Medi-Span (Wolters Kluwer)

Artificial intelligence is claimed, and the learned component is not Medi-Span's. The product is named Medi-Span Expert AI, and Wolters Kluwer says it integrates advanced AI with evidence based drug data. What is described is a Model Context Protocol server that returns curated Medi-Span content, drug dosing guardrails and decision logic to other companies' agents and language models, alongside the same content delivered as licensed data and clinical screening APIs.

The screening itself runs on expert curated content: severity levels, dose ranges and documented interactions. No model trained or operated by Medi-Span is described in its published material, so the AI claim cannot be located in the product. That point concerns the mechanism and says nothing about the quality of the content, which is where the product's value sits. Wolters Kluwer says it expects to expand Expert AI into further workflows.

The oversight structure is described for the screening content and left open for agents. Interaction content carries a severity, a level of supporting documentation, management guidance and onset, and pregnancy, lactation, age and gender conflicts carry a quantified significance so a customer can choose which alerts to receive.

Wolters Kluwer also offers Command Center, which its own disclaimer describes as a way to tailor which Medi-Span Clinical alerts reach clinicians so they are not distracted or desensitized by unnecessary ones. In each case a clinician or pharmacist in the host system decides what to do. The missing part is the agent side.

The Model Context Protocol server promises clear visibility into the content an agent used, but nothing states what an agent may do with that content unattended, at what point a case should go to a person, or who is answerable when an agent misapplies content that was supplied correctly.

Medi-Span publishes how its content is made, and in close detail. Its Content and Editorial Process page names the primary sources (literature, product labeling, pharmacology studies and clinical trials) and a multidisciplinary team of nearly 150 advanced degreed clinicians.

It describes daily surveillance of industry activity and literature, weekly research on medication guides, REMS and United States drug shortages, review of monographs by practicing specialists, and a clinical quality control step that tests content inside the software against edge cases a clinician verifies. It also states that new FDA or Health Canada approvals trigger creation or revision of a monograph.

What is not published is any versioning, change history or correction rate that a customer could audit. The Model Context Protocol server's tools, the granularity of what it returns and the decision logic it applies are described only in outline.

The delivery routes are described, and the parties behind the hosted ones are not. Medi-Span reaches customers four ways: flat files that customers load into their own systems; clinical screening APIs that customers host locally; the same screening as web services Wolters Kluwer runs, whose documented inputs include a patient's age, weight, body surface area, renal function, conditions, pregnancy and lactation status and pharmacogenomic factors; and a Model Context Protocol server that other companies' agents query.

The first two involve no transfer to Wolters Kluwer at all. For the web services and the agent server, no hosting provider, region or subprocessor is named, and nothing says whether patient attributes sent for screening are logged or retained. There is no model layer of Medi-Span's own to disclose, since it supplies content to other companies' models, so the open questions are infrastructure and retention.

Named customers and vendor reported figures, with no published evaluation. Wolters Kluwer states that 98 percent of customers are satisfied with the quality of Medi-Span data, 99 percent are likely to renew their data license, and hundreds report productivity gains of 50 percent or more, without method, sample or date.

Named deployments include The Christie NHS Foundation Trust, announced in May 2026 as the first NHS implementation of Medi-Span, and a combined offering for eligible Epic customers from July 2026. The Expert AI page cites outside studies showing general purpose models failing basic medication tasks; those figures describe general models, not Medi-Span or an agent grounded on it. No study of the content's accuracy, of outcomes attributable to its alerts, or of agents using the Model Context Protocol server is published.

General assurances only. Wolters Kluwer's corporate AI principles commit to privacy and security in general terms, and nothing specific to Medi-Span says what happens to patient information. That matters for the routes Wolters Kluwer runs.

The screening web services take patient attributes as documented inputs, including age, weight, renal function, conditions, pregnancy and lactation status and pharmacogenomic factors, and the agent server answers queries in whatever context an agent sends. Nothing states whether those inputs are retained, for how long, whether they are deidentified, or whether they inform product development. Customers who use flat files or locally hosted APIs send nothing. No safety engineering is published: nothing describes how the agent server guards against returned content being misapplied downstream, or how a safety event is handled.

No business associate status and no privacy document reaching the product is published for Medi-Span. The Wolters Kluwer privacy and cookies notice covers the website, which does not reach the service. Customers who take Medi-Span as flat files or locally hosted APIs run it inside their own systems and send Wolters Kluwer no patient information, so the gap is narrower than it looks.

It applies to the web services and the Model Context Protocol server, which take patient attributes as inputs. For those routes no plain statement of scope is published, either that they are not intended to receive protected health information or that a business associate agreement is available for them. The sibling product UpToDate publishes the first kind of statement in its own terms; no equivalent is stated for Medi-Span.

No SOC 2 report, ISO 27001 certificate, trust center, penetration testing cadence or vulnerability disclosure route is published for Medi-Span. Wolters Kluwer publishes privacy and security certification pages for products in other divisions, naming ISO 27001 and SOC 2, and describes a group security program aligned to ISO 27001 and the NIST Cybersecurity Framework under executive oversight.

Alignment with a framework is not an attestation against it, and certificates scoped to other products do not cover this one. The gap has become more consequential with the agent server. A server that exposes tools to language models carries risks that content licensing did not, including prompt injection through returned content and exposure of the patient context an agent sends.

Medi-Span is drug data built into other companies' software and makes no device claim of its own, so the regulatory question sits with the systems that present it. The screening content is patient specific in places, such as dose range checks by age, weight and renal function, and it carries its basis with it: severity, level of supporting documentation and management guidance, so a clinician can review why an alert fired. The agent server moves the question.

Content that another company's agent summarizes into a recommendation may be harder to review than the entry it came from, and Wolters Kluwer publishes no regulatory position on agent use. The gap concerns agent use, not the content itself.

Policy language with no evaluation behind it. Wolters Kluwer publishes corporate AI principles covering privacy and security, transparency and explainability, governance and accountability, and fairness. Medi-Span's editorial page adds a concrete independence policy: it keeps long standing relationships with pharmaceutical companies for timely product information, publishes under proprietary editorial policies intended to keep content unbiased, and does not give pharmaceutical companies statistics on how end users use the product.

That addresses commercial bias in the content, not how the agent server behaves. No evaluation of agents grounded on Medi-Span, no coverage analysis and no subgroup result is published. The question specific to this product is thin evidence. Dose ranges are published for adult, pediatric and neonatal patients, where pediatric and neonatal evidence is sparse, and nothing says how that thinness is signaled to an agent that will state an answer with the same confidence either way.

Mechanisms exist that let a clinician catch a wrong output, and nothing stands behind the output. Alerts arrive inside the host system with severity and supporting documentation, a clinician or pharmacist decides what to do, and Command Center lets an organization tune which alerts fire. The agent server promises visibility into the content an agent used, which is traceability rather than recourse.

No warranty, indemnity, correction commitment or published error rate is offered for the content or for the server. The risk shifts with the consumer. A pharmacist reading an alert notices when it looks wrong; an agent applies what it is sent at machine speed.

Named systems with content delivered into them. Medi-Span is sold as embedded drug data for electronic prescribing and clinical screening, claims processing and drug utilization review, dispensing and formulary systems, and Wolters Kluwer says 7 of the 10 top grossing record vendors use it.

Since July 2026, eligible Epic customers, including rural hospitals, federally qualified health centers, state and tribal health organizations and small health systems, can take it inside Epic as part of a combined Wolters Kluwer offering with simplified implementation and testing. Outside the United States, The Christie NHS Foundation Trust and the Lebanese Hospital Geitaoui University Medical Center, through Guardia Systems and IQVIA EMR, are named implementations from 2026.

The clinical APIs take patient attributes from the host system and return screening results, and the agent server adds a standard interface for agents. No record vendor marketplace listing or integration documentation is published that would show the depth of any named integration.

Several delivery routes and no residency commitment. Flat files and locally hosted APIs run in customers' own systems, which places hosting and residency with the customer and settles most of the question for those routes by design. The web services, marketed to healthcare systems internationally, and the Model Context Protocol server are services Wolters Kluwer operates, and no hosting provider, region, residency commitment or availability commitment is published for either.

For an agent builder the availability question is practical. An agent whose grounding source is unreachable either fails visibly or falls back to its own training data, and which of those happens is decided by whoever builds the agent.

How the product is sold is discoverable; what it costs is not. Medi-Span is licensed as data that customers renew, and it is organized into named content sets: clinical screening, drug data, drug pricing data and patient education, plus the Price Rx pricing service, veterinary medicines and Expert AI for agent developers. Eligible Epic customers can buy it bundled with UpToDate Lexidrug and patient education content, with simplified procurement.

The modules are therefore clear before contact, as is the fact that the selection drives the cost. Missing is any figure or unit of charge, and whether an existing data license extends to agent access through the server or is a separate arrangement.

Coverage is named in detail. The buyer types listed are payers and pharmacy benefit managers, drug manufacturers and wholesalers, providers and record vendors, health technology companies, retail pharmacies and system integrators, and financial and consulting firms. Wolters Kluwer says the data reaches 17 of the 20 top grossing pharmacy benefit managers, 95 percent of the 20 top health insurers and 4 of the 5 top grossing retailers.

The workflows listed include electronic prescribing, drug utilization review, medication therapy management, formulary management, dispensing and analytics. Dose screening covers adult, pediatric and neonatal patients, and monograph work tracks FDA and Health Canada approvals. Implementations outside the United States were announced in 2026 at an NHS trust in England and a university hospital in Lebanon, and veterinary medicines were added in August 2026. No performance evidence is published for any setting, and much of the reach runs through other vendors' software, so what a given clinician sees depends on choices those vendors made.