Cause-specific mortality: understanding uncertain tips of the disease iceberg.

STUDY OBJECTIVE: To determine the extent to which individual diseases, when recorded as being present shortly before death, were certified as causes of death. DESIGN: Retrospective cohort study in which the "subjects" were computerised linked records. SETTING: Six districts in the Oxford Regional Health Authority area (covering a population of 1.9 million people). SUBJECTS: Linked abstracts of hospital records and death certificates for people who died within four weeks and, for some diseases, within one year of hospital admission. MAIN OUTCOME MEASURES: The percentage of people with each disease for whom the disease was recorded as the underlying cause of death, was recorded elsewhere on the death certificate, or was not certified as a cause of death at all. RESULTS: Three broad patterns of certification are distinguished. Firstly, there were diseases that were usually recorded on death certificates when death occurred within four weeks of hospital care of them. Examples included lung cancer (on 91% of such death certificates), breast cancer (92%), leukaemia and lymphoma (90%), anterior horn cell disease (89%), multiple sclerosis (89%), myocardial infarction (90%), stroke (93%), aortic aneurysm (87%), and spina bifida (89%). These diseases were also usually certified as the underlying cause of death. Secondly, there were diseases which, when present within four weeks of death, were commonly recorded on death certificates but often not as the underlying cause of death. Examples included tuberculosis (on 76% of such certificates; underlying cause on 54%), thyroid disease (49%; 21%), diabetes mellitus (69%; 30%) and hypertension (43%; 22%). Thirdly, there were conditions which, when death occurred within four weeks of their treatment, were recorded on the death certificate in a minority of cases only. Examples of these included fractured neck of femur (on 25% of such certificates), asthma (37%), and anaemia (22%). Not surprisingly, there was "convergence" in certification practice towards the common cardiovascular and respiratory causes of death. There was also evidence that conditions regarded as avoidable causes of death may not have been certified when present at death in some patients. CONCLUSION: When uses are made of mortality statistics alone, it is important to know which category of certification practice the disease of interest is likely to be in. Linkage between morbidity and mortality records, and multiple cause analysis of mortality, would considerably improve the ability to quantify mortality associated with individual diseases.