Estimation of Cancer Incidence Trends Adjusted for Changes in Screening and Detection Processes.

Cancer is a major public health issue, and monitoring its incidence is important to suggest and evaluate the impact of preventive interventions. However, estimating trends in cancer incidence is often difficult due to changes in screening or other detection processes over time, which can artificially inflate or deflate the observed incidences. We propose a new method for estimating trends in cancer incidence adjusted for such changes, using a constrained Almon distributed lag model. Unlike other approaches, our method does not rely on any knowledge of cancer progression, or detailed evolution of screening practice over time. It requires the registration of the stages (I-IV) of detected cancers while assuming that the distribution of these stages remains constant in the absence of any change in screening practice. Our method is able to recover the real underlying cancer incidence in simulated data reproducing either no change or a gradual or sudden change in screening practice. For illustration, it is applied to registry data from the canton of Geneva, Switzerland, to estimate breast cancer incidence for the period 1991-2016, where it downwardly corrects the observed incidence when an organized screening program was started.