The ECOG-ACRIN Cancer Research Group and the National Cancer Institute have launched the Tomosynthesis Mammographic Imaging Screening Trial (TMIST), a randomized trial that will enroll 165,000 U.S. women to compare two types of digital mammography for breast cancer screening: tomosynthesis (known as three-dimensional, or 3-D) and conventional (two-dimensional, or 2-D).
Led by Professor Constantine Gatsonis, Brown University’s Center for Statistical Sciences and the Department of Biostatistics will serve as the hub for handling the massive statistical challenges of the study, starting from its innovative design through to the eventual analysis of the results.
Q: What are the main measures of the study and what precisely will that data tell us about breast cancer screening?
The primary "endpoint" in the study is the occurrence of an advanced, life threatening breast cancer in a participant, during a period of 4.5 years from study entry. The trial will compare the proportions of women in the two arms of the study – the 3D arm or the 2D arm – who experience such a cancer within four and a half years from randomization. The key idea is that the more effective screening modality will be the one with the fewer advanced cancers. Notice that this way of evaluating mammography screening is not focused on how accurate it is, but on how it affects a woman's life.
Q: Why is the trial so large?
The reason for the large enrollment is that the likelihood of an advanced breast cancer (or any breast cancer, for that matter) in a 4.5 year period is fairly low, thus requiring a large sample size to ensure adequate statistical power to detect a meaningful difference between the two arms of the trial.
Q: Are there particular innovations in the study design?
Here are three important innovations:
- Definitive trials of the impact of breast screening in the past have had mortality as the primary endpoint. To perform such a comparison in this setting would require an unrealistically long follow-up and a large sample size. TMIST uses an innovative endpoint (the occurrence of an advanced cancer), which can be assessed earlier than breast cancer mortality and is prevalent enough to permit a study to be completed in a realistic timeframe.
- In addition to the primary comparison, TMIST was designed to generate a rich, population-based biospecimen bank which will include (a) specimens of all cancers found in the participating women and also specimens from a subset of the benign biopsies and (b) blood samples and mouth swabs from all study participants who volunteer to provide them. This biospecimen bank will be an enormously valuable resource for genetic and genomic research in breast cancer.
- TMIST will also generate a vast database of imaging scans and thus provide a population-based resource for radiomics research to develop machine-learning tools for breast cancer detection and prediction of outcomes. It will also enable large-scale radiogenomics research, combining information from imaging and genetic/genomic studies.
Q: How is Brown involved in the design, conduct and analysis of the study?
The Center for Statistical Sciences (CSS) and the Department of Biostatistics at Brown are the statistical nerve center for this trial. Our work is done in our capacity as partners in the Statistics and Data Management Center of ECOG-ACRIN, the collaborative group that launched the trial. However, we are the statistics entity solely responsible for this trial.
CSS faculty provided methodologic leadership in the design of the trial, working closely with the PI, Dr Etta Pisano from Harvard Medical School. CSS faculty and staff biostatisticians are leading the development of the infrastructure for data collection and data monitoring and reporting. During the course of the trial, Brown biostatisticians will monitor quality, completeness and accuracy of the incoming data, conduct analyses,and interface with entities such as the Data and Safety Monitoring Board. After the data collection is complete, Brown biostatisticians will analyze the data for this multifaceted study and collaborate with the PI and other study investigators in developing manuscripts.
Q: What do you expect will be some of the challenges in analyzing the study data?
Although we have long experience in large screening trials, this one is at least three times larger in size and involves very extensive and multifaceted data and biospecimen collection. So this trial is very much a project of the health data science era. The analysis of large, high-dimensional data from several sources (clinical, patient reports, imaging, genetic, genomic) will undoubtedly present significant challenges and, I hope and expect, will also lead to new methodologic advances.