The increasing parallel use of ICD-9 and ICD-10 complicates the compar
ability of coded diagnoses. This is the reason why we developed a symm
etric table for interactive conversion between ICD-9 and CD-10, based
on a vector space text-retrieval method that resulted in unambiguous m
apping from ICD-9 to ICD-10 in 64%, from ICD-10 to ICD-9 in 87%, of al
l three- and four-character classes of the tabular list. Out of the re
maining 13% of multi-valued relations, a table for automated mapping f
rom ICD-10 to ICD-9 was created. In 9% of cases, the selection offered
no problems. A compromise between preserving information content and
maintain ng the logical integrity had to be found in 2.4%; in 1.6% aut
omated mapping was impossible because of newly defined concepts and st
ructural differences between ICD-9 and ICD-10 that are not counterbala
nced by a consistent system of residual categories. We recommend that
in a future revision of the ICD, compatibility with the then existing
classification system should be considered.