A Modeling Study on Overdiagnosis

By the Cancer Rose Collective, March 12, 2022


According to a modeling study based on data from Breast Cancer Surveillance Consortium (BCSC), about one in six to seven screened breast cancer cases is overdiagnosed.

This study first highlights that overdiagnosis in breast cancer screening is real.

Results of study

An average of 15.4% (95% CI: 9.4%-26.5%) of screened cancer cases were estimated to be overdiagnosed, reports lead author Marc D. Ryser* of Duke University in Durham, North Carolina, and colleagues.

* Ryser: Marc Daniel Ryser, Assistant Professor of Population Health Sciences. Dr. Marc Ryser is an expert in mathematical and statistical modeling. His research uses biological, clinical, and population-level data to inform and guide the early detection and prevention of cancer.

Below are the results by age group and type of cancer detected (Figure 3 and Table 3).

Beyond the average values, we can observe (Fig 3) that for all age groups, the rate of overdiagnosis can reach maximum values higher than 20%, and according to Table 3, the rate of overdiagnosis at the first screening reaches a maximum value of 28%, at 58 years 21.1%, at 66 years 25.4% and at 74 years 31.9%

In this model study, an interesting finding is that the rate of overdiagnosis increased with age and almost doubled depending on the age range analyzed: 11.5% (95% CI, 3.8%-28.3%) at the first screening at age 50 to 23.6% (95% CI, 17.7%-31.9%) at the last test at age 74.

Comparison with previous data

The authors note, "comparison of our estimates against those from other studies is not straightforward because of differences in overdiagnosis definitions and screening practices."

They conclude that their results regarding overdiagnosis are both higher than previous modeling studies (ranging from 1% to 12%, depending on the studies cited in the article) because of differences in screening practices, diagnostic practices, and modeling assumptions, but lower than other studies that have shown rates much higher than the average in this study.
For example, the Canadian screening trial estimated an overdiagnosis rate of 30% (Baines CJ, To T, Miller AB. Revised estimates of overdiagnosis from the Canadian National Breast Screening Study. Prev Med. 2016;90:66-71. [PMID: 27374944] doi:10.1016/j.ypmed.2016.06.033 ) for cancers detected by screening.
In a population-based study, Bleyer and Welch (Bleyer A, Welch HG. Effect of three decades of screening mammography on breast-cancer incidence. N Engl J Med. 2012;367:1998- 2005. [PMID: 23171096] doi:10.1056/NEJMoa1206809 ) estimated that 31% of all diagnosed breast cancer cases were overdiagnosed.

The authors conclude with the hope that their findings will join a consensus and facilitate decision-making regarding mammography screening.

Conclusions of the Editorial "Reducing the Burden of Overdiagnosis in Breast Cancer Screening and Beyond

The editorial published in conjunction with the study emphasizes the importance of informing women about what this overdiagnosis represents.
(Marcondes FO, Armstrong K. Reducing the Burden of Overdiagnosis in Breast Cancer Screening and Beyond. Ann Intern Med. 2022 March 1. doi: 10.7326/M22-0483. Epub ahead of print. PMID: 35226534.)

Authors underline: « Women who are considering having mammography screening should be counseled about the risk for unnecessary cancer treatment using this information."
Estimating that about 60% of the 280,000 cases of breast cancer diagnosed each year in the United States are discovered through mammography screening, eliminating overdiagnosis could save 25,000 women the cost and complications of unnecessary treatment.

"Substantial advances" in critical areas need to be made, according to the authors, including:
- Develop a better predictive capability to accurately identify tumors that will not progress
- Improve the accuracy of screening technologies to reduce the risk of overdiagnosis and improve the ability to detect breast cancer that has not been detected by mammography
-Implement prevention strategies to reduce the overall rate of breast cancer diagnosis, such as providing counseling on lifestyle changes and screening for genetic risk.

The authors of the editorial conclude: « Screening tests, whether for cancer or other conditions, can provide great benefit by detecting disease when it is more easily treatable. However, the risk of labeling millions of persons as having a disease without improving their outcomes is very real. For now, the key to navigating these tradeoffs remains open and effective physician–patient communication, rigorous evaluation of all proposed screening strategies, and continued investment in early detection research. We look forward to the day when making an early diagnosis always helps our patients achieve better outcomes. »

And the findings go beyond breast cancer screening.
"As screening and diagnostic testing continues to grow in clinical practice, the issue of overdiagnosis is being felt far beyond cancer screening. For some conditions, changing definitions have led patients to be labeled with a predisease state on the basis of a test result that was previously considered in the normal range. Although there are strong arguments in favor of early treatment to prevent long-term complications in many conditions, the reality is that, just as with cancer screening, there is little doubt that some patients diagnosed through a screening test would never have progressed and are likely to be receiving unnecessary treatment."

Comments and criticisms, opinion of Dr. V.Robert, statistician

1-A modeling study

The study remains a modeling study, which means that the results of a model depend on a chosen model and conditions of validity, at best unverifiable and at worst questionable. For example, the authors are obliged to consider that a breast cancer is either definitively non-evolving or inexorably evolving, with no possibility that the evolutionary status of cancer changes over time. It is not clear that things are that simple.

Another example is that the authors are obliged to build their model by considering that all progressive breast cancers evolve at the same rate and that this rate remains constant throughout the evolutionary period. In practice, there are most likely different distributions of progression rates for each type of breast cancer, and it is not clear that progression rates cannot vary over time.

2- The data on mortality from causes other than breast cancer used by the study do not seem well adapted.

On the one hand, after checking reference 25 of the study, which corresponds to the source of these data (Contribution of Breast Cancer to Overall Mortality for US Women): for a population of women aged 50 to 80 years, these data are not derived from direct measurements of mortality but from data estimated from projections (in other words, from models).

On the other hand, the data correspond to a cohort of women born in 1971. Since the median age of the women included in the study is 56 years, the cohort born in 1971 is adapted for the mortality of women included in 1971 + 56 = 2027. Or, if you prefer, the cohort adapted to have the mortality of women aged 56 in 2000-2018 should be born between 1944 and 1962. Whatever the reasoning, it is clear that the cohort considered to obtain the mortality data is too recent by at least a decade. This is not neutral since the tables in Reference 25 show a non-negligible decline in mortality over time.

3-The definition of screen-detected cancers is questionable.

Screen-detected cancers are considered to be those that meet the following two conditions: screening mammograms BI-RADS 3 to 5 + diagnosis of cancer within the next 12 months.
With criteria such as these, interval cancers are likely to be classified as screen-detected cancers (BI-RADS 3 + negative complementary examinations = screening showing no cancer; if a cancer occurs 11 months after the screening mammogram, it is an interval cancer, and yet it will be classified as a screened cancer). Even if these cases are not very frequent, they are part of the data used to adjust the parameters, and adjusting on "garbage in, garbage out" data can only give garbage results.

4- It is wrong to pretend that the study found that the overdiagnosis rate is 15%.

The reality is that the study shows that the overdiagnosis rate is somewhere between 9% and 27% (and any value within that range is possible, no more 15% than 9% or 27%).

Figure 3 from the study:

Depending on the age range, the percentage of overdiagnosis can vary up to 25% or even 32%.

Unfortunately, it is a very common mistake to take the result of a study (estimated rate from a sample) for the reality (real and unknown, rate in the population). And it is an even more common mistake to believe that the estimated rate is more likely to be close to the real rate than any other value in the confidence interval.


Based on such a study, we cannot arbitrarily consider that the debate on the frequency of overdiagnosis is closed, with a definitive frequency of 15%, as the study's authors would like.

On the other hand, the model may be interesting for answering questions about the evolution of the frequency of overdiagnosis as a function of the age of the women screened or about the evolution of the frequency of overdiagnosis as a function of the interval between two screening mammograms.    

Professor Alexandra Barrat of the University of Sydney, interviewed by Amanda Sheppeard, associate editor and reporter for Oncology Republic and The Medical Republic, said there are different methods for estimating the potential rate of overdiagnosis through breast cancer screening.
She said the study demonstrated the inevitability of overdiagnosis in screening for a number of cancers. "I think there is a need in the professional community for greater acceptance of what the evidence shows about breast cancer screening."
"We just need to recognize that this is inherent in a cancer screening program."

As a result, and beyond the evidence, according to A.Barrat, the study helped to underscore the importance of informed consent in breast cancer screening.

Conflicts of interests

Dr Ruth Etzioni - Individual investor and stock in Seno Medical  https://senomedical.com/clinical/product-applications

Seno’s first clinical product targets the diagnosis of breast cancer and will be used in addition to screening mammography, integrating opto-acoustics and ultrasound.

Cancer Rose est un collectif de professionnels de la santé, rassemblés en association. Cancer Rose fonctionne sans publicité, sans conflit d’intérêt, sans subvention. Merci de soutenir notre action sur HelloAsso.

Cancer Rose is a French non-profit organization of health care professionals. Cancer Rose performs its activity without advertising, conflict of interest, subsidies. Thank you to support our activity on HelloAsso.