Abstract
Causal studies of algorithm-assisted prior authorization in Medicare Advantage find savings without detectable harm, while litigation, Senate, and inspector general records describe warranted care denied to patients least able to contest it. Both can hold for one technology under different contracts. In a model where reviewers keep a share of averted spending and predict which denials get appealed, two contract terms decide whether prediction screens or targets: the captured share and whether appeals or independent audits trigger oversight. Four implications, three preregistered, are read from public data on the largest insurer's 2020–2021 integration of a review vendor. Denial growth concentrated in post-acute care, and the integrated reviewer paired the highest denial rate with a higher overturn rate than internal review. These patterns are inconsistent with necessity-only tightening but do not identify the mechanism. The model favors decoupling reviewer pay from savings and auditing denials independently of appeals.

![Author ORCID: We display the ORCID iD icon alongside authors names on our website to acknowledge that the ORCiD has been authenticated when entered by the user. To view the users ORCiD record click the icon. [opens in a new tab]](https://preprints.apsanet.org/engage/assets/public/apsa/logo/orcid.png)