Artificial intelligence to augment imaging could be pricey
January 31, 2024
I’ve seen a slew of articles suggesting that artificial intelligence (AI) could lower medical costs, often while improving quality of care. AI can interpret images, review laboratory results, identify worrisome clinical patterns, and optimize schedules. Alas, some early evidence suggests that the US health care delivery system will use AI to raise costs.
StatNews reported this fall that AI companies are promoting AI enhanced dental imaging that will dramatically increase detection of tiny imperfections in teeth, leading to dentists being able to justify substantially more dental work. The Food and Drug Administration (FDA) has approved 8 AI programs to enhance dental imaging so far, and there are more on the way. In many instances, the imperfections found by very sensitive AI systems would not have needed dental intervention for years, and sometimes these would never have needed intervention. One dental office in Oklahoma reported an increase of 30% in dental fillings in the first months of operation.
KFF Health News reported this month on new AI add-ons for mammography that are approved by the FDA, but for which there are no specific authorized billing codes. Therefore, radiologists only obtain higher reimbursement for using these AI tools if they bill patients directly. These AI systems could offer a payoff to the radiologists by decreasing time to read each mammogram and increasing accuracy. But I’m not surprised that many radiologists are implementing the technology to increase revenue, rather than to decrease resource cost.
One large academic study shows that AI augmented mammography increases cancer detection by 20%, which was statistically significant. The study also decreased radiologist screen time substantially. However, that study has not been going on long enough to determine whether these additional detections save or prolong lives. Some early cancers detected would never have caused harm, and some people with positive screenings would have died of other causes. A 2022 study suggested that about 15% of breast cancers detected by screening were “overdiagnosis,” which is more common in women over 65.
Implications for employers:
Don’t believe the hype that AI will necessarily lower health care costs. The incentives in US health care finance are to use these new tools to increase provider revenue, rather than to offer lower total medical costs.
Be cautious about providing coverage for new technologies that have not been demonstrated to improve quality. Most employers use health plans which have well-developed technology assessment programs.
Ask vendor partners what safeguards they have in place to mitigate any risks of this new technology, including the level of human oversight they provide over decisions made by AI.
Thanks for reading. You can find previous posts in the Employer Coverage archive
Please “like” and suggest this newsletter to friends and colleagues. Thanks!