Micromedex

The moat is the dataset at its strongest anywhere in this index, and the reason is legal rather than commercial. Micromedex is a decades old curated drug and toxicology corpus whose content is recognised as a compendium under United States federal law. That recognition attaches to the content and to the editorial process behind it, not to any software feature, and it is what a large share of customers are actually buying.

The artificial intelligence arrived in September 2025 as a search capability layered on that corpus. Remove it and the product is unchanged in its essential function, which is what it was for decades.

Graded C on mechanism, consistent with the other content organisations in this category, and the note says plainly that the grade describes what the artificial intelligence does rather than how good the product is. A buyer should read this axis alongside the regulatory one, where the same fact that holds centrality down is the source of the product's unusual standing.

The essentials are present and stated. Every result carries a citation, and the citation opens the underlying source in a single click, so verification costs one action rather than a search. The corpus is bounded to the company's own curated content, and a customer quoted by the vendor makes the point that the value is seeing artificial intelligence results come from trusted data with a click through to the source.

That is the baseline this category should meet, and it does. What is absent is anything beyond it. No statement describes what the system does when the content does not cover a question, when sources within the corpus disagree, or when a query falls outside drug and toxicology altogether. No abstention behaviour is described and no confidence signal is surfaced.

One framing on the vendor's own material deserves a gentle correction rather than a penalty, and it recurs across this category. The company describes a commitment to artificial intelligence transparency through features like suggested searches and follow up prompts. Those are usability affordances. They help a clinician ask a better question; they disclose nothing about how the system produced its answer. Naming them as transparency blurs a distinction this index treats as load bearing.

The corpus is well described, with curation from primary literature, in line referencing and daily updates all stated. The generative layer is not described at all. No model class, no model provider, no retrieval architecture, no ranking method and no version information were located in two retrieval passes.

One claim makes the gap sharper than a simple absence would. The company states that the artificial intelligence in the product is clinically validated. Nothing published describes what that validation consisted of, who performed it, against what reference standard, on how many queries or with what result. A validation claim without a method is weaker than no claim, because it asserts precisely the thing a buyer would otherwise know to ask about.

That is the single most answerable question on this record and it should be put directly: what was validated, by whom, and where can the result be read. The contrast within the category is available and unflattering. One competitor names its model family and platform provider outright, and another published a five dimension evaluation with a named panel and reported figures.

The corpus is well described and the generative layer is not described at all, which splits this record cleanly. On the corpus, curation from primary literature, in line referencing and daily updates are stated, so a buyer knows what the knowledge base is built from and how current it is, and for a reference product that is the substantive half of this axis.

On the generative layer, two retrieval passes located no model class, no model provider, no retrieval architecture, no ranking method and no version information, so the component that decides which of that curated content a clinician sees, and how it is summarised, is entirely opaque. One product line complicates the otherwise favourable architecture and is easy to miss.

The clinician facing reference product needs no patient record to function, but the payer offering supplies drug information for coverage determinations and pricing for claims adjudication, which sits close to claims and member data even where the reference content does not contain it, and the boundary between the two is undescribed.

Nothing states whether clinician queries into the search are retained or used to improve the system, and for a product operating in more than eighty countries the residency and query handling questions are the first a non United States buyer will need answered. Ask for the model provider, query retention, and the payer boundary.

Stronger than most of this category, on three independent artefacts, and the limits of each are stated.

A peer reviewed usability evaluation of the combined drug and disease product exists, published in 2023, with authors from Brigham and Women's Hospital, Mass General Brigham, Harvard Medical School and a college of pharmacy alongside authors from the vendor. It observed clinicians using the tool, collected structured feedback and identified areas needing refinement, including navigation difficulty and gaps in comprehensiveness. Peer reviewed study of an actual product in this category is rare, and a study that publishes what needed fixing is rarer.

Recognition as a compendium under federal law is a formal governmental designation reached through review of the evidentiary process behind the content, which is a different and in some ways higher bar than a customer satisfaction ranking. And the product was named number one for point of care drug reference in the 2026 Best in KLAS report for the second time, with customer feedback specifically citing reduced medication errors and standardisation.

Held at B rather than A for three reasons stated plainly. The peer reviewed study measured usability rather than correctness. It predates the current artificial intelligence search by two years and evaluated an earlier assistant. And the vendor co authored it. The Best in KLAS ranking measures customer experience, not output accuracy, and the compendium designation covers content rather than the model built over it.

The clinician facing reference product needs no patient record to function, which is the favourable architecture common to this category.

One product line complicates that and it is easy to miss. The payer offering supplies drug information for health plan coverage determinations and drug pricing for claims adjudication, reimbursement approvals and formulary optimisation. Work of that kind sits close to claims and member data even where the reference content itself does not contain it, and the boundary between the reference layer and the payer workflow is not described.

Nothing published states a retention period, a data residency position, an encryption practice, or whether clinician queries into the artificial intelligence search are retained or used to improve the system. The corporate privacy documentation was identified but not opened this pass, which is recorded explicitly rather than left implicit.

For a product operating in more than 80 countries, the query handling and residency questions are the ones a non United States buyer will need answered first.

No statement was located in two retrieval passes. No business associate agreement, no covered entity or business associate language, no execution path and no tier at which an agreement becomes available.

The architectural case for that being tolerable is reasonable on the clinician facing side, since a drug reference does not require a patient record. It is weaker on the payer side, where the product supports coverage determinations and claims adjudication and the customer is a health plan handling member data at scale.

Graded on what a buyer can find rather than as a finding that no instrument exists. A vendor selling into health systems and health plans in more than 80 countries has executed these agreements many times; none of the terms are published, and the corporate legal documentation was not opened this pass.

The pattern across this category is now consistent enough to state: the three largest content publishers here all assert or imply compliance and none publishes the instrument, while the smallest vendor graded publishes its agreement in full at a public address.

No certification was located in two retrieval passes: no SOC 2 report of either type, no ISO 27001, no HITRUST, no trust centre, no penetration testing statement and no vulnerability disclosure programme.

Customer feedback collected by an independent ranking organisation praises integration and interoperability, and one quoted customer specifically values that the artificial intelligence search required no integration work and no firewall changes. That speaks to deployment friction rather than to security posture, and the record should not let the two be confused.

Recorded as a retrieval outcome rather than as a finding that controls are absent. A company of this scale, formed from a major technology vendor's health division and selling into government departments and poison control centres as well as hospitals, will hold substantial security documentation. None of it was published against this product, and the specific ask is the report and its type.

No device clearance exists and none is needed. The regulatory status this product does hold is of a different kind and it is unique in this category.

Micromedex is recognised as a drug compendium under United States federal law. That designation means the content is used in determining medically accepted indications for coverage purposes, particularly for off label uses of anticancer drugs. The practical consequence is that what this reference says can decide whether a treatment is paid for. This index has recorded the pattern of clinical content becoming a reimbursement instrument in several forms; this is the strongest instance found, because the link is statutory rather than contractual or by convention.

Held at B rather than A, and the reason is the most important sentence on this record. The designation attaches to the compendium and to the editorial process behind it. It does not attach to the generative search layer built over that content in September 2025. A clinician reading a synthesised answer is not reading the compendium entry; they are reading a model's account of it, and the federal recognition says nothing about that transformation.

That is a fifth form of the certification non transfer problem this index already tracks, and it is the most consequential yet, because here the credential carries a payment consequence. Establish in writing whether a coverage determination may rest on an artificial intelligence generated summary or must rest on the underlying monograph.

No governance artefact was located: no published artificial intelligence principles, no responsible use framework, no external advisory body, no bias evaluation, no subgroup analysis, no error taxonomy and no published evaluation methodology.

One modest credit belongs on the record. A senior executive has published a set of questions organisations should ask before adopting artificial intelligence in healthcare. A vendor publishing buyer side diligence questions is a small but real contribution, and it invites exactly the scrutiny this record applies.

Against that sit two framings that should not pass unremarked. Describing suggested searches and follow up prompts as a commitment to artificial intelligence transparency treats interface design as disclosure. And asserting that the artificial intelligence is clinically validated, without publishing the validation, is a governance claim standing on nothing a third party can inspect.

The bias question here is specific and untouched. Drug evidence is unevenly distributed across populations, with well documented gaps in paediatric, pregnancy, renal impairment and older adult data. A search layer that synthesises across that corpus will answer with the same confidence whether the underlying evidence is dense or thin, and nothing published describes how sparse evidence is signalled.

The reference layer carries a real mechanism and the claim made about the artificial intelligence layer is the problem. On the mechanism: content is curated from primary literature with in line referencing, so a clinician receiving an answer can follow it to the source and judge it, which is the right control for a reference product and the reason this is not lower.

On the claim: the company states that the artificial intelligence in the product is clinically validated, and nothing published describes what that validation consisted of, who performed it, against what reference standard, on how many queries or with what result. A validation claim without a method is weaker than no claim at all, because it asserts precisely the thing a buyer would otherwise know to ask about and invites them to stop asking.

It is also the single most answerable question on this record, since a validation that happened has a design, a sample and an outcome that can be described in a paragraph. The contrast inside this category is available and unflattering: one competitor names its model family and platform provider outright, and another published a multi dimension evaluation with a named panel and reported figures. No warranty, indemnity or remediation commitment was located. Ask what was validated, by whom, against what standard, and where the result can be read.

Integration is stated by the vendor and corroborated by structured third party customer feedback, which is a stronger combination than most of this category offers. The company states integration with existing electronic health records, with pharmacy benefit management systems and with clinical processes, and customer feedback collected by an independent ranking organisation specifically praises integration and interoperability, with one customer describing a long standing and settled deployment.

Context retrieval support means reference content can be surfaced from within the record in the context of the patient being viewed, which is the standard built for this job rather than a bespoke build, and the same standard credited on the DynaMed record.

Held at B rather than A because no marketplace listing on an electronic health record vendor's own storefront was located on the company's material this pass, no conformance statement was published, and no implementation effort figure is given. One customer quote also notes that the artificial intelligence search itself required no integration, which is a convenience for adoption and a reminder that the generative layer and the integrated content layer are not necessarily reaching the clinician by the same route.

Hosted service delivered through web, a mobile application and integration into the electronic health record, in use across more than 80 countries and across an unusually wide institutional base including hospitals, poison control centres, government departments, universities, pharmaceutical organisations and health plans.

That geographic reach is worth reading against the pattern this category has produced. Three other vendors here publish explicit geographic limits on their generative layers, two of them excluding the European Union. Here the underlying reference is stated to reach more than 80 countries and nothing states whether the artificial intelligence search released in September 2025 is available everywhere the content is. Given how consistently that limitation appears elsewhere, silence should not be read as parity.

No hosting provider, cloud region, data residency statement or retention position was located, and no alternative deployment model is described for an institution requiring processing inside its own environment.

No pricing of any kind was located in two retrieval passes: no tier structure, no per seat or per institution rate, no basis, no implementation fee and no published position on whether the artificial intelligence search is included in an existing subscription or sold as an addition.

That last question is the one to force open early, and the pattern across this category makes it a reasonable worry rather than a hypothetical. Both of the other large reference publishers here sell their generative layer as a paid addition to a content subscription. A buyer already licensing this content should assume the same and get the answer in writing before renewal rather than after.

Every route on the vendor's material leads to a contact or demonstration request. There is no self serve tier, no trial route published for the clinician facing product and no rate card for any of the distinct buyer types the company names, which is notable given how varied that buyer list is: a poison control centre, a university and a health plan are not buying the same thing on the same terms.

Breadth within a defined domain, with the buyer types enumerated rather than implied, which is unusual and useful. Coverage spans drug information, interactions, intravenous administration and toxicology, and the stated user base runs across providers, poison control centres, government departments, universities, pharmaceutical organisations and health plans in more than 80 countries. Distinct product lines exist for the clinician and for the payer.

Toxicology deserves specific mention because it is genuinely differentiated. Poison control centres are a demanding user with no tolerance for a vague answer, and no other vendor in this category serves them.

Held at B rather than A because the domain is drug and toxicology rather than general clinical reference, so a buyer needing disease level guidance is directed to the combined product with a partner's content rather than to this one. And no performance evidence is published for any individual setting or user type. The peer reviewed evaluation that exists studied clinicians using the combined tool, not a poison control centre or a payer reviewer, so demonstrated usefulness covers a fraction of the stated footprint.