Reliability of cancer mortality statistics in Ontario: a comparison of incident and death diagnoses, 1979-1983.

We compared the underlying cause of cancer death listed on death certificates, to the registry diagnosis from the incident file in the Ontario Cancer Registry (OCR). For the 68,772 cancer deaths having both a registry diagnosis and a cancer cause of death, 79.3% agreed between the two sources at the third digit level of ICD-9; this rose to 85.8% when sites were aggregated into about 30-site groups (positive predictive value 85.8%, sensitivity 82.9%). The most common sites, accounting for greater than 80% of all cancer deaths, all had agreement rates above 80%. Sites of questionable reliability, comprising less than 10% of all cancer deaths, included liver and larynx, and most other ill-defined and unspecified sites. Recommendations to improve the quality of published cancer mortality statistics include combining colon and rectum, and the non-Hodgkin's lymphomas. Caution in the use and interpretation of statistics for cancers of the liver and larynx is suggested owing to poor reliability.