Vigilant Algorithms and the Contagion of Risk in Automated Opioid Risk Assessment.

Amid the ongoing opioid crisis, U.S. primary care providers must balance the effectiveness of opioid analgesics with increasing regulatory pressures to mitigate addiction and overdose risks. Over the past two decades, growing concerns about opioid misuse have translated into tightened prescribing restrictions, often leaving patients with chronic pain conditions without viable alternatives (Knight et al. ; Brewer ). At stake is the broader question of who today is entitled to opioid pain relief—and under what conditions—when the accepted risks of opioid use disorder (OUD), overdose and drug diversion loom large in public and policy discourse. One major regulatory strategy has been the establishment of prescription drug monitoring programmes (PDMPs), which are state‐managed electronic databases that track controlled substance prescriptions across providers and pharmacies. In nearly all U.S. states and territories, PDMPs allow prescribers and pharmacists to review a patient's prescription history before making a prescription. Although originally designed as surveillance infrastructures, PDMPs have increasingly been extended through proprietary algorithmic tools that promise more actionable insights: risk scores and decision‐support metrics that flag data patterns associated with opioid misuse. The most widely implemented of these analytic extensions is NarxCare, a machine‐learning based risk scoring system developed in 2010 by an Ohio physician and acquired in 2014 by Bamboo Health (then Appriss Health). Since then, NarxCare has been integrated into 52 of 54 state and territorial PDMPs, purportedly informing over a billion encounters annually (Bamboo Health , ). Its rapid adoption reflects state and federal funding (Chiarello ) and lobbying support from the pharmaceutical industry (Kabella et al. ), cementing its dominance alongside only a few alternatives, such as the Veterans Health Administration's STORM tool (see E. M. Oliva et al. ).