Predicting attendance for breast screening using routinely collected data

The aim of this paper is to predict attendance if the age range for routine invitation to breast screening were to be extended. The response to the most recent screening invitation is modelled for women eligible for screening if the age range were extended. The independent variables include (i) the woman's characteristics: her screening history; the deprivation score of the area she lives in and (ii) the characteristics of the screening: whether the screening took place in a mobile van or at a static site; and time of the year. The predictive ability of the regression model is tested by goodness of fit measures and by predicting attendance for a holdout sample of the data and for women who participated in a demonstration project. The modelling of attendance is quite successful in that most hypothesised variables have the expected sign. Moreover, the predictive ability of the model is satisfactory in terms of goodness of fit statistics and in terms of accuracy of predictions for a holdout sample. The model predicts less well for the demonstration project possibly because it is less representative of usual screening practice.