[Comparison of ICD coding between mortality statistics and study-intern retrospective re-coding].

Objectives: The aim of this study was to assess agreement between study-intern nosologist death certificate coding and official mortality statistics ICD codes.

Methods: During the follow-up of a historical cohort study conducted in Germany, original death certificates were obtained for deceased cohort members and ICD coded by a study-intern nosologist. Additionally, ICD codes for these study subjects were obtained from a state statistical office responsible for mortality statistics. A weighted inter-observer agreement for these two sources was calculated.

Results: In total, 406 ICD codes were available from both sources. 219 (53.9%) of these ICD codes completely agreed on the highest level possible (three- or four-digits). Agreement was found on the three-digit level with a difference at the fourth digit in 42 (10.3%) causes of death. Agreement within diagnosis groups or within chapters of disease was found in 21 (5.2%) and 49 (12.1%) causes of death, respectively. The weighted kappa for the overall inter-observer agreement was 0.67 [95% confidence interval (CI): 0.63-0.71]. Within chapters of disease, agreement was higher for neoplasms (kappa=0.88; 95% CI: 0.83-0.93) than for cardiovascular diseases (kappa=0.69; CI: 0.62-0.76).

Conclusions: Overall level of agreement between the two coding sources was not very good. In Germany, the quality of coding has not improved substantially in the past two decades. The introduction of automatic coding systems, multicausal coding and a mortality register could improve the quality of ICD coding in Germany.