Lung cancer has the second highest cancer incidence for both genders, second only to prostate cancer in men and breast cancer in women. There is a continued need for effective secondary prevention in the form of early detection and early treatment. However, screening trials have produced ambiguous results as to their effectiveness and in particular have failed to show a mortality benefit due to screening in the context of a randomized clinical trial (RCT). It has been suggested that it is the nature of tumor progression in lung cancer that limits the effectiveness of early detection methods. There may be very aggressive cancers that metastasize rapidly at a small size and there may be very slow-growing tumors that can be detected on screening scans but which are unlikely to shorten a patient’s life expectancy. This latter phenomenon is commonly referred to as “over-diagnosis”. We evaluate data on tumor size and stage from two data-sets on lung cancer screening, namely the Mayo Lung Project (MLP) and the Mayo CT study, in order to examine the evidence for over-diagnosis directly. In contrast to classical estimation methodology used to evaluate screening data (i.e. Walter and Day) from a single screening study, data from different screening modalities (CT and chest x-ray) allows for clustering of cancers with respect to tumor progression dynamics. An initial homogeneity analysis of the two data-sets suggests the presence of two or more clusters having distinct expected size at stage transitions. We use a model that parameterizes tumor growth and stage transition by the evolutionary parameters of branching fraction (f) and the cell mutation rate (u) in order to simulate likelihoods for cancers from both the MLP and the Mayo CT study. Clustering of MLP cancers based on a likelihood similarity matrix indicates the presence of two distinct clusters (A and B) having expected mean size at stage transition of 94 mm (CI: 64,100) and 6.1 mm (CI: 5,20) respectively. The tumor stage and size distribution evident in the Mayo CT results are consistent with a reduction in representation of Cluster A cancers due to higher sensitivity of the prevalence CT screen and detection of Cluster B cancers at smaller sizes than previously observed. These results indicate that in order to achieve a reduction in lung cancer mortality by CT screening may require a lower detection threshold than current estimates suggest.

* Join us for light refreshments and meet our guest from 2:00 to 2:10 in the lobby of Duncan Hall. The colloquium begins at 2:10 and ends at 3:10. Open to the general public.