Secondary Data Analysis:

1
Secondary Data Analysis

• Opportunities and Pitfalls

2
Who am I and why am I here?

• Laurel A. Copeland, BS MPH PhD
• VA Health Services Research Development field
  program investigator UTHSCSA Department of
  Psychiatry Assistant Professor - Research
• copelandl_at_uthscsa.edu
• http//czresearch.com/dropbox

3
What Can Secondary Data Analysis Do for You?

• Provide preliminary data for grant proposals
• Facilitate publication while you apply for grants
• Explore new areas

4
My CV

• Shope, Jean T Copeland, Laurel A Maharg, Ruth
  Dielman, TE Butchart, Amy T (1993). Health
  Education Quarterly 20(3) 373-390.
• Shope, Jean T Copeland, Laurel A Dielman, TE
  (1994). Alcoholism Clinical and Experimental
  Research 18(3) 726-733.
• Shope, Jean T Copeland, Laurel A Marcoux, Beth
  C Kamp, Mary E (1996). Journal of Drug
  Education, 26(4) 323-337.
• Shope, Jean T Copeland, Laurel A Maharg, Ruth
  Dielman, TE (1996). Alcoholism Clinical
  Experimental Research, 20(5) 791-798.
• Copeland, Laurel A Shope, Jean T Waller,
  Patricia F (1996). Journal of School Health,
  66(7) 254-260.
• Barry, Kristen Lawton Fleming, Michael F
  Manwell, Linda Copeland, Laurel A (1997).
  Journal of Family Practice, 45(2) 151-158.
• Zimmerman, Marc A Copeland, Laurel A Shope,
  Jean T Dielman, TE (1997). Journal of Youth and
  Adolescence, 26(2) 117-141.
• Barry, Kristen Lawton Fleming, Michael F
  Manwell, Linda Copeland, Laurel A Appel, Scott
  (1998). Family Medicine, 30(5)366-371.
• Blow, Frederic C Barry, Kristen Lawton
  BootsMiller, Bonnie J Copeland, Laurel A
  McCormick, Richard Visnic, Stephanie (1998).
  Journal of Psychiatric Research, 32 311-319.
• Valenstein, Marcia Barry, Kristen Lawton Blow,
  Frederic C Copeland, Laurel A Ullman, Esther
  (1998). Psychiatric Services, 49(8) 1043-1048.
• Shope, Jean T Copeland, Laurel A Kamp, Mary E
  Lang, Sylvia W (1998). Journal of Drug Education,
  28(3) 185-197.
• Kales, Helen C Blow, Frederic C Copeland,
  Laurel A Bingham, Raymond C Kammerer, Ericka E
  Mellow, Alan F (1999). American Journal of
  Psychiatry, 156(4) 550-556.
• Blow, Frederic C Barry, Kristen Lawton
  Copeland, Laurel A McCormick, Richard Lehmann,
  Laurent Ullman, Esther (1999). Psychiatric
  Services, 50(3) 390-394.
• Maio, Ronald F Shope, Jean T Blow, Frederic C
  Copeland, Laurel A Gregor, MaryAnn Brockmann,
  Laurie M Weber, Janet E Metrou, Mary E (2000).
  Annals of Emergency Medicine, 35(3)252-257.
• Kales, Helen C Blow, Frederic C Bingham,
  Raymond C Copeland, Laurel A Mellow, Alan M
  (2000). Psychiatric Services, 51(6) 795-800.
• Roberts, J Scott Blow, Frederic C Copeland,
  Laurel A Barry, Kristen Lawton Van Stone,
  William (2000). Journal of Geriatric Psychiatry
  and Neurology, 13 Summer 78-86.
• Blow, Frederic C Walton, Maureen A Barry,
  Kristen Lawton Coyne, James C Mudd, Sharon A
  Copeland, Laurel A (2000). Journal of the
  American Geriatrics Society, 48(7) 769-774.
• Kales, Helen C Blow, Frederic C Bingham,
  Raymond C Roberts, J Scott Copeland, Laurel A
  Mellow, Alan M (2000). American Journal of
  Geriatric Psychiatry, 8301-309.
• Blow, Frederic C Ullman, Esther Barry, Kristen
  Lawton Bingham, C Raymond Copeland, Laurel A
  McCormick, Richard Van Stone, William (2000).
  American Journal of Orthopsychiatry, 70(3)
  389-400.

5
My CV without 2 Data

• Shope, Jean T Copeland, Laurel A Maharg, Ruth
  Dielman, TE Butchart, Amy T (1993). Health
  Education Quarterly 20(3) 373-390.
• Shope, Jean T Copeland, Laurel A Dielman, TE
  (1994). Alcoholism Clinical and Experimental
  Research 18(3) 726-733.
• Shope, Jean T Copeland, Laurel A Marcoux, Beth
  C Kamp, Mary E (1996). Journal of Drug
  Education, 26(4) 323-337.
• Shope, Jean T Copeland, Laurel A Maharg, Ruth
  Dielman, TE (1996). Alcoholism Clinical
  Experimental Research, 20(5) 791-798.
• Copeland, Laurel A Shope, Jean T Waller,
  Patricia F (1996). Journal of School Health,
  66(7) 254-260.
• Barry, Kristen Lawton Fleming, Michael F
  Manwell, Linda Copeland, Laurel A (1997).
  Journal of Family Practice, 45(2) 151-158.
• Zimmerman, Marc A Copeland, Laurel A Shope,
  Jean T Dielman, TE (1997). Journal of Youth and
  Adolescence, 26(2) 117-141.
• Barry, Kristen Lawton Fleming, Michael F
  Manwell, Linda Copeland, Laurel A Appel, Scott
  (1998). Family Medicine, 30(5)366-371.
• Blow, Frederic C Barry, Kristen Lawton
  BootsMiller, Bonnie J Copeland, Laurel A
  McCormick, Richard Visnic, Stephanie (1998).
  Journal of Psychiatric Research, 32 311-319.
• Valenstein, Marcia Barry, Kristen Lawton Blow,
  Frederic C Copeland, Laurel A Ullman, Esther
  (1998). Psychiatric Services, 49(8) 1043-1048.
• Shope, Jean T Copeland, Laurel A Kamp, Mary E
  Lang, Sylvia W (1998). Journal of Drug Education,
  28(3) 185-197.
• Kales, Helen C Blow, Frederic C Copeland,
  Laurel A Bingham, Raymond C Kammerer, Ericka E
  Mellow, Alan F (1999). American Journal of
  Psychiatry, 156(4) 550-556.
• Blow, Frederic C Barry, Kristen Lawton
  Copeland, Laurel A McCormick, Richard Lehmann,
  Laurent Ullman, Esther (1999). Psychiatric
  Services, 50(3) 390-394.
• Maio, Ronald F Shope, Jean T Blow, Frederic C
  Copeland, Laurel A Gregor, MaryAnn Brockmann,
  Laurie M Weber, Janet E Metrou, Mary E (2000).
  Annals of Emergency Medicine, 35(3)252-257.
• Kales, Helen C Blow, Frederic C Bingham,
  Raymond C Copeland, Laurel A Mellow, Alan M
  (2000). Psychiatric Services, 51(6) 795-800.
• Roberts, J Scott Blow, Frederic C Copeland,
  Laurel A Barry, Kristen Lawton Van Stone,
  William (2000). Journal of Geriatric Psychiatry
  and Neurology, 13 Summer 78-86.
• Blow, Frederic C Walton, Maureen A Barry,
  Kristen Lawton Coyne, James C Mudd, Sharon A
  Copeland, Laurel A (2000). Journal of the
  American Geriatrics Society, 48(7) 769-774.
• Kales, Helen C Blow, Frederic C Bingham,
  Raymond C Roberts, J Scott Copeland, Laurel A
  Mellow, Alan M (2000). American Journal of
  Geriatric Psychiatry, 8301-309.
• Blow, Frederic C Ullman, Esther Barry, Kristen
  Lawton Bingham, C Raymond Copeland, Laurel A
  McCormick, Richard Van Stone, William (2000).
  American Journal of Orthopsychiatry, 70(3)
  389-400.

6
Is it scientifically valid to use data for
purposes for which it was not originally
collected?

• Yes, because.
• No, because
• Maybeplease use caution

7
Today and next week

• Uses of secondary data
• Common biases encountered with secondary data
• Methods of adjustment
• Sources of secondary data
• Possible effects of HIPAA on this type of
  research

8
Different Uses

• Health Care Delivery
• Quality assessment
• Geographic variation under-/over-utilization
• Adverse events
• Outcomes of a particular treatment
• Cost
• Natural history
• incidence, prevalence
• prognosis
• Association not causation
• Other
• Regulatory

9
Some Examples
To show you the range of uses To show you the
work is publishable
10
Regulatory enforcement

• TENET HEALTHCARE - 2002
• Medicare Report Jan-Oct 2004 included these
  findings
• At Alvarado Hospital, prosecutors accused Tenet
  of running covert kickback arrangements
• Document requests were received from federal
  prosecutors in L.A.
• False claims were uncovered and the company
  agreed to pay 22.5M to settle allegations
• Redding Medical Center was determined to be
  performing unnecessary cardiac surgery
• Tenet sold some holdings to Hospital Partners
  of America to satisfy part of the 2003 settlement
  with federal officials
• Whistleblowers were to receive over 8M

11
(No Transcript)
12
Whose data set is it anyway? Sharing raw data
from randomized trials

• Andrew J Vickers
• Trials. 2006 May 16715
• http//www.trialsjournal.com/content/pdf/1745-6215
  -7-15.pdf

13
Risk of death in elderly users of conventional
vs. atypical antipsychotic medications
Wang PS, Schneeweiss S, Avorn J, Fischer MA,
Mogun H, Solomon DH, Brookhart MA. N Engl J Med.
2005 Dec 1353(22)2335-41 http//content.nejm.or
g/cgi/reprint/353/22/2335.pdf
14
Does Medicare coverage of colonoscopy reduce
racial/ethnic disparities in cancer screening
among the elderly?

• Shih YC, Zhao L, Elting LS (2006). Health Aff
  (Millwood). Jul-Aug 25(4)1153-62.
  http//content.healthaffairs.org/cgi/content/full/
  25/4/1153

15
Cancer incidence in Kentucky, Pennsylvania, and
West Virginia disparities in Appalachia

• Lengerich EJ, Tucker TC, Powell RK, Colsher P,
  Lehman E, Ward AJ, Siedlecki JC, Wyatt SW (2005).
  J Rural Health 21(1)39-47.
• http//www.blackwell-synergy.com/toc/jrh/21/1

16
Can you think of examples of secondary data
analysis that you have read in the medical
literature?

• Any that have changed your practice or your
  research direction?

17
Methods of Learning

• Lecture
• In class epidemiologic exercise
• Journal article evaluation
• Proposal for your own secondary data analysis

18
Exercise 1 Policy Decision on FundingOf
Regional Trauma Center (handout)
19
Secondary Data Exercise 1

• A Regional Trauma Center (which encourages the
  surrounding hospitals to refer patients with
  serious injuries to it for expert care) is
  seeking additional funds from next years health
  budget for more equipment and staff. A local
  politician (who would rather spend such money on
  a new hospital named after his father) criticizes
  this request for additional resources by claming
  that Regional Trauma Centers do not, in fact,
  save lives, and he submits the following data to
  back up his claim.
• You are asked for your opinion on these data. How
  should you respond?

Severity of trauma surrounding hospitals surrounding hospitals regional trauma center regional trauma center
of cases of deaths of cases of deaths
Total 7266 458 3354 212
Case-fatality rate 458/7266 6.3 212/3354 6.3
20
Secondary Data Exercise 1

• A Regional Trauma Center (which encourages the
  surrounding hospitals to refer patients with
  serious injuries to it for expert care) is
  seeking additional funds from next years health
  budget for more equipment and staff. A local
  politician (who would rather spend such money on
  a new hospital named after his father) criticizes
  this request for additional resources by claming
  that Regional Trauma Centers do not, in fact,
  save lives, and he submits the following data to
  back up his claim.
• You are asked for your opinion on these data. How
  should you respond?

Severity of trauma surrounding hospitals surrounding hospitals regional trauma center regional trauma center
of cases of deaths of cases of deaths
Mild 3734 37 687 3
Moderate 1887 94 1238 37
Severe 1645 327 1429 172
Total 7266 458 3354 212
Case-fatality rate 458/7266 6.3 212/3354 6.3
21
Bias

• Systematic error in measurement or a systematic
  difference (other than the one of interest)
  between groups
• Selection
• For cohorts, assembly, migration, contamination,
  and referral bias
• Measurement
• Confounding

22
Bias Anticipate and Control

• Restriction (may lose generalizability)
• Matching
• Stratification
• Standardization
• Multivariate Adjustment
• Assuming the worst (sensitivity analyses)
• Discussing possible effects on your results

23
Example Hospital Mortality Report Cards

• Originally unadjusted
• Hospitals without trauma centers, doing primarily
  elective surgery, etc., looked really good
• Made hospitals who took care of the sickest of
  the sick look bad

24
Quality Assessment

• Data Quality Garbage in, garbage out
• Risk Adjustment To remove the confounding effect
  of different institutions providing care to
  patients with dissimilar severity of illness and
  case complexity

25
Interpret data carefully
26
Data Quality

• If no reliable (accurate and adequate) data are
  available, questions about risk adjustment are
  moot
• Inconsistent practices in assigning standard ICD9
  codes names to diseases existbut this has
  improved over time (Do you know who assigns the
  codes?) http//www.eicd.com/
• Lack of specificity of ICD9 system in some
  diagnostic areas, especially with regard to
  severity

27
Administrative Data Limitations

• Need patient-specific identifiers to link
  episodes
• Need multi-year data when outcomes are infrequent
• Limited generalizability if restricted by type of
  institution or hospital, by type of payor, or by
  location/region
• May lack important clinical variables known to be
  related to outcomes (especially clinical tests or
  qualitative evaluations of severity)

28
Risk Adjustment

• Controls for those patient characteristics that
  are related to the outcomes of interest
• Removes the confounding effect of different
  institutions providing care to patients with
  dissimilar severity of illness and case
  complexity
• Addresses regional variations
• Inadequate case-mix adjustment can lead to
  misclassification of outlier status

29
Risk Adjustment

• Primary data collection vs. administrative data
• Disease-specific vs. generic
• Commercial vs. developed for your study
• Predictors vary by outcomes being predicted

30
Essential Elements of Risk Adjustment

• Outcome-specific
• Contains specification of the principal diagnosis
• Contains demographics as proxies for preexisting
  physiological reserve
• Measure count of comorbidities and all the most
  important comorbidities to assume their own
  empirically derived coefficients

31
Classification of Disease States

• ICD-9 too many specific codes (n10,000)
• Clinical Classifications for Health Policy
  Research (CCHPR) good for chronic illness and
  longitudinal care http//www.ahrq.gov/data/hcup/h
  is.htm
• Primary diagnosis good for studies that focus on
  a single episode of care

32
Risk Adjustment Charlson

• Advantages
• Commonly used case-mix classification system in
  the health care industry
• System with which most clinicians and reviewers
  are familiar

33
Risk Adjustment Charlson

• Disadvantages
• Principal diagnosis not differentiated
• Original work did not specify ICD-9 codes that
  went into the disease categories
• Developed on inpatients predicting mortality may
  not be well suited to outpatients at low risk of
  death
• Not good for longitudinal care / chronic illness

34
Demographic Factors to Consider in Risk Adjustment

• Age (e.g., age-adjusted Charlson)
• Proxies of Social Support
• Marital status
• Race
• Gender
• SES (occupation, employment status, education)
• Proxies of Socioeconomic Status
• Health insurance status
• Home address zip code average income

35
Race and Gender

• Dont adjust for automatically
• Ideally adjust for variation in the patients
  physiological reserve and disease burden but not
  for variation in care rendered to patients

36
Propensity Scores

• Useful when dataset is small, to conserve power
• Need a good proxy to develop a propensity score
• Ask propensity for what? (Tx)
• Published schema may include predictors you want
  to study separately
• Best for non-randomized studies of treatment
  effect where you want to adjust for the factors
  that may have influenced the treatment choices

37
Your Assignment

• Read other two articles
• E-mail me, by this Thursday, a 1-page proposal
  for a database analysis related to your area of
  interest
• copelandl_at_uthscsa.edu