Breast cancer natural history models and risk prediction in mammography screening cohorts

In this thesis, the foundations are laid for a new natural history model for breast cancer—specifically designed to take advantage of detailed screening cohorts. Three diverse applications of this model are then presented.

Study I develops the statistical framework for the natural history model, and shows with simulations that the model parameters can be estimated based on only the information available at diagnosis. It also describes how to adjust for random left truncation—an important aspect to consider when studying prospective cohorts.

In Study II, the newly developed natural history model is applied to a Swedish mammography screening cohort. It estimates the population-level distributions of age at onset and tumor volume doubling time. As an extension, the tumor volume doubling time is allowed to depend on the age at onset. The study also estimates the rate of symptomatic detection and screening sensitivity as functions of tumor size. Simulations are used to validate the estimates.

Study III shifts the focus from inference to risk prediction. The natural history model is modified to incorporate risk factors separately in each of the four components of the model. Short-term risk prediction is then performed on the screening cohort and the results are compared to a conventional approach to breast cancer risk prediction. The study also develops novel predictions based on, for example, having experienced tumor onset, having a tumor detected at the next screening, and having a tumor detected before it reaches a certain size if attending the next screening.

In Study IV, the model is used to study the effect that certain acquisition parameters used in mammography have on the detectability of the breast cancer tumor. With the model, it is possible to more directly study the mammography screening sensitivity, compared to the ad hoc definition of sensitivity commonly seen in the screening literature. It was identified that the compressed breast thickness—in addition to the percent mammographic density and latent tumor size—was inversely associated with the screening sensitivity.

List of scientific papers

I. Rickard Strandberg & Keith Humphreys. Statistical Models of Tumour Onset and Growth for Modern Breast Cancer Screening Cohorts. Mathematical Biosciences. 2019, 318, 108270.
https://doi.org/10.1016/j.mbs.2019.108270

II. Rickard Strandberg, Kamila Czene, Mikael Eriksson, Per Hall, Keith Humphreys. Estimating Distributions of Breast Cancer Onset and Growth in a Swedish Mammography Screening Cohort. Cancer Epidemiology, Biomarkers and Prevention. 2022, 31 (3), 569-577.
https://doi.org/10.1158/1055-9965.EPI-21-1011

III. Rickard Strandberg, Kamila Czene, Per Hall, Keith Humphreys Novel Predictions Of Breast Cancer Risk In Mammography Screening Cohorts. [Manuscript]

IV. .Rickard Strandberg, Maya Alsheh Ali, Kamila Czene, Per Hall, Keith Humphreys. Modelling the effects of Mammographic Density and Compressed Breast Thickness on Mammographic Sensitivity: A Natural History Approach. [Manuscript]