Doximity launched DoseCheck on 29 October 2026, bringing a specific medicines-safety question into focus: after an AI has generated a dose, what checks that the claim matches an authoritative record?
For UK clinicians, that announcement also helps explain the value of iatroX’s partnership with Datapharm, which brings regulator-approved medicines information from emc directly into the clinical AI workflow. Both developments move attention towards the information and safeguards around a model. They address different parts of the process.
Understanding that difference is useful when choosing a reference tool, checking an unfamiliar medicine or teaching prescribing reasoning. “Which source supports this?” and “Was this exact dosing claim verified?” deserve separate answers.
What DoseCheck adds to Doximity Ask
Doximity’s launch explanation describes a deterministic verification layer. It extracts dosing claims, including the medicine, amount, route, frequency and relevant indication, and checks them against a structured database. That database draws on sources including FDA labelling; verified doses link to supporting details.
Doximity released its own model-testing findings alongside the launch. The companion explainer examines those results and the limits of the reported evaluation.
The important architectural idea is that generating a sentence and checking a structured claim are different operations. A fluent explanation can contain a misplaced decimal, an unsuitable regimen or an omitted qualification. A separate check can test whether a recognised claim agrees with the reference rules it covers.
That still leaves questions about coverage: which products, populations and complex regimens are represented, and what happens when the system cannot verify a claim? Our DoseCheck launch explainer examines the announcement and its reported findings in more detail.
What the Datapharm partnership brings to iatroX
Datapharm’s own announcement confirms that healthcare professionals can access emc medicines information within iatroX’s AI workflow, including current Summaries of Product Characteristics, or SmPCs. The partnership connects the answer process to maintained product information.
An SmPC describes a particular medicinal product: its authorised uses, administration, contraindications, warnings, interactions and other characteristics. emc explains how companies publish approved information and maintain it as changes are authorised. That gives an answer a product-specific reference point. A reader can locate the licensed indication, examine the conditions attached to administration and trace an interaction statement to the underlying record.
We announced the partnership in May. DoseCheck gives that existing relationship a fresh context: medicines AI needs a dependable connection to the underlying product record, alongside appropriate checks on what it says.
For iatroX, this foundation supports questions extending beyond a numerical dose, including formulation differences, interactions, pregnancy information and product-specific precautions.
The shared principle, and the real difference
The shared principle is that an AI’s remembered account of a medicine is an insufficient foundation for a consequential answer. Relevant, maintained source material needs to participate in the workflow.
The distinction is where the documented intervention occurs.
Datapharm provides iatroX with an authoritative medicines-information foundation. DoseCheck is described as checking generated dosing claims against structured rules. Those documented roles distinguish source grounding from a separate deterministic check of a generated claim.
iatroX’s published methodology describes retrieval, source-grounded synthesis, output checks and uncertainty handling. A useful way to work with that process is to state the UK product and clinical purpose explicitly, examine the answer’s source links, and read the relevant product-information section when a condition or exception determines the interpretation.
A useful comparison therefore separates three things: the authority of the source, the method used to construct or check an answer, and the clinical decision being made. For a particular question, identify which of those tasks you need help with. That makes the choice of workflow more precise and the resulting answer easier to assess.
Three UK examples that expose the questions
These examples are exercises in reading the source and identifying missing context. They do not provide treatment regimens.
Apixaban: the indication changes the question
A question such as “Is this apixaban dose correct?” leaves out crucial information. The UK product information distinguishes stroke prevention in non-valvular atrial fibrillation, treatment of venous thromboembolism, prevention of recurrence and thromboprophylaxis after certain operations. Its conditions also vary across these uses.
The first task is to identify the clinical purpose and treatment stage, then locate the corresponding product-information section. Matching a familiar number to a medicine name is insufficient.
Renal assessment adds another source question. The MHRA’s DOAC safety advice explains the relevance of calculated creatinine clearance and links to updated renal guidance. A useful learning prompt is: “Which patient details and renal-function measure would I need before evaluating this claim?”
That tests whether the reasoning preserved the conditions attached to the source.
Nitrofurantoin: the formulation must survive the summary
emc lists both macrocrystalline capsules and prolonged-release capsules containing nitrofurantoin. The product descriptions and administration sections identify the relevant preparation.
For a clinician comparing an answer with a prescription, the practical question is whether both refer to the same formulation. An ingredient-level summary can be informative while leaving that unresolved.
A useful exercise is to open the two records and identify the pharmaceutical form before reading the administration section. Then examine the AI answer: does it preserve the formulation, or silently generalise from one preparation?
This gives “check the source” a specific meaning. The source must describe the product under discussion, and the summary must retain the distinction that makes it relevant.
Tacrolimus: brand identity can be clinically meaningful
Tacrolimus illustrates why an active ingredient alone may be too broad a search key. The MHRA’s reminder about oral tacrolimus highlights reports of toxicity and graft rejection associated with inadvertent switching between products, and the importance of brand identity and specialist oversight of changes.
The learning question is therefore wider than arithmetic: “Which exact product does this statement concern, and what information would be lost if its name were shortened?”
A generic summary might explain the medicine’s purpose adequately while obscuring a product distinction that matters in practice. Source-grounded answers should make those distinctions visible. A numerical match, considered alone, would not settle the question.
Five source questions worth asking of any medicines AI
The same questions apply when assessing iatroX, Doximity or another platform:
- Which jurisdiction and exact product support this answer? A correct reference for one market or formulation may not resolve the product question in front of you.
- Can I open the supporting passage? A source name is less informative than a link to the relevant document and section.
- What context was matched? Indication, treatment stage, route, formulation, age, weight, organ function and interacting medicines can change what needs checking.
- What was the check designed to establish? Source retrieval, citation checking and deterministic regimen verification answer different questions.
- What happens outside coverage? Missing context, conflicting information and unsupported regimens should be recognisable to the reader.
These questions are more useful than asking whether a platform is simply “accurate”. They expose what a particular answer establishes and which uncertainties remain.
For a team evaluating a tool, a small set of anonymised educational cases can make the discussion concrete. Include an ambiguous indication, a formulation distinction and a product whose brand matters. Review the source trail and the treatment of missing information, rather than scoring presentation alone.
Turning the distinction into prescribing learning
For learners preparing for the PSA, GPhC registration assessment or medical examinations, a useful exercise is to explain why a superficially plausible answer cannot yet be accepted. The learning objective becomes recognising the missing condition, identifying the appropriate source and showing how that changes the question.
iatroX Medicines provides evidence-linked UK medicine guides within the platform. Alongside question practice, that creates a route from an uncertain answer to a more precise product-information question.
For CPD, the useful record is equally specific: the assumption you initially made, the source distinction you discovered and the change you intend to make in future checking. iatroX CPD supports turning a learning need into a focused session and a reflection you review. Reading a launch announcement is only the starting point; the educational value comes from examining how you would assess the next medicines answer.
Frequently asked questions
What does an SmPC add to an AI medicines answer?
It provides product-specific information about authorised uses, administration, contraindications and precautions. For UK questions, the value is being able to identify the relevant product record and examine the conditions behind a summary, including distinctions between indications or formulations.
Does a verified dose settle whether a prescription is appropriate?
A match can support a particular dosing claim within the verifier’s coverage. It does not, on its own, establish that the indication, product and patient context were correctly identified or that the overall treatment decision is appropriate.
How can I use this comparison for CPD without prescribing from it?
Choose a medicines-information question that revealed a gap in your checking. Identify the exact product and source section, record the missing context, and reflect on how you would handle that uncertainty next time. Use an educational case without identifiable patient information.
Put the source question into practice
Explore iatroX Medicines, then turn a source question into a focused learning session with iatroX CPD.