slide 1

New Measures to Access the Quality of Race/Ethnicity Reporting in State Databases

David Zingmond, MD, PhD
AHRQ monthly grantees call
January 10, 2012

UCLA logo

slide 2

Aim

  • Develop validated audit measures for race/ethnicity reporting that can be used for any state's statewide databases

slide 3

Background: Data Auditing

  • Current data auditing rules are blunt instruments for determining the accuracy (or adequacy) of race/ethnicity reporting in statewide hospital databases
    • Provide minimum criteria for flagging hospital data
      • Flags for hospitals exceeding rates for missing or unknown race/ethnicity
      • Flags for extreme variation in reporting (100% or 0% for categories)

slide 4

Unknown race/ethnicity

  • Combined race/ethnicity measure
  • 3.4% unknown race/ethnicity (mean across 349 hospitals reporting discharges in 2009)
    • 17 hospitals > 10% unknown
    • 1 hospital > 20% unknown

slide 5

Candidate Audit Measures

  • Reflect self-reported race/ethnicity
  • Data are readily available for use by those performing the data audits
  • Patient-level record comparisons are best
    • not available for every patient or for every state (e.g. where patient-level data linkages are not routinely performed)

slide 6

Available Data for Comparison

  • Patient-level data
    • Birth certificates (mother self-report; not all hospitals have births)
    • Cancer registry (abstracted with data supplemented by name-based algorithm and death certificates; small subset with self-report)
    • Death certificates (institutional reporting)
  • U.S. Census data (self-report)

slide 7

Patient Level Data
% agreement = # agree with GS / # Total *100

  • This type of measure can be used for single category agreement or overall agreement across all categories
  • If GS (gold standard) is truly self-report, then this measure can also be used for validation purposes

slide 8

Measures for Agreement - Summary Level Data

  • Hospital Summary Level Data
    Estimated disagreement =
    Abs(Reported — Predicted)/2 + % Unknown
    Reported = Distribution of race categories
    Predicted = Population mean predicted distribution using zip-code level distribution for each patient

*agreement = 100 - disagreement

slide 9

PDD vs CCR

  PDD vs. SR PDD vs. CCR
Race 0.91 0.92
Ethnicity 0.91 0.95
Alt Ethnicity (unk = NH) 0.95 0.95
White - White 0.94 0.93
Non - White - Non - White Race 0.82 0.91
Hispanic - Hispanic 0.66 0.66
  PDD vs. SR (combined race/ethnicity) PDD vs. CCR (combined race/ethnicity)
Combined Race/Ethnicity 0.90 0.92
NH White - NH White 0.95 0.96
Non - White - Non - White 0.79 0.83
Hispanic - Hispanic 0.66 0.66

N = 16,653

slide 10

PDD versus Birth

Race 70.0
Ethnicity 85.8
White 71.6
NW 65.4
Hisp 89.1
NH 82.1

Race/Eth 85.8
NHW 91.1
minorities 83.9

N = 513,456

slide 11

Scatter plot showing agreement between patient discharge data and birth versus patient discharge data and Census.

slide 12

Other observations

  • Too much scatter for good matching for prediction
  • Populations are not exact matches (mothers versus all adult patients)

slide 13

Further work on Data Audits

  • Revised metrics
  • Match populations for derivation of metrics
  • Compare proposed metrics to current insensitive, context-free metrics
  • Apply metrics to hospitals across time to see if there have been changes in performance during the observation period

slide 14

Complex flow chart showing data collection and reporting process. First row of process from left to right shows boxes connected by arrows for Patient, Clerk, Registration System, Medical Record, H I T, Record Abstract, O S H P D. Second row shows flow from Medical Record to C C R Registrar to Record Abstract to C C R. Third row shows flow from Medical Record to Decedent Affairs to Record Abstract to O. Vital Statistics to C C R. Fourth row shows flow from Medical Record to Live Births to O. Vital Statistics.

slide 15

Complex flow chart showing self-report process. First row of process from left to right shows boxes connected by arrows for Patient, Clerk, Registration System, Medical Record, H I T, Record Abstract, O S H P D. Second row shows flow from Medical Record to C C R Registrar to Record Abstract to C C R to Clinical Trials Database.


Internet Citation: New Measures to Access the Quality of Race/Ethnicity Reporting in State Databases. Healthcare Cost and Utilization Project (HCUP). July 2014. Agency for Healthcare Research and Quality, Rockville, MD. www.hcup-us.ahrq.gov/datainnovations/raceethnicitytoolkit/ca20.jsp.
Are you having problems viewing or printing pages on this Website?
If you have comments, suggestions, and/or questions, please contact hcup@ahrq.gov.
Privacy Notice, Viewers & Players
Last modified 7/31/14