The interpretation of time trends.

The requirements for valid comparison over time of cancer incidence and mortality are given in the opening paragraphs of this chapter. Among the more important of these is comparability between ICD revisions. Yet those entrusted with the decennial revisions of the International Statistical Classification of Diseases, Injuries and Causes of Death (ICD) have to try to reconcile several frequently contradictory demands. These include (a) meeting the needs of an ever widening number of users (hence, the change in title for the 10th revision to International Statistical Classification of Diseases and Related Health Problems); (b) incorporation of advances in the understanding of the biology of disease while avoiding ephemeral notions and classifications; and (c) the preservation of the continuity of time series. These desiderata have not always been met: some of the resulting problems have been discussed above. Despite the foregoing, the comparison of time trends in cancer incidence and mortality is a valid and useful exercise particularly when the investigator systematically considers the possible sources of bias and error. A low rate, if based on over a million person-years of observation, is, if underregistration can be ruled out, likely to reflect reality. Nevertheless, the smaller the number of cases and the greater the standard error of the rate, notably for the less frequent sites and the rarer histological types, the more attention needs to be paid to questions of possible artefact. Should the data for a given area or time period appear out of line with what might be expected, it may be worthwhile to check with the registry in question to see whether the trend could be explained on the basis of local circumstances, custom or practice. A formal review of the evidence for an apparent change over time, drawing on the experience of epidemiologists, pathologists and clinicians, as for example in non-Hodgkin's lymphoma, can be very useful (Levine and Hoover, 1992). Hitherto, much of the study of time trends of both incidence and mortality, notably those comparing international data, has been based on data published at the level of three digits in the ICD. Although such analyses will continue to be illuminating, future work is likely to give much greater emphasis to subsite and histological type, information that is difficult to publish routinely.(ABSTRACT TRUNCATED AT 400 WORDS)