In late September 2024, patients and family members at Adventist Health in Bakersfield, California, observed disturbing behavior from a nurse assigned to intensive care and a post-anesthesia recovery unit. The nurse walked barefoot on the unit, talked to herself, and acted abrasively, according to family reports cited in federal investigation materials. One patient described severe, unrelieved pain while under that nurse’s care and believed he had been given fentanyl and morphine even though the IV drip appeared ineffective.
Subsequent investigators from the Centers for Medicare and Medicaid Services (CMS), responding to a complaint, found that the nurse — who had been hired through a travel nursing agency a few weeks earlier — had been taking medications from a secured cabinet and documenting them as having been administered to patients.
Auditors found that the hospital used a machine learning system that monitored medication access and dispensing patterns to identify potential drug diversion. That software generated alerts indicating suspicious activity related to the nurse’s medication handling. The published account emphasizes that these types of systems can detect patterns humans might miss and can flag individual staff members for further review.
Despite the system’s alerts, hospital managers failed to act on them, according to the auditors cited in the report. The case underscores a central point in the STAT coverage: technology that detects risk is not sufficient unless human teams review and respond appropriately. The article describes the incident as one where institutional response — or lack of it — allowed diversion to continue.
Hospitals store and dispense controlled substances that are both critical for patient care and have high potential for abuse. The STAT report summarizes that employees occasionally steal these medications and may be under the influence while caring for patients. That dual reality — therapeutic need and addiction risk — makes hospital medication-management systems an important locus for diversion-prevention efforts.
CMS investigators were tasked with reviewing a complaint about the incident and concluded that the nurse had removed medications from a secured cabinet while documenting administration to patients. The article links family and patient complaints, investigator findings, and auditor reports about the ignored machine learning alerts. Additional operational details about the hospital’s internal processes, the exact content or timing of the software alerts, and subsequent disciplinary or legal outcomes were not reported in the publicly available portion of the article.
The STAT coverage frames the episode as an example of the promise and limitations of algorithmic surveillance. On the one hand, AI and machine learning tools can surface anomalous patterns that merit human investigation and can help connect individuals to treatment resources. On the other hand, such systems are not self-executing safeguards: they require human oversight, clear response protocols, and accountable follow-through to prevent harm.
The article highlights that an algorithm’s detection is only the first step. Effective prevention depends on operational factors — who receives alerts, how escalations are handled, whether staff are trained to interpret signals, and whether leadership acts on recommendations. In this instance, auditors found those human steps were not taken, allowing diversion to continue despite automated warnings.
The STAT report uses the Adventist Health episode to illustrate that technology can be a valuable component of diversion surveillance but cannot replace human responsibility and systems of response. It emphasizes the need for integrated approaches that combine analytic detection with timely human action.
Because the article excerpt available publicly is a STAT+ exclusive, some specifics — including full details of the machine learning tool’s alerts, exact auditing timelines, internal hospital communications, any employment or regulatory penalties, and follow-up patient-safety measures — were not reported in the accessible portion of the story.
Note: This write-up summarizes the facts reported in the STAT article excerpt. The original reporting was published as a STAT+ subscriber-exclusive piece; readers seeking the complete investigative detail and broader analysis referenced in the article would need access to the full STAT+ story.