Death, Where is Thy Record? Death Ascertainment for Veterans Presented by Denise M. Hynes, Ph.D., R.N. Director, VIReC Research Health Scientist, MCHSPR COE

1
Death, Where is Thy Record? Death Ascertainment
for Veterans Presented byDenise M. Hynes,
Ph.D., R.N.Director, VIReCResearch Health
Scientist, MCHSPR COE
VIReC CyberSeminar Series 2006 VA Databases
and Methods

2
VIReC VA-NDI Mortality Data Merge Project (SDR
03-157)Min-Woong Sohn, PhD (PI)Denise Hynes,
PhDNoreen Arnold, M.S.Charles Maynard, PhD
3
Why is Mortality Ascertainment Using Multiple
Data Sources Important?

• Mortality is a common outcome measure in
  epidemiologic and clinical trials research
• Accuracy of vital status important for
  calculation of other related outcomes
• Conflicting results using single data sets
• More deaths in Medicare data than BIRLS Death
  File
• Cost effective VA use of the NDI

4
Mortality Data Merge Project Objectives

• Analyze and compare four sources for death data
  available within the VA
• Determine the benefit of using the NDI
• Develop a methodology to use all sources cost
  effectively
• Propose a strategy to maintain a Death Registry

5
NDI

• National Death Index
• Gold standard for death ascertainment
• Maintained by National Center for Health
  Statistics
• States report all deaths from death certificates
  filed in the state
• Expensive
• 0.15 per record per year
• 5.00 per record for cause of death

6
NDI

• 10 million veterans
• 1.5 million per year to search for deaths
• 800,000 deaths 1999-2002
• 4 million to get the cause of death
• 15-month lag
• 2004 deaths available April-May 2006
• No deaths outside the US

7
VA Death Sources

• BIRLS Death File
• Main VA Death File
• Medical SAS Inpatient Datasets
• VA inpatient deaths
• SSA Death Master File
• Over 78 million deaths
• Medicare Vital Status File
• Deaths in Medicare beneficiaries who are known to
  the VA
• Received Annually

8
Methodology

• Cohort
• Known to VA 1997-2002 via VHA or VBA
• 8.6 million veterans
• Deaths in 1999-2002
• Presumed Living
• Not Dead Yet
• VHA VBA activity after 12/31/2002 or known to
  Medicare
• Unknown Status

9
Methodology

• Random samples of deaths for NDI search
• validate death
• Random sample of presumed living and unknown
  status for NDI and SSA Epidemiological search
• identify unknown deaths
• validate presumed living status
• validate unknown status

10
Methodology
11
Results

• Deaths in multiple sources same date (88)
• 98.3 in NDI
• 98.2 in NDI and year and month of death matched
• Deaths in multiple sources different date (8)
• 97.0 in NDI if year same
• 84.3 in NDI if year different
• Routine to select the best date of death
• Deaths in only one source (4)
• Varied 73.3 for BIRLS and 95.8 for Medicare in
  NDI with year and month match

12
Results

• Presumed Living
• No new deaths identified in NDI
• 90.6 presumed living according to the SSA
• 7.2 unknown status
• 1.6 invalid demographics
• 0.6 deaths in 2004

13
Results

• Unknown Status Under 65 (97)
• 6 deaths identified by NDI 0.3 of sample
• 87.6 presumed living
• 8.0 unknown status
• 4.0 invalid demographics
• 0.4 deceased

14
Results

• Unknown Status - 65 and older (3)
• 179 deaths identified by NDI 9 of sample
• 5.3 presumed living
• 20 unknown status
• 73.9 had invalid demographics
• 0.8 deceased
• only 0.3 of total population

15
Results

• Overall Sample 3,000 Alive 1/1/1999
• 292 NDI deaths 1999-2002
• Combined Sources
• Sensitivity 98.3
• Specificity 99.8
• 2 unidentified deaths in unknown status (0.4)
• 3 unidentified deaths in presumed living (0.4)
• 2 had issues with demographics
• 1 had VHA use two years after date of death
• 5 deaths not in NDI

16
Conclusions

• Not cost effective to use the NDI to build a
  registry
• Registry could be built using the four available
  sources
• Include presumed living status in registry
• VHA National Data Systems Death Database