By the end of the contract period, 36 cancer registries in 23 countries were providing data.
Each collaborating centre follows a common protocol. The protocol demands that they provide, each year, data on all recorded cases of childhood leukaemia and lymphoma. These data are provided in the form of an updated file, each line comprising the following minimal information on each registered case number, date of birth, sex, date of diagnosis, place of residence, basis of diagnosis, and histological diagnosis. In addition, each centre provides annual estimates of the childhood population, in the maximum detail (age/sex grouping, and place of residence) as possible. Collation of the data from the different centres is carried out at IARC. The case data are transformed into a standard format for analysis, including recoding of cancer diagnoses to standard ICD-O codes using the "CONVERT" programs (Ferlay, 1994). All data sets are submitted to a standard set of controls to check for internal inconsistencies-eg in age vs date of birth. Impossible or unlikely combinations of values for recorded cases are returned to the registry for checking.
The estimate excess radiation exposure to the populations of the areas under study, for the first and subsequent years post-accident are provided by UNSCEAR. Although national data on radiation levels, and estimated personal exposures are produced in most of the participating countries, the methods used are not comparable, so that the greater detail available is of no value in the analysis of international datasets.
By the end of the contract period (30/6/95), the protocol required participants to have completed data submission to December 1992, and the majority of centres had done so - see Table A1. All received data sets had been verified for internal consistency, as described above. The case and population at risk data are used to calculate incidence rates (by diagnosis, age group, sex, geographic subunit). The coherence of the estimated incidence with expected values is checked. Other controls on data quality include calculation of percentage of cases histologically verified, or registered as "Death Certificate Only" cases. Trends in these various estimators over time are an important indicator of the value of the data in a comparative study.
The study involves collection and analysis of data on new cases of leu kaemia and lymphoma in children aged 0-14 in Europe, and comparison of incidence rates with estimated population-level exposure to radiation from the accident at Chernobyl in 1986. Leukaemia is used to monitor possible carcinogenic effects since it is the cancer most sensitive to radiation dose, appears earliest following exposure, the background level varies rather little, and there are no known other variables which strongly influence risk. The diagnostic distinction between lymphoid leukaemia and non-Hodgkin lymphoma is rather arbitrary, so lymphoma data are recorded to. ensure that any observed trends reflect change in risk and not change in nosology. The data on leukaemia/lymphoma derive from cancer registries in Austria, Bulgaria, Byelorussia, Czechoslovakia, Denmark, Estonia, Finland, France, Germany, Hungary, Italy, Lithuania, Netherlands, Norway, poland, Russia, Slovenia, Sweden, Switzerland and UK. Registries supply on an annual basis data on 8 variables for all new cases of leukaemia and lymphoma in their area. All participants have provided case data since 1980 at least, and for many from earlier periods also. Participants supply annual population estimates, by age, sex and place of residence. The data on registered cases and population are used to calculate the risk of cancer; these risk estimates are then compared with estimates of population-level exposure to radiation supplied by UNSCEAR. The estimates are prepared by a uniform modelling procedure and hence are comparable between countries. Data analysis is carried out centrally. The major emphasis is on temporal and geographic patterns of incidence. Although rates could be calculated for quite small areas, they would be based on very small numbers; it is proposed to restrict analysis to a priori large regions which correspond to different levels of estimated radiation exposure prepared by UNSCEAR although it is feasible to prepare estimates for different geographic units. Prior estimates suggest that increases in cancer incidence outside the vicinity of the accident will be undetectable against the 'normal' background incidence of cancer. The study is designed to test this hypothesis. Within Russia and Byelorussia the populations which received relatively large doses are big enough to give the study sufficient power to compare numbers of cases observed with those predicted by current dose-effect model