Evaluation of Alternative Matching Criteria by Race/Ethnicity and Sex in NHIS-NDI Linked Mortality files

1
Evaluation of Alternative Matching Criteria by
Race/Ethnicity and Sexin NHIS-NDI Linked
Mortality files

• Gloria Wheatcroft and Jennifer Parker
• Office of Analysis and Epidemiology

Centers for Disease Control and
Prevention National Center for Health Statistics
2
Background

• Individual demographic, behavioral, and health
  characteristics are strongly associated with
  mortality
• These factors may also be associated with the
  presence and accuracy of survey information used
  for matching to the NDI
• Inferences from studies that use survey data
  linked to mortality data to examine the
  correlates of mortality could be affected by
  differential matching

3
Prior matching evaluation

• Liao et al. (AJPH 1998) evaluated three criteria
  for matching using the1986-1990 NHIS linked to
  NDI through 1991
• Exact social security number (SSN) match
• (A) or SSN unknown and matching score for
  demographic items gt recommended cutoffs
• (A) or SSN unknown and either 8 or more
  demographic items match or fewer than 8 match but
  with high matching score, gt32.5

4
Liao et al (AJPH1998) Death Rates Men, 45-64
years
A Exact match on social security B (A) or SSN
unknown and score for demographic items gt
recommended cutoffs C (A) or SSN unknown and
either gt7 items match or lt than 8 match but with
high score
5
Liao et al (AJPH1998) Death Rate Ratios for
Hispanic, relative to Non-Hispanic White, Men age
45-64 years
A Exact match on social security B (A) or SSN
unknown and score for demographic items gt
recommended cutoffs C (A) or SSN unknown and
either gt7 items match or lt than 8 match but with
high score
6
Objective

• To examine effects of alternative matching
  criteria on mortality estimates obtained using
  NHIS-NDI Linked Mortality files
• Death rates
• Ratios between groups defined by race/ethnicity
  (non-Hispanic white, non-Hispanic black, and
  Hispanic), age, and sex
• Comparison by nativity and health status

7
Methods

• NHIS-NDI Linked Mortality files
• NHIS1989-1994 linked to NDI through 2002
• Mortality ascertained using four matching
  criteria
• Least conservative, Current NCHS recommendation,
  Moderately conservative, Most conservative
• Death rates and hazard ratios were calculated
  using survival analysis programs in Stata
• Standard errors were not estimated but cell sizes
  were reasonably large

8
Matching Criteria
9
Number of Deaths by Matching Criteria
Non-Hispanic white
Non-Hispanic black
Hispanic
Least conservative criteria Current NCHS
criteria Moderately conservative criteria Most
conservative criteria
10
Male versus Female Death Rate Ratios, by Age and
Race/Ethnicity
Non-Hispanic black
Hispanic
Death Rate Ratio
11
Death Rate Ratios for Race/Ethnicity, by AgeMales
Non-Hispanic black relative to non-Hispanic white
Hispanic relative to non-Hispanic white
Rate Ratios
12
Death Rate Ratios for Race/Ethnicity, by Age
Females
Non-Hispanic black relative to non-Hispanic white
Hispanic relative to non-Hispanic white
Rate Ratio
13
Hazard Ratios for Hispanic Females, by
Education (Relative to Non-Hispanic White
Females)
Hazard Ratio (Cox proportional hazards)
Hazard Ratios controlled for age
14
Hazard Ratios for Hispanic Females, by Health
Status (Relative to Non-Hispanic White Females)
Hazard Ratio (Cox proportional hazards)
Hazard Ratios controlled for age
15
Hazard Ratios for Hispanic Females, by Years in
the United States(Relative to Non-Hispanic White
Females)
Hazard ratio (Cox proportional hazards)
Hazard Ratios controlled for age
16
Previous study

• Finch BK et al Validity of Self-rated Health
  among Latino(a)s Am J Epidemiol 2002
• Examined whether self-rated health had
  differential mortality risks for Latino(a) adults
  by acculturation
• Concluded that self-reported poor health status
  was a weaker marker of mortality for the less
  acculturated
• Used 1989-1994 NHIS linked to the NDI through
  1997

17
Finch et al (AJE 2002) Hazard ratio for Health
Status, Hispanics (Poor/Fair Health relative to
Good/Excellent Health)
Hazard ratio (Cox proportional hazards)

• Adjusted for age, sex, marital status, education,
  employment, origin and income
• NHIS-NDI 1989-1994/1997

18
Hazard ratio by Health Status, Hispanics
(Poor/Fair Health relative to Good/Excellent
Health)
Hazard ratio (Cox proportional hazards)
Adjusted for age, sex, marital status,
education and income Finch et al used NHIS/NDI
1989-1994/1989-1997, also adjusted for employment
and origin
19
Summary

• Matching criteria differentially affect mortality
  estimates by race/ethnicity, sex, and age
• Groups most affected are Hispanics, especially
  Hispanic females and younger Hispanics
• Differential matching may affect inferences
• Education, Health status, Nativity

20
Summary

• Different criteria change number of matched
  deaths by about /- 5
• Small effect of different criteria on many
  inferences
• Non-Hispanic black, non-Hispanic white
• Older Hispanics
• Hispanics who have lived in the U.S. for a long
  time

21
Conclusions

• Overall associations likely to be robust to
  different matching criteria
• No current gold standard
• Sensitivity analysis would be appropriate for
  some studies