EHR-Embedded AI Outperforms Pharmacokinetics, Triages Antiepileptic Reviews

Predictive Model DevelopmentPredictive External Validation
Journal: Clinical and Translational Science

An EHR-integrated ensemble AI predicts AED levels more accurately than population PK for 3 of 4 drugs (e.g., phenytoin RMSE 4.15 vs 16.12 μg/mL), enabling smarter TDM triage. Pharmacists can prioritize CBZ/PHE/VPA reviews and targeted dose adjustments using interpretable drivers (time since last dose, dose, weight), while retaining PK support for phenobarbital.

EHR Logs Quantify Post-Alert CDS Time Tradeoffs

Predictive Model DevelopmentPredictive Process Outcomes
Journal: Jamia Open

Vendor-agnostic topic modeling of EHR audit logs turns “CDS exhaust” into a Fitbit-like monitor, flagging post-alert workflow shifts within the 1-hour window (+1.3 to +2.2 minutes handling CDS, ~−2.9 minutes documentation). Pharmacy informatics and leaders gain scalable visibility to prioritize alert tuning or de-implementation, protect pharmacist verification throughput, and steer CDS governance without laborious surveys or opaque vendor metrics.

Pharmacists Use EHR Risk Flags to Prioritize Paxlovid

Predictive Model DevelopmentPredictive Real-World Deployment
Journal: Prevention Science : the Official Journal of the Society for Prevention Research

An EHR-embedded risk model (19 factors), deployed at Kaiser Permanente Washington, flags the top 10% of COVID-positive adults who account for ~48% of 14-day hospitalizations/deaths (AUC 0.825), letting pharmacists prioritize Paxlovid counseling and complex interaction checks within five days. Tunable thresholds and equity monitoring align outreach with staffing, strengthening stewardship and reducing missed high-risk patients.

Biomarkers plus urine output flag ICU AKI, dialysis risk

Predictive Model Development
Journal: Intensive Care Medicine Experimental

Explainable XGBoost flags ICU patients at admission who are likely to develop AKI or need RRT, modestly outperforming logistic regression (AUC 0.76/0.92 vs 0.74/0.90) using urine output plus endostatin/NGAL. This helps ICU pharmacists triage worklists for renal-dose adjustments and nephrotoxin stewardship, turning early signals into targeted interventions and streamlined EHR workflows.

Model Passports Standardize AI Lineage, Simplify Pharmacy Audits

Predictive Guidelines & StandardsPredictive Real-World Deployment
Journal: Computational and Structural Biotechnology Journal

AIPassport brings verifiable, EHDS/HealthDCAT-AP-aligned model passports to hospital AI/CDS, automating lineage and provenance across MLflow, DVC, Git, and MINIO so pharmacists spend less time documenting and more time reviewing risk. Proven at scale (>14,300 cases), it streamlines governance, cross-site audits, and reproducibility for on-prem deployments while keeping PHI out of artifacts.