Plain Language Summary

What is it about? Breast cancer is a leading cause of death among women in the United States. While current screening methods help reduce breast cancer mortality, they are not perfect. Artificial intelligence (AI) can analyze mammography images and may help doctors find breast cancer more accurately. This study looks at how using AI to read mammograms might impact breast cancer outcomes for women over the long term. The study also evaluates whether using AI to help with breast cancer screening is cost-effective.

Overall, the study finds that while using AI to assist doctors in reading mammograms may improve diagnostic accuracy, AI is not cost-effective. This is because AI finds small precancerous breast lesions, many of which will never be harmful but are costly to treat. This research significantly contributes to understanding the balance between improved screening outcomes and the associated costs.

How was the research conducted? The research is based on a microsimulation model, which estimates the outcomes for a large, simulated group of individuals over time. This analysis used data from national cancer databases to simulate standard mammography for women aged 40 to 74. Researchers then compared the outcomes of traditional digital breast tomosynthesis, a kind of mammography,  with those enhanced by AI. By doing so, they showed how AI might contribute to false positives, false negatives, cancer stage at diagnosis, and breast cancer deaths. The chosen method allowed researchers to assess long-term outcomes and cost-effectiveness, which are difficult to measure in short-term studies.

What were the results? The study found that adding AI to breast cancer screening reduced false negatives by 2.1 cases and false positives by 49 cases per 1000 women screened. This led to a decrease in advanced breast cancer cases and a reduction of 0.13 breast cancer deaths per 1000 women. However, the use of AI increased the lifetime screening costs significantly, resulting in an incremental cost-effectiveness ratio of $303,279 per quality-adjusted life year gained. This is above the standard willingness-to-pay threshold, indicating it is not cost-effective at the current price. Importantly, the study reported a 21% increase in detection of noninvasive cancer cases, which will increase the cost of screening but will not result in health benefits.

Why are the results important? For health technology assessment agencies, these results highlight the need to consider both the clinical benefits and the economic impact of AI in screenings. In practice, while AI could slightly improve screening accuracy, its high costs and potential to identify lesions that are not clinically important may limit widespread adoption. Long term, these findings could prompt additional development of AI models to improve the ability to detect the highest-risk breast cancers or an adjustment of pricing models to make them more cost-effective.

What are the strengths and weaknesses of this study? A main strength of the study is its comprehensive modeling approach, which provides valuable insights into long-term outcomes and costs of AI-assisted breast cancer screening. One important limitation is that the study uses data from controlled studies of AI, which may not apply to all real-world settings. Future research could focus on real-world studies to validate these findings and analyze specific patient populations where AI may be most cost-effective in breast cancer screening.

^{Note: This content was created with assistance from artificial intelligence (AI) and has been reviewed and edited by ISPOR staff. For more information or for inquiries on ISPOR’s AI policy, click here or contact us at info@ispor.org.}