Synapse Medicine

The machine learning here sits upstream of the clinical decision rather than inside it, and that distinction is worth understanding before comparing this vendor to others in the category. Natural language processing reads manufacturer product documentation, health authority recommendations and research papers, and turns that unstructured text into the structured medication knowledge the platform then reasons over.

So the model builds the corpus; conventional pharmacological logic makes the recommendation. That is a genuinely different risk profile from a learned recommender, and arguably a harder one to audit, because an extraction error does not announce itself as a wrong answer once but propagates silently into every recommendation touching that drug.

A second product line applies the same approach to pharmacovigilance, automatically coding and prioritising adverse drug reaction reports, which is a clearer case of the model doing the work. Graded B because without the extraction layer the company would be maintaining a database by hand, which is precisely the incumbent product it was founded to replace.

The clinical products are advisory and reach a prescriber or a pharmacist who decides, with medication reconciliation producing a list for a clinician to confirm rather than an order. The pharmacovigilance product occupies a different position that deserves separate attention: automating the coding and prioritisation of adverse drug reaction reports means the model influences which safety signals a national pharmacovigilance centre examines first, and a report deprioritised is a report that may wait.

That is oversight of the safety system itself rather than of an individual prescription. Retrieval located no acceptance rate, no override figure, no confidence threshold and no description of how a miscoded report is caught, and for the pharmacovigilance line that last question is the one that matters most.

The method is named and the sources are named, which is more than most: the company states that it uses natural language processing to analyse and classify medication information, and identifies the three source types it ingests. What is absent is the number that matters for a system of this design.

An extraction pipeline turning regulatory text into structured drug knowledge has an accuracy, a recall and a failure mode, and none of those is published, nor is any description of how extracted content is validated before it reaches a clinician, whether a pharmacist reviews it, or how often the corpus is refreshed against updated labelling.

For a product whose whole architecture rests on reading documents correctly, publishing the extraction accuracy would be the single most informative disclosure available and it is the specific ask to put to this vendor.

The institutional signal is strong and the published evidence was not located. Deployments include Assistance Publique Hopitaux de Paris, which is among the largest hospital groups in Europe, the Hospices Civils de Lyon, the Bordeaux teaching hospital, more than a hundred hospitals in total, and dozens of telemedicine companies.

More striking, the French national medicines agency and the national digital health agency are named as users, so the institutions that regulate medicines and set digital health standards are customers of the product. A research partnership with the national health research institute on detecting drug interactions is also on record. None of that is evidence of benefit, and this index grades adoption and evidence separately as a matter of standing practice.

Scope stated plainly: two retrieval passes located no outcome study, no accuracy evaluation and no peer reviewed publication on the product itself, and a targeted search of the founders' pharmacovigilance literature is owed before this grade is treated as settled.

Medication reconciliation requires assembling a complete and accurate list of what a patient is actually taking, which means the platform handles full medication histories rather than single prescriptions, and the telemedicine integrations extend that to patients outside any hospital's walls.

The governing framework is European rather than American, so the General Data Protection Regulation and the French public health code set the obligations, and the company's institutional customer base implies those obligations have been assessed by demanding buyers. Retrieval located no retention schedule, no de identification statement and no description of whether patient data contributes to improving the extraction models.

That last question has a specific edge here, because the corpus is built from published sources rather than from patient data, so a clear statement that patient data never enters model training would be straightforward to make and would distinguish this vendor sharply from its peers.

This grade follows the index's practice of assessing a vendor against the framework that actually governs it rather than penalising it for lacking an American instrument it has no occasion to sign. Synapse Medicine is French, its customers are French and European health institutions, and the obligations that bind it are the General Data Protection Regulation and the French public health code rather than the United States privacy rule, so the absence of a published business associate agreement is not by itself a deficiency.

Two things hold the grade at C. The company has stated an intention to operate in the United States, and a business associate agreement becomes necessary the moment that happens, with nothing published to indicate readiness. And no European equivalent is published either: no data processing agreement terms, no subprocessor list and no privacy documentation was located in two passes, so a buyer in either jurisdiction sees nothing before contact.

Retrieval located no published certification of any kind, no trust centre, no penetration testing cadence and no vulnerability disclosure policy. The specific gap for a French vendor is named rather than generic. Hosting personal health data for third parties in France requires certification as an Hebergeur de Donnees de Sante under article L.1111-8 of the public health code, a scheme built by the national digital health agency on top of ISO 27001 and issued as two certificates.

Version 2.0 of that framework, approved in 2024, adds sovereignty requirements obliging health data to be hosted physically within the European Economic Area with any transfer outside made public, and since 16 May 2026 only providers conforming to version 2.0 may continue hosting activity. A company serving Assistance Publique Hopitaux de Paris and the national medicines agency is necessarily operating inside that regime, whether certified itself or through a certified host. Publishing which, and confirming version 2.0 conformity against a deadline that has now passed, is the highest value disclosure available to this record.

No United States clearance was located and none would be expected for a company whose operations are European, though the stated ambition to sell in the United States makes it a question rather than a non issue. The live question is in the home market.

Software intended to inform prescribing decisions generally falls within the European medical device framework, and the Medical Device Regulation classifies decision informing software more strictly than the directive it replaced, so a platform advising on drug interactions and reconciliation across more than a hundred hospitals plausibly requires a conformity assessment. Two retrieval passes located no CE mark, no device class, no notified body and no statement of regulatory position. A direct competitor in this category publishes exactly that information, including the class and the route, so the disclosure is achievable and its absence here is a gap rather than an industry norm.

Retrieval located no bias assessment, no performance reporting by any patient characteristic, no governance documentation and no post deployment monitoring statement. The exposure specific to this architecture is a coverage question rather than a demographic one, and it becomes acute with the company's stated international expansion.

A knowledge base assembled by reading French and European regulatory documentation inherits the scope of European drug approvals, European labelling conventions and European recommendations, so the same extraction pipeline meeting American or Japanese labelling encounters different documents, different approved products and different conventions for expressing the same warning.

Nothing published describes how coverage and extraction quality are assessed when the corpus crosses a jurisdiction, and a clinician in a new market has no way to know whether the answer they receive rests on thoroughly indexed sources or thinly indexed ones.

The distribution model is the notable feature. Alongside direct hospital deployment at more than a hundred institutions including the largest French university hospital groups, the platform is sold modularly to software developers through an application programming interface, and dozens of telemedicine companies embed it, including one of the largest international virtual care providers.

That makes this vendor a medication intelligence layer inside other people's products as well as a product in its own right, which is the same structural position drug knowledge base incumbents occupy, reached by a different route.

Held at B because no specific hospital information system certifications, marketplace listings or interoperability standard conformance statements were located, and because French hospital software estates differ enough from American ones that this integration record says little about readiness for the market the company says it wants to enter.

Delivery is cloud based, offered both as a direct product and as an interface other software embeds, which means patient data may reach the platform through a partner's product rather than through a hospital's own systems and the contractual chain lengthens accordingly. Retrieval located no named hosting provider, no region and no residency commitment.

The residency question has a definite answer available in this jurisdiction rather than an open ended one, since the French health data hosting framework in its current version requires personal health data to be hosted physically within the European Economic Area and any transfer outside it to be made public, so the vendor could resolve this axis with a single sentence naming its host and confirming its conformity. Graded C rather than lower because the regulatory environment constrains the answer even where the vendor has not published it.

No price, unit, pricing basis, contract shape or implementation fee was located, and the modular structure makes the unit question sharper than usual, since the same underlying medication intelligence is sold to a hospital, to a telemedicine company embedding it, and to a pharmacovigilance centre, three buyers with entirely different volumes and value.

Funding is well documented at roughly 39 million dollars across three rounds while pricing is not, which is the ordinary asymmetry of a venture backed company but worth naming. One disclosure of a different kind belongs on the record and it is a credit rather than a gap: the company states plainly that it is completely independent of the pharmaceutical industry, and its investor base of a healthcare professionals' insurer, a mutual insurance fund and public investment banks is consistent with that claim. In a category where industry funded medication content is a live commercial model, an unprompted independence statement is the disclosure a buyer most needs and fewest vendors make.

Coverage follows the medication rather than a specialty, and the settings are unusually varied: acute hospital care at major university hospital groups, virtual care through telemedicine partners, national pharmacovigilance through the adverse event coding product, and public health institutions including the medicines regulator and the digital health agency.

That range of institutional type is wider than any other vendor in this category and reflects a platform sold as infrastructure rather than as a workflow tool. Held at B on a currency caution the record states rather than hides: the international footprint described in company material, spanning offices in several countries and operations in the United States and Japan, comes from a 2022 funding announcement describing plans over the following two years, and no confirmation that the expansion occurred as described was located, so a buyer outside France should verify local presence directly rather than relying on it.