**5. Conclusions**

Screening tends to detect breast cancers when they are smaller and at earlier stage and has a marked effect on the detection of in situ cancer [40]. The detection of some of these

cancers, which are very slow growing or indolent, provides little or no benefit and should be considered as overdetection. In some cases, however, the degree of overdetection has been greatly overestimated [3,4,21], or at least the estimates are not on solid ground.

Currently, because there are no reliable and accepted in vivo tests to determine which cancers will be progressive and no way of determining which women will die of other causes before an undetected progressive cancer is found clinically, overdetection is a necessary tradeoff for the substantial mortality and morbidity reduction opportunities provided by screening. Nevertheless, overdetection, and more importantly, overtreatment remain issues that must be addressed, particularly for in situ disease.

At the same time, there are opportunities for the improvement of detecting aggressive cancers through screening methods that are preferentially sensitive to these cancers, such as contrast-enhanced imaging [41], or approaches such as breast tomosynthesis that are intended to make cancers more conspicuous by overcoming tissue superposition effects. In addition, the radiomic analysis of screening images could allow better in vivo characterization to more accurately determine which findings require biopsy [42,43] or are indicative of aggressive disease. For example, Tabar et al. have created subclassifications of mammographic findings previously generically referred to as "DCIS" to distinguish acinar findings, which tend to be lower risk from the more ominous presentations accompanied by linear calcifications in the major lactiferous ducts, likely indicating that neoductgenesis is already occurring [44].

Harms associated with overdetection can also be mitigated by reducing subsequent overdiagnosis, through more definitive pathology tools for analysis and characterization of samples from needle biopsy, thereby reducing the amount of over (and, in some cases, under) treatment [35,37–39,45–48].

While it is tempting to reduce overdetection by limiting screening to those who are perceived to be at greatest risk of developing cancer, often referred to as "personalized screening" [49–53], this does not really target the problem as it is the aggressiveness of the cancer or lack thereof, not the likelihood of incidence that is the main driving factor for overdetection. Personalized screening is an excellent concept and research in this area could be of grea<sup>t</sup> value, but until we have risk prediction tools whose negative predictive value is extremely high, this would be a dangerous path to follow as it would result in the missed opportunity for the earlier detection and effective treatment of many potentially lethal cancers.

Finally, women, the general public and health care providers must be clearly, accurately and appropriately informed of the benefits provided by screening as well as the limitations or harms associated with phenomena such as overdetection. This has often been conducted poorly or inaccurately and without any weighting related to quality of life, ascribing inappropriately high harms to false positive screens and overdetection [3]. Others have suggested more useful and positive approaches to communication [54]. To allow informed decision making, and as an indicator of respect for human life, it is critically important that this is performed more effectively in the future.

**Author Contributions:** Conceptualization, M.J.Y.; methodology, J.G.M. and M.J.Y.; software, J.G.M.; validation, J.G.M. and M.J.Y.; formal analysis, J.G.M. and M.J.Y.; writing—original draft preparation, M.J.Y.; writing—review and editing, M.J.Y. and J.G.M. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Data Availability Statement:** All data supporting the results reported in this article can be found in the published work listed in the references. Input parameters and results produced from use of the OncoSim-Breast model can be obtained upon reasonable requests to the authors.

**Acknowledgments:** The authors are grateful to Claude Nadeau at Statistics Canada and Jean Yong at The Canadian Partnership Against Cancer for their valuable assistance. Jean Seely (University of Ottawa) reviewed early drafts of the manuscript and provided helpful suggestions for improvement. Eileen Rakovitch (Odette Cancer Centre) generously shared her insights on avoiding overtreatment of in situ breast cancer.

**Conflicts of Interest:** Yaffe holds shares in Volpara Health Technologies, a manufacturer of software for analyzing medical images and conducts some collaborative research in breast cancer imaging with GE Healthcare under an agreemen<sup>t</sup> with his institution. Neither organization was involved in any way with this project.
