We developed a method to identify maternal deaths (deaths to women wit
hin 365 days of delivery) by linking Tennessee vital records. A comput
erized algorithm compared personal identifiers from the death certific
ates of reproductive-aged women to maternal identifiers on birth and f
etal death certificates. For each decedent record which met the study
criteria, the algorithm calculated a ''match score'' by comparing comm
on elements in both files. The algorithm awarded full credit for data
elements that agree exactly, partial credit for elements in partial ag
reement, and subtracted credit for information that mismatched. Match
scores ranged from 0 to 35 for the 9,009 deaths in women 10-55 years o
f age during the three study years, with the majority of scores (96.3%
being 0 for ''no match.'' Match scores of 1 to 8 were obtained by 153
(1.7%) of decedent records, while scores greater than 9 were obtained
by 184 (2.0%) of decedent records. We used nurse-abstracted hospital,
autopsy, and coroner records as our standard to verify the linkages.
Manual review of personal identifiers showed that scores of 12 or less
were not a match while scores of 13 or more indicated ''true'' matche
s. Based on this cutoff, the linkage algorithm yielded 130 maternal de
aths. Of these, 32 (25%) were classified as truly pregnancy-related up
on medical record review by an obstetrician. The remaining 98 deaths w
ere associated only temporally with pregnancy. During the same time pe
riod, 16 individuals were identified to the State Health Department on
their death certificates as dying from pregnancy-related causes, incl
uding one not identified by the linkage process. Thus, computer linkag
e supplemented by record review proved to be a reliable and efficient
means of identifying maternal deaths and doubled ascertainment during
the study period. Limitations of ascertainment by this method are due
primarily to pregnancy terminations that do not generate a fetal vital
record (spontaneous and induced abortions, ectopic pregnancies), whic
h makes vital records linkage impossible.