AI Integration for McKesson EnterpriseRx Prescription Verification

The integration targets the prescription verification queue—the central workflow where pharmacists review new and refill prescriptions for safety and accuracy. AI connects via McKesson's EnterpriseRx API to access the prescription object (Rx), patient profile, and medication history. The AI agent acts as a pre-screen, analyzing each script in the queue for potential issues like drug-drug interactions, allergy conflicts, dosage appropriateness, and prior authorization (PA) requirements before the pharmacist's final review. This creates a two-step flow: AI flags and annotates, the pharmacist adjudicates.

Implementation involves an event-driven architecture. When a prescription enters the verification status, a webhook or API call triggers the AI service. The AI model, grounded in the pharmacy's specific patient data and formulary, returns structured recommendations (e.g., "Flag: Potential interaction between amiodarone and simvastatin. Suggested Action: Review for rhabdomyolysis risk."). These are injected back into the prescription record as custom notes or alert fields visible within the native EnterpriseRx verification screen. This keeps the pharmacist in the platform, maintaining audit trails and final approval authority.

Rollout is typically phased, starting with non-controlled substance, new-to-therapy prescriptions to build confidence. Governance is critical: all AI recommendations must be logged with a confidence score, and pharmacist overrides are captured to continuously refine the model. The impact is operational: reducing manual look-up time, catching complex interactions early, and ensuring PA flags are identified at the point of verification—turning a potential 48-hour delay into a same-day action.

The integration is triggered from the EnterpriseRx verification queue. When a new or modified prescription enters the queue, a secure webhook or API call is made from the McKesson platform to a dedicated AI service endpoint, passing key data objects: PatientProfile, MedicationHistory, CurrentRx, and AllergyList. The AI service, hosted in a compliant cloud environment, processes this payload using a clinical LLM grounded in drug databases and patient context to generate a risk assessment summary and recommendations.

The AI service returns a structured JSON payload containing fields like potential_interactions, dosage_appropriateness_flag, prior_auth_likelihood, and clinical_notes. This payload is injected back into the EnterpriseRx workflow via its API, typically populating a custom field group or an embedded iFrame within the pharmacist's verification screen. This allows the pharmacist to see AI-generated insights alongside the standard prescription data, maintaining their final approval authority. Audit trails are maintained by logging all API calls, input data hashes, and AI outputs to a separate governance database.

For rollout, we recommend a phased approach: start with non-controlled substance new prescriptions in a single location. Governance is critical; establish a weekly review board where pharmacists and the AI lead review flagged cases to calibrate model confidence thresholds and refine prompts. This architecture ensures the AI acts as a copilot, augmenting the pharmacist's review without creating a hard dependency or bypassing required human oversight.

Implementation begins by establishing a secure, read-only data feed from EnterpriseRx's API or database—pulling only the prescription data, patient age, and medication history required for the AI safety check. This data is never stored long-term and is processed within a private, HIPAA-compliant environment. The AI's output—a recommendation flag (e.g., POTENTIAL_INTERACTION, DOSAGE_CHECK, PRIOR_AUTH_FLAG) and supporting rationale—is injected back into the verification queue via a secure API call, appearing as a non-binding alert within the pharmacist's existing workflow screen. This ensures the AI acts as a copilot, not an autopilot, preserving the pharmacist's final approval authority.

A phased rollout is critical for clinical acceptance and risk management. We recommend starting with a silent pilot: the AI runs in the background on a subset of prescriptions (e.g., 10% of non-controlled substance new Rxs) and its recommendations are logged but not shown to pharmacists. This allows for performance benchmarking and tuning against the platform's own clinical decision support. Phase two introduces selective visibility, where alerts are shown only for high-confidence, low-risk recommendations (e.g., duplicate therapy checks). The final phase expands to full integration, with all alert types active and configurable by pharmacy management via a simple dashboard that controls sensitivity thresholds and can disable specific check types.

Governance is built into the workflow. Every AI-generated alert is logged in an immutable audit trail within EnterpriseRx (e.g., appended to the prescription notes or a dedicated audit table), recording the input data, model version, recommendation, and whether the pharmacist accepted or overrode it. This creates a feedback loop for continuous model improvement and provides clear documentation for compliance reviews. Role-based access ensures only authorized managers can adjust AI settings, and regular drift monitoring checks that the model's performance remains consistent as formularies and prescribing patterns change. This structured approach de-risks the integration, builds trust with clinical staff, and ensures the AI augments—rather than disrupts—the safe, efficient operation of the pharmacy.