A coalition of lawmakers is attempting to force a vote to end a Medicare prior authorization pilot, warning that the artificial intelligence program’s tendency to automatically reject all requests for medical treatment has not been subjected to a rigorous double-blind study.
The Centers for Medicare & Medicaid Services (CMS) experiment, which allows proprietary algorithms to determine if seniors require post-acute care, is currently facing pushback from House Democrats. While observational data suggests the AI rejects nearly all physician requests for MRI scans, cancer screenings, and basic life support, experts emphasize the need to control for confounding variables before definitively stating the software is harmful.
It is certainly tempting to look at a machine that just denied 14,000 consecutive requests for oxygen therapy and assume the worst, but we must remember that correlation does not equal causation. We simply do not have the sample size yet to prove these patients wouldn't have been denied care by a human.
Lawmakers pointed to a recent retrospective cohort study published in the Journal of the American Medical Association (JAMA), which found a statistically significant association between the deployment of the AI and seniors suddenly expiring in hospital waiting rooms. However, the study’s authors cautioned against prematurely halting the trial, noting that the patients' pre-existing condition of being elderly may have contributed to their mortality independently of the algorithmic intervention.
UnitedHealthcare, which provided the underlying training data for the pilot—and disclosed a minor financial conflict of interest regarding its recent $3.2 billion in quarterly profits linked directly to unpaid claims—stressed that ending the experiment now would introduce severe selection bias into the dataset. The insurer noted that the AI’s decision-making methodology has yet to be fully peer-reviewed, meaning its habit of refusing all medical care could technically still be classified as a placebo effect.
For now, researchers advise Medicare beneficiaries to consult with their primary care physicians before assuming their algorithmically rejected claims will result in death, as long-term efficacy data on whether seniors actually need medicine remains largely preliminary.