Build, Test, And Scale Accountable Pharmacy Artificial Intelligence

Predictive PerspectivesPredictive Guidelines & Standards

Authors:Angus, Derek C; Khera, Rohan; Lieu, Tracy, et al.

Journal:Jama

DOI:10.1001/jama.2025.18490

Directs AI to become accountable, evidence-driven health technology by recommending total-product-lifecycle governance, standardized evaluation/monitoring (algorithmovigilance), a national federated data infrastructure, and aligned incentives. This roadmap helps pharmacy informatics teams, pharmacy leaders, and clinicians deploy and monitor EHR‑CDS, AI scribes, DTC apps, and operational tools safely—shifting workflows toward validated, continuously assessed automation.

October 13, 2025

Quick Take

JAMA Summit calls for total product lifecycle governance, standardized algorithmovigilance, and a federated national data infrastructure to enable continuous evaluation and monitoring of artificial intelligence (AI) tools

note >1,200 US Food and Drug Administration (FDA)–cleared AI medical devices (majority in imaging).

Pharmacy leaders and informatics teams should operationalize algorithmovigilance for Electronic Health Record–embedded clinical decision support (EHR-CDS), AI scribes, direct-to-consumer (DTC) medication apps, and operational AI tools to enable validated, continuously monitored automation that preserves safety, optimizes workflow, and delivers measurable outcomes.

Why it Matters

Hospital pharmacies are seeing a rapid increase in EHR-embedded and operational AI (eg, sepsis alerts, AI scribes, prior-authorization tools) whose clinical effects are often unquantified

because performance depends on the human–computer interface, training, and deployment context, pharmacists face uncertainty about the reliability of medication decision support and reconciliation.

Rigorous evaluation and continuous monitoring are resource intensive, and many health systems lack the data infrastructure or local expertise to validate models in real-world settings, leaving pharmacy teams to handle alert triage, mitigate automation bias, and manage workflow and safety risks.

Absent a federated data infrastructure, standardized algorithmovigilance, and aligned incentives, pharmacy informatics cannot scale evidence-driven CDS tuning or stewardship—exacerbating constraints on safety, workflow optimization, and medication stewardship.

What They Did

Convened a multidisciplinary JAMA Summit including clinicians, informaticists, regulators, industry developers, payers, and patient representatives to assess AI in health care.

Synthesized targeted literature, regulatory policies, reporting standards, and real-world deployment examples across clinical, direct-to-consumer, business-operations, and hybrid AI tools relevant to care delivery and EHR workflows.

Reviewed existing evaluation approaches (FDA and Office of the National Coordinator frameworks, trial designs, observational methods) and compared practical monitoring and testing strategies for EHR-embedded clinical decision support and operational tools.

Produced a consensus, non-systematic expert synthesis summarizing summit discussions and gaps to inform lifecycle governance and evaluation priorities.

What They Found

>1,200 FDA-cleared AI devices (majority in imaging) and ~90% adoption of some imaging AI in US systems, yet many EHR-embedded clinical decision support tools lack robust evidence—creating substantial verification demands for pharmacy teams.

Market scale: >350,000 mobile health apps and approximately 30% of adults have used one

many DTC tools avoid FDA oversight, increasing pharmacists’ workload for medication reconciliation and safety tracking.

Rigorous evaluations are scarce—few randomized trials

the Apple Heart Study required >400,000 participants to validate a smartwatch atrial fibrillation tool, highlighting evaluation scale and cost that exceed most health systems’ capacities.

Summit prioritized four levers—total product lifecycle governance, standardized algorithmovigilance, a federated national data infrastructure, and aligned incentives (eg, the Health Information Technology for Economic and Clinical Health [HITECH] Act’s $35 billion that supported >97% EHR adoption). Cross‑stakeholder lifecycle oversight plus federated data were identified as the drivers of systemic improvement.

Takeaways

Operationalize algorithmovigilance within pharmacy informatics: require vendor model cards and labeled performance, surface model outputs in the Electronic Health Record (eg, Epic) and monitoring dashboards, and define pharmacist review triggers and de‑adoption workflows.

Fund staffing and training: allocate pharmacist and informatics time, provide AI literacy and change‑management education, and establish a lifecycle governance committee that includes pharmacy, IT, quality, legal, and patient representation.

Treat AI like a medicine—require predeployment evidence and continuous postmarket algorithmovigilance where pharmacists act as stewards: triaging alerts, verifying clinical context, and preserving human judgment.

Join federated evaluation efforts (eg, Coalition for Health AI/Joint Commission sandboxes or Sentinel‑style networks), demand vendor transparency, and align incentives so monitoring is funded and sustained.

Strengths and Limitations

Strengths:

Multidisciplinary Summit produced a consensus roadmap synthesizing literature, policy, and deployment examples to prioritize lifecycle governance, algorithmovigilance, federated data, and aligned incentives.

Recommendations align proposed levers with existing reporting and monitoring tools and offer pragmatic steps health systems and pharmacy informatics can use to operationalize stewardship.

Limitations:

This is a consensus, non‑systematic synthesis grounded in expert opinion and lacks multi‑site causal evaluations or randomized evidence to validate the proposed interventions.

Feasibility of implementing a federated data infrastructure, securing sustained funding, and absorbing operational burdens remains unresolved and may limit generalizability.

Bottom Line

This consensus roadmap provides pharmacy leaders a practical, actionable blueprint to pilot total product lifecycle governance and algorithmovigilance for EHR‑embedded AI, contingent on institutional commitment to staffing, data infrastructure, and sustained monitoring.