity and race
1997-2000
2003-2005
NOTE Pain severity was not collected in
2001-2002. SOURCE Wilper AP et al. Health
Affairs. 2008Jan-Febw84-w93.
29
Age adjusted injury visit rates using alternative
definitions of injury for state benchmarking
1,500.0
1,400.0
1,300.0
Visits per 10,000 population
1,200.0
1,100.0
1,000.0
900.0
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
SOURCES CDC/NCHS, NHAMCS-ED, 1995-2005 data
files and NEISS-AIP data from WISQARS.
30
NHAMCS Methodology
31
NHAMCS Sample design

• 112 geographic PSUs
• 500 hospitals
• 400 EDs and 250 OPDs
• 37,000 ED and 35,000 OPD visits
• 4-week reporting period

32
NHAMCS Scope

• General medicine, surgery, pediatrics, ob/gyn,
  substance abuse, and other clinics are in-scope
• Ancillary services are out of scope

33
Data collection

• Gaining cooperation
• Advance letters
• Endorsement letters
• Public relations materials
• Data collection procedures
• Induction visit by Census field representative
  (FR)
• FR training of office/hospital staff

34
2009 NHAMCS Patient Record Form
35
Items available on the public use file

• Patient characteristics
• age, race, sex, ethnicity
• Visit characteristics
• reason for visit, diagnosis, medication
• Provider characteristics
• physician specialty, hospital ownership
• Contextual variables based on patient zip code
• of poverty, median HH income, adults with
  bachelors degree or higher, urban/rural

36
Data processing

• Data are coded and keyed by SRA International
• Quality control procedures
• Edit checks by NCHS

37
Coding systems used

• A Reason for Visit Classification (NCHS)
• ICD-9-CM
• diagnoses
• external causes of injury
• procedures
• Drug coding system (NCHS)

38
Data User Considerations
39
Encounter vs. person data

• NHAMCS are record-based surveys
• Estimates are in terms of visits and not persons
• Not population-based surveys (NHIS)
• Cannot calculate incidence or prevalence rates
  from NHAMCS estimates

40
Sample weight

• Sample data MUST be weighted to produce national
  estimates
• Estimation process
• Adjusts for survey and item nonresponse
• Makes several ratio adjustments within and across
  physician specialties and hospitals

41
Sampling error

• NHAMCS are not simple random samples
• Clustering effects
• Providers within PSUs
• Visits within physician practice or hospital
• Must use generalized variance curve or special
  software (e.g., SUDAAN) to calculate SEs for all
  estimates, percents, and rates

42
Reliability criteria

• Estimate considered reliable if
• Estimate is based on 30 raw cases or more and
• Relative standard errors (RSEs) of estimate is
  less than 30 percent
• Combine multiple years of data to increase
  reliability

43
Nonsampling error

• Frame coverage
• Reporting and processing errors
• Biases due to survey and item nonresponse
• Incomplete responses

44
Minimizing nonsampling error

• Improve sample frame for better coverage
• Encourage uniform reporting and eliminate
  ambiguities
• Pretest survey items and procedures
• Perform quality control procedures consistency
  and edit checks
• Train Census field representatives

45
NHAMCS Response rates
ED
46
Attempts to improveresponse rate

• Publicity
• Eliminating questions that have a high item
  non-response
• Methodological studies
• PR material

47
HIPAA

• No directly identifiable information collected
• PHS Act 308(d) / Title 15
• Data Use Agreement w/ Limited Dataset
• IRB approval w/ waiver of patient authorization
• Accounting Document

48
Accessing the data
49
Microdata files

• Downloadable files
• NHAMCS, 1992-2006
• CD-ROMs
• NHAMCS, 1992-2005
• Tapes/cartridges (NTIS)
• NHAMCS, 1992-1997

50
Tools to access public-use files

• SAS input statements, variable labels, value
  labels, and format assignments for 1993-2006
• SPSS syntax files
• STATA .do and .dct files for 2002-2006

51
Other tools to access data
52
Accessing non-public use data

• Research Data Center
• Access to information not available on public use
  files
• Patient zip code linked income, education, or
  urbanicity status
• Provider physician gender and age, board
  certification, teaching hospital, medical school
  affiliation, ED size
• Supplement data CCSS
• Geographic state and county FIPS codes

53
Data Center rules

• Submit a proposal
• Cannot use data to identify patients or providers
  or geographic location of providers
• Cannot remove data files
• Fee onsite / remote / file construction
• 2 NCHS RDCs Hyattsville,MD Altanta, GA
• 9 Census Bureau RDCs are located in Boston, MA
  Berkeley, CA Los Angeles, CA Washington, DC
  Chicago, IL Ann Arbor, MI New York, NY Ithaca,
  NY and Durham, NC.

54
I need more information on NHAMCS data

• Call Ambulatory Hospital Care Statistics Branch
  at [phone]
• Public Use Documentation
• or

55
http//www.cdc.gov/nchs/about/major/ahcd/ahcd1.htm
56
Thank You

• Margaret Warner
• mwarner_at_cdc.gov

57
Future Directions and Challenges

• Future- Potential for significant growth with
  electronic medical records
• Increased data sets
• More timely data
• Increased access
• Challenge- need to increase usefulness,
  especially for evaluating prevention and
  screening programs

58
Research Recommendation 1

• For applicable datasets, electronically link ED
  health care visits longitudinally to future
  health outcomes, including costs and other
  financial implications, while maintaining
  de-identification of the data.

59
Research Recommendation 2

• For data collected electronically, provide timely
  access to the data. For data not collected
  electronically, provide access to the data no
  later than 1.5 years after data collection

60
Research Recommendation 3

• Improve completeness of data collection for
  clinically relevant and/or historical data
  elements, such as external causes of injury
  codes.

61
Research Recommendation 4

• Provide easy access to data that can be parsed
  into smaller jurisdictions (such as states) for
  policy and/or research purposes, while
  maintaining confidentiality.