Technical Report No. 15-91
Combining diagnostic categories to improve agreement between death certificate and autopsy classifications of cause of death for atomic bomb survivors, 1950-87
Carter RL, Ron E, Mabuchi KSummary
Several investigators have observed less-than-desirable agreement between death certificate diagnoses and autopsy diagnoses for most specific causes of death, and even for some causes grouped by major disease category. Our results from data on 5130 autopsies of members of the Life Span Study cohort of atomic bomb survivors in Hiroshima and Nagasaki conducted prior to September 1987 were equally discouraging. Among diseases with more than 10 cases observed, confirmation rates ranged from 13% to 97% and detection rates from 6% to 90%. Both rates were greater than 70% for only 6 of 60 disease categories studied and for only 1 of 16 categories defined by major International Classification of Disease categories (neoplasms). This deficiency suggests cautious interpretation of results from studies based on death certificate diagnoses. To determine whether any groupings of diagnoses might meet acceptable accuracy requirements, we applied a hierarchical clustering method to data from these 5130 cohort members. The resulting classification system had 10 categories: breast cancer; other female cancers; cancers of the digestive organs; cancer of the larynx; leukemia; nasal, ear, or sinus cancer; tongue cancer; external causes; vascular disease; and all other causes. Confirmation and detection rates for each of these categories were at least 66%. Although the categories are broad, particularly for nonneoplastic diseases, further divisions led to unacceptable accuracy rates for some of the resulting diagnostic groups. Using the derived classification system, there was 72% agreement overall between death certificate and autopsy diagnoses compared to 53% agreement for a second system obtained by grouping strictly by major disease category. Eighty-seven percent agreement was observed for a similar classification system with vascular disease grouped with all other nonneoplastic diseases. Further agglomeration achieved very little additional improvement. Accuracy rates for some of the categories of the 10-category diagnostic system defined above varied with various covariates. For example, accuracy decreased with increasing age at death for most of these categories. Thus, subpopulations exist for which accuracy rates can be expected to be either better or worse than for the whole population. Although these results do not necessarily dictate which diseases and/or populations should be studied in future cause-specific mortality investigations, they do provide investigators with useful information pertinent to the planning of their study, analysis of the data, and interpretation of the results.
Several investigators have observed less-than-desirable agreement between death certificate diagnoses and autopsy diagnoses for most specific causes of death, and even for some causes grouped by major disease category. Our results from data on 5130 autopsies of members of the Life Span Study cohort of atomic bomb survivors in Hiroshima and Nagasaki conducted prior to September 1987 were equally discouraging. Among diseases with more than 10 cases observed, confirmation rates ranged from 13% to 97% and detection rates from 6% to 90%. Both rates were greater than 70% for only 6 of 60 disease categories studied and for only 1 of 16 categories defined by major International Classification of Disease categories (neoplasms). This deficiency suggests cautious interpretation of results from studies based on death certificate diagnoses. To determine whether any groupings of diagnoses might meet acceptable accuracy requirements, we applied a hierarchical clustering method to data from these 5130 cohort members. The resulting classification system had 10 categories: breast cancer; other female cancers; cancers of the digestive organs; cancer of the larynx; leukemia; nasal, ear, or sinus cancer; tongue cancer; external causes; vascular disease; and all other causes. Confirmation and detection rates for each of these categories were at least 66%. Although the categories are broad, particularly for nonneoplastic diseases, further divisions led to unacceptable accuracy rates for some of the resulting diagnostic groups. Using the derived classification system, there was 72% agreement overall between death certificate and autopsy diagnoses compared to 53% agreement for a second system obtained by grouping strictly by major disease category. Eighty-seven percent agreement was observed for a similar classification system with vascular disease grouped with all other nonneoplastic diseases. Further agglomeration achieved very little additional improvement. Accuracy rates for some of the categories of the 10-category diagnostic system defined above varied with various covariates. For example, accuracy decreased with increasing age at death for most of these categories. Thus, subpopulations exist for which accuracy rates can be expected to be either better or worse than for the whole population. Although these results do not necessarily dictate which diseases and/or populations should be studied in future cause-specific mortality investigations, they do provide investigators with useful information pertinent to the planning of their study, analysis of the data, and interpretation of the results.