The use of models for estimating overall incidence trend

Several approaches to evaluate an overall trend including Registries with different length of activity with a defined period of time have been evaluated in the cancer trend study of the Italian Network of Cancer Registries (AIRT). The study used the nine Registries of the pool AIRT in the period 1986-1997. Data for 1986-1993 from the Registry of Ragusa have been excluded to simulate the inclusion of a Registry with partial observation. Four different approaches have been used: Golden standard (GS) as a weighted mean of the observed standardised (European population) rates in the whole pool AIRT; 2) Raw trend (TG) computed with partial data of Ragusa; 3) Model trend (TM) estimated by means of a log-linear model including age, registry and period as independent variables and applied on the partial data set; Adjusted trend (TC) computed on complete data for eight Registries, observed 1994-1997 data of Ragusa and estimated data for Ragusa for 1986-1993 computed by (3) model. The trend estimates, as estimated annual percent change of the standardised rates (EAPC) have been compared with GS. The results of this preliminary analysis do not recommend the use of raw trend and indicate adjusted trend as the most reliable method for trend estimate.