Public-Use NHIS Linked Mortality Files (Lochner K. AJE 168:336-344, 2008).

Public-Use NHIS Linked Mortality Files

<<Lochner_K_08_[NHIS_Mort_Files].pdf>>
Folks,

This paper shows the similarity in findings between using the public-use version of the NHIS Linked Mortality Files (released last September?) and the restricted use data available through the NCHS Research Data Center.   The former had modified information that might otherwise be used to identify individuals.  The authors ran some models and revealed that results can be reasonably close for their example.  However they also stated:  "Moreover, caution in using the public-use files is urged for researchers requiring more detail on timing of death or age or when examining the mortality patterns of small subgroups of the population, such as numerically small racial/ethnic minority groups, very old individuals, or young adults. This is particularly the case when cause-specific analyses of such numerically small demographic subgroups are performed."  So the paper offers some good (and some bad) news.

Carl

Lochner K, Hummer RA, Bartee S, Wheatcroft G, Cox C.  The Public-Use National Health Interview Survey
Linked Mortality Files:  Methods of Reidentification Risk Avoidance and Comparative Analysis
     Am. J. Epidemiol. 2008 168: 336-344; doi:10.1093/aje/kwn123.
        http://aje.oxfordjournals.org/cgi/content/abstract/168/3/336?etoc

The National Center for Health Statistics (NCHS) conducts mortality follow-up for its major population-based
surveys. In 2004, NCHS updated the mortality follow-up for the 1986–2000 National Health Interview Survey
(NHIS) years, which because of confidentiality protections was made available only through the NCHS Research
Data Center. In 2007, NCHS released a public-use version of the NHIS Linked Mortality Files that includes a limited
amount of perturbed information for decedents. The modification of the public-use version included conducting
a reidentification risk scenario to determine records at risk for reidentification and then imputing values for either
date or cause of death for a select sample of records. To demonstrate the comparability between the publicuse
and restricted-use versions of the linked mortality files, the authors estimated relative hazards for all-cause
and cause-specific mortality risk using a Cox proportional hazards model. The pooled 1986–2000 NHIS Linked
Mortality Files contain 1,576,171 records and 120,765 deaths. The sample for the comparative analyses included
897,232 records and 114,264 deaths. The comparative analyses show that the two data files yield very similar
results for both all-cause and cause-specific mortality. Analytical considerations when examining cause-specific
analyses of numerically small demographic subgroups are addressed.

confidentiality; epidemiologic methods; health surveys; longitudinal studies; mortality