the naphsis/nchs collaboration past successes and future challenges salt lake city, ut june 3 rd –...
Post on 31-Mar-2015
213 Views
Preview:
TRANSCRIPT
The NAPHSIS/NCHS CollaborationPast Successes and Future Challenges
Salt Lake City, UT June 3rd – 7th, 2007
Estimation of US Life Tables for Minority Populations: Issues of Data
Quality and Availability
Elizabeth Arias, Ph.D.Mortality Statistics BranchDivision of Vital Statistics
National Center for Health Statistics
Will it be possible to estimate life tables for groups other than the White
and Black populations?
Data Limitations and Availability Vital Statistics and Medicare Data
US Life Table Methodology
Example: Comparison of White and Hispanic US Life Tables for Decennial Period 1999-2001
Data Limitations and Availability
Race/Ethnic Misclassification on Death Certificate
Race/Ethnic Classification on Centers for Medicare and Medicaid Services (CMS) Medicare Data
Race/Ethnic Misclassification on Death Certificates
Results based on study “The Validity of Race and Hispanic Origin
Reporting on Death Certificates in the United States,” by Arias, et al. (2007)
Study used the National Longitudinal Mortality Study (NLMS) – linkage of Current Population Surveys (1973, 1978-1998) with NCHS Mortality Data (Mortality follow-up 1979-1998)
For Sample of Decedents identified in the NLMS, results show that reporting on the death certificate is excellent for White and Black populations; less than optimal for API and Hispanic populations; and, poor for AIAN population.
There is some improvement between 1980s and 1990s for most of these groups
Ratio of CPS (self) to Death Certificate Report for the Sample of NLMS Decedents
1979-89 1990-98
White 1.00 1.00
Black 1.00 1.01
AIAN 1.46 1.30
API 1.12 1.07
Hispanic 1.03 1.05
Application of Validity Study to the Estimation of Life Tables
Evaluation of misclassification was carried out by age, sex, region, rural/urban residence, and co-ethnic concentration
These results can be used to adjust observed death rates
These adjusted death rates can then be used to estimate life tables
Ideally, this would be all that is needed……But,
The Role of CMS Medicare Data in the Production of US Life Tables
NCHS has traditionally used Medicare data to estimate mortality at the oldest ages for its Decennial Life Tables (Annual Life Tables since 1997)
It is believed that Medicare coverage is better because age-reporting is verified with date of birth, whereas the denominators of Vital Statistics rates come from Census estimates, which are not verified for age reporting
The 1999-2001 US Decennial Life Table Method blends Vital and Medicare rates to estimate mortality for ages 65 – 100
• q(x) for ages 65-94 are blended with progressive weight given to Medicare Data
• q(x) for ages 95-100 are derived exclusively from Medicare Data
CMS Medicare Race/Ethnic Classification
CMS Medicare Data derives its Race and Ethnic information from the Social Security Administration
Race and ethnic data is collected by SSA when individuals complete form SS-5 for SS Card.
Between 1936 and 1980 Race categories included in the SS-5 application were limited to: White, Negro or Other
As per OMB Directive No. 15 SSA revised the SS-5 (1980) by expanding the options to 5 categories, combining Race and Hispanic Origin : White (Non-Hispanic) Black (Non-Hispanic) Asian or Pacific Islander American Indian or Alaskan Native Hispanic
CMS Medicare Race/Ethnic Classification
The result is that CMS Race/Ethnic Categories are a combination of pre-1980 and post-1980 SSA Race/Ethnic Categories: 0=Unknown 1=White (Non-Hispanic) 2=Black (Non-Hispanic) 3=Other 4=Asian, Asian American or Pacific Islander 5=Hispanic 6=American Indian or Alaskan Native
Use of CMS Medicare for Groups Other than White or Black
How do we disentangle AIANs, APIs, and Hispanics from the combination of pre-1980 and post-1980 categories?
Experiment: Use NLMS – CMS linked Data. NLMS was recently linked to 1991-1995 CMS Medicare files
Compare CPS and CMS-Medicare Classification CPS has provided respondents with full-
range of race/ethnicity since 1977
Sensitivity: Percent of NLMS Respondents Correctly Identified by CMS – Medicare (1991-95)
Hispanic NHWhite NHBlack NHAIAN NHAPI
UK 2.0 1.5 1.6 1.5 2.0
NHWhite 78.5 97.7 2.1 37.6 4.9
NHBlack 2.0 0.1 95.1 4.0 0.4
Other 8.7 0.6 1.1 47.6 83.7
API 0.1 0.0 0.0 0.1 8.8
Hispanic 8.7 0.0 0.0 0.2 0.1
AIAN 0.0 0.0 0.0 8.9 0.0
Total 100.0 100.0 100.0 100.0 100.0
Predictive Value Positive: Percent of Respondents Identified by CMS who Self-Identified in the same group in NLMS
Hispanic NHWhite NHBlack NHAIAN NHAPI Total
UK 5.1 85.8 7.0 0.4 1.4 100.0
NHWhite 3.5 96.0 0.2 0.2 0.1 100.0
NHBlack 1.2 1.5 96.7 0.2 0.1 100.0
Other 15.4 23.7 3.2 8.2 41.3 100.0
API 1.6 24.1 0.8 0.3 63.1 100.0
Hispanic 95.1 3.6 0.7 0.2 0.3 100.0
AIAN 0.9 19.2 4.7 69.7 0.0 100.0
Effects on Estimation of Life Tables for these Populations
NLMS-CMS link suggests the majority of Hispanics, AIANs, and APIs are not easily identifiable in CMS Medicare Data.
Does it Matter? Can we do without Medicare Data?
Exploration Hispanic Mortality Compared to White Mortality
1999-2001 Following 3 Graphs compare Vital q(x) between
Hispanic (observed and adjusted for DC under-report) and White populations
Age-Specific Mortality, Total Population
0.000010
0.000100
0.001000
0.010000
0.100000
1.000000
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100
Age
THISPOBS TWHITEOBS THISPADJ
Age-Specific Mortality, Male Population
0.000010
0.000100
0.001000
0.010000
0.100000
1.000000
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100
Age
HISPMALEOBS WHITEMALEOBS HISPMALEADJ
Age-Specific Mortality, Female Population
0.000010
0.000100
0.001000
0.010000
0.100000
1.000000
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100
Age
HISPFEMOBS WHITEFEMOBS HISPFEMADJ
What do Mortality Patterns Say?
Even after Adjustment for DC misclassification, Hispanic Mortality remains lower.
Lower Mortality for Hispanics is concentrated in the older ages, except for Hispanic Females who show advantage throughout full age range.
Next, Closer look at Mortality at ages 65 and above
Mortality Ages 65-100, Total
0.000000
0.050000
0.100000
0.150000
0.200000
0.250000
0.300000
0.350000
0.400000
0.450000
65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100
Age
THispanic
TWhite
MedicareWhiteT
THISPADJ
Male Mortality Age 65-100, Males
0.000000
0.050000
0.100000
0.150000
0.200000
0.250000
0.300000
0.350000
0.400000
65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100
Age
HispMale
WhiteMale
MedicareWhiteMale
HispMaleADJ
Mortality Ages 65-100, Females
0.000000
0.050000
0.100000
0.150000
0.200000
0.250000
0.300000
0.350000
0.400000
0.450000
Age
HispFemale
WhiteFemale
MedicareWhiteFemale
HispFemalADJ
Effects of Using Medicare Data on Life Expectancy
Previous 3 graphs show that for the white population CMS Medicare shows higher mortality at ages 65-100 than Vital Statistics
What impact does including CMS Medicare Data have on Life Expectancy Estimates?
Experiment: Quantify the Effect of Excluding Medicare Data and Closing the Life Table at age 85 for the White Population:
Comparison of Life Expectancy - Blending Vital and Medicare Data for Ages 65-100 vs. Vital
Statistics Closed at Age 85 – White Population Decennial Method Close at Age 85 Diff
Total White Population• Birth 77.22 77.58 0.36• 65 17.63 18.06 0.43• 85 5.99 6.38 0.39
White Male • Birth 74.60 74.90 0.30• 65 16.01 16.38 0.37• 85 5.23 5.66 0.43
White Female• Birth 79.74 80.16 0.42• 65 18.95 19.43 0.48• 85 6.38 6.73 0.35
Comparison of White and Hispanic Life Expectancy Using Estimates based Solely on Vital Statistics, Closing
Table at Age 85 1999-2001 Decennial Period
Total Hispanic Total White Diff• Birth 79.53 77.58 1.95• 65 19.76 18.06 1.70• 85 7.88 6.38 1.50
Hispanic Male White Male• Birth 76.69 74.90 1.79• 65 18.10 16.38 1.72• 85 7.33 5.66 1.67
Hispanic Female White Female• Birth 82.31 80.16 2.15• 65 21.10 19.43 1.67• 85 8.23 6.73 1.50
How Do Vital Statistics Estimates Compare to NLMS Estimates of Hispanic Life Expectancy?
Vital Statistics NLMS Total Hispanic Diff
• Birth 79.53 80.12 0.59• 65 19.76 20.15 0.39• 85 7.88 8.03 0.15
Hispanic Male • Birth 76.69 77.25 0.56• 65 18.10 18.28 0.18• 85 7.33 7.37 0.04
Hispanic Female• Birth 82.31 83.35 1.04• 65 21.10 21.86 0.76• 85 8.23 8.50 0.27
Summary and Future Research and Exploration
Data Quality and Limitations Pose Challenges to the Production of Life Tables for Minority Populations.
Are they insurmountable? Perhaps, Perhaps Not On the plus side: We have been able to
identify and quantify Race/Ethnic misclassification on DC and use this information to correct the resulting under-count of deaths for affected groups
Summary and Future Research and Exploration
On the negative side: we may not be able to use Medicare data for a very long time, if ever for minority populations
We may need to accept that if we want life tables for Hispanics, APIs, and AIANs we will need to rely solely on Vital Statistics Can we accept this alternative? For example, can we accept that Hispanic life expectancy is
higher than that of NHWhites?• Previous studies using the NLMS, NHIS-NDI all show that
Hispanics indeed have lower mortality than NHWhites.• One study attributes the advantage to the Salmon Bias Effect,
but finds that this Effect applies only to Foreign Born Mexicans and Central/South Americans (Palloni and Arias, 2004)
• Could the large gap in mortality at the oldest ages we observed be due to Salmon Bias?
Next Steps Explore upcoming NLMS – CMS 1996-
2000 Medicare linkage for possibility of re-classification of CMS categories for use in the 1999-2001 Decennial Life Tables.
Repeat comparative exercises for AIANs and APIs
Explore other statistical modeling techniques for estimates of old-age mortality for these populations.
top related