2025 SAIL Review - Clinical AI’s Mixed Reality

• AI-enabled electrocardiogram (ECG) alerts were associated with a 17% relative reduction in 90-day mortality when clinicians acted on flags (3.6% vs. 4.3%; HR, 0.83).

• Pharmacy must treat AI-generated documentation as untrusted input—verify medication histories, plan for increased verification workload, and be prepared to contest AI-driven payer denials that delay discharge.

• Ambient generative-AI documentation has achieved near-universal adoption across health systems and is now a primary upstream input to medication reconciliation, but documented hallucinations (fabricated clinical content) threaten medication safety and impose a measurable verification burden on pharmacy teams.

• Health plans are increasingly using AI to flag early disease and to auto-deny claims, which can delay discharge, create financial toxicity for patients, and increase pharmacy time spent on coverage, prior authorizations, and appeals.

• Real-world evidence is mixed: some AI clinical decision support (AI-CDS) tools show substantial clinical benefit while others show neutral or no improvement. Pharmacy engagement in tool evaluation, lifecycle stewardship, and clinical decision support governance is essential to protect safety and prioritize informatics resources.

• Produced the SAIL 2025 Year in Review: a committee-curated perspective synthesizing literature, industry reports, and news from 2024–2025; the symposium convened roughly 150 stakeholders.

• Synthesized evidence across six priority domains—documentation, clinical decision support, payer practices, clinician–AI interaction, economics, and regulation—drawing on recent trials, economic analyses, and legal reports.

• Anchored conclusions in cited primary studies and policy sources rather than new patient-level data, incorporating multidisciplinary viewpoints from clinicians, informaticists, payers, regulators, and patients.

• Framed the document as a perspective/synthesis focused on practical implementation, oversight, and monitoring considerations for health systems rather than original research.

• AI ECG alerts reduced 90-day mortality by 17% (3.6% vs. 4.3%; hazard ratio, 0.83); alerts prompted confirmatory testing and escalation of care.

• Ambient AI documentation was pursued across 43 responding health systems; one multicenter cohort reported clinician burnout fell from 77% to 51% after AI scribe implementation, while a large longitudinal study found no significant reduction in documentation time or productivity (Net Promoter Score ≈ 0).

• AI clinical decision support produced mixed results: a wearable sensor reduced 90-day rehospitalizations by 38%; EchoNext doubled positive predictive value for structural heart disease detection though half of high-risk patients still lacked echocardiography; an intracranial-hemorrhage triage trial showed no improvement in sensitivity, specificity, or turnaround time.

• Operational and safety signals included fabricated clinical content from AI scribes and increasing payer use of AI to auto-deny claims; approximately 70% of payers and providers were pursuing or implementing generative AI projects, creating an immediate medication-verification workload for pharmacy teams.

• Ambient AI notes are now a major upstream input to medication reconciliation, but they can contain fabricated clinical content—pharmacists must treat documentation as a starting point that requires corroboration before influencing orders.

• Given heterogeneous trial outcomes, prioritize adoption of AI tools that demonstrably change clinical management and ensure implementation includes resourced intervention workflows; accuracy alone does not guarantee improved real-world outcomes.

• As payers deploy AI for coverage decisions, medication access and discharge timing may become less predictable; pharmacy should monitor downstream safety, equity, and financial impacts and engage in payer and policy discussions.

Strengths:

• The Year-in-Review draws on about 150 stakeholders plus curated literature, reports, and news to reflect real-world clinical AI deployment and operational experience.

• The perspective anchors discussion in randomized trials, observational cohorts, economic analyses, and regulatory reports, presenting both impactful and null AI implementations.

Limitations:

• As a perspective, it synthesizes existing work without new patient-level data or pooled meta-analysis, so reported effect sizes and return-on-investment estimates are approximate.

• Source selection was curated by a committee with limited detail on search methods or formal bias assessment, which could influence representativeness and emphasis.

AI has produced selective, demonstrable clinical benefits while introducing new medication-safety risks and verification burdens; pharmacy should actively monitor deployments, require vendor transparency and local validation, budget for life‑cycle monitoring, and treat AI-generated documentation as untrusted until independently verified.