Statistical analysis of donation--transfusion data with complex correlation

Blood-borne transmission of disease is estimated from linked data records from blood donors and transfusion recipients. However, such data are characterized by complex correlation due to donors typically contributing many donations and recipients being transfused with multiple units of blood product. In this paper, we present a method for analysing such data, by using a modification of a nested case-control design. For recipients who develop the disease of interest (cases) and their matched controls, all donors who contributed blood to these individuals define clusters or 'families' of related individuals. Using a Cox regression model for the hazard of the individuals within clusters of donors, we estimate the risk of transmission, and a bootstrap step provides valid standard errors provided the clusters are independent. As an illustration, we apply the method to the analysis of a large database of Swedish donor and recipient records linked to the population cancer register. We investigate whether there is an increased risk of cancer in recipients transfused with blood from donors who develop cancer after donating. Our method provides a powerful alternative to the small 'look-back' studies typical of transfusion medicine and can make an important contribution to haemovigilance efforts.