MedScrub

The grade describes the mechanism, not the quality. The de identification engine is genuine model work and it is the differentiator: reliably locating all eighteen HIPAA identifiers across messy clinical text, reversibly, is a hard natural language problem and it is what the company actually builds. But the clinical reasoning a buyer relies on is explicitly performed by third party models the customer selects, GPT-4, Claude, Gemini, Mistral or a local model under Ollama, and MedScrub is deliberately agnostic between them. So the AI producing the clinical output is not MedScrub's AI. That is a coherent and arguably safer architecture, and it is also the reason this is not an A.

No oversight mechanism is described for the clinical output. There is no confidence signal, no threshold, no abstention behaviour and no required review step for the visit summaries or the chart question answering. One structural point matters more than the missing parameters, and it is the third instance of this pattern in the lane: because the clinical reasoning is performed by a model the customer selects, the oversight and error properties of the output depend on a component MedScrub neither controls nor evaluates. The same summary produced through a frontier model and through a small local model are different artefacts with different failure rates, and nothing here tells a buyer which to expect. Establish who is accountable for output quality when the buyer chose the model.

The most complete answer to the whose model is it question found anywhere in this index, because the answer is whichever model you choose and the company names them: GPT-4, Claude, Gemini, Mistral and local models under Ollama. Model agnosticism stated as a design property rather than left vague is real transparency, and it lets a buyer reason about where content goes and what it costs. Held at B because the single most important number is missing. For a de identification engine, RECALL ON IDENTIFIER DETECTION IS THE SAFETY PROPERTY: one missed identifier defeats the entire architecture, since the whole argument rests on nothing identifiable reaching the model. The company describes HIPAA grade FHIR accuracy without publishing a figure, a method or an evaluation set. Ask for measured recall across all eighteen identifier categories and what happens on a miss.

Claims without substantiation. No customer is named, no funding is disclosed, no deployment is described and no independent evaluation was located. The two quantified claims, 78 percent token cost savings and HIPAA grade FHIR accuracy, carry no baseline, method or denominator. The token saving figure is at least the kind of claim a buyer can verify quickly in a trial, which is more than most of this category offers, but it measures cost rather than correctness.

The strongest privacy architecture in this category and it rests on several independent mechanisms rather than a single promise. All eighteen HIPAA identifiers are removed before content reaches any model, so a consumer or frontier model never receives identifiable data. The substitution is reversible so utility is preserved. The proxy can be self hosted. Local models are supported under Ollama, meaning the entire path can run without content leaving the premises. And patient data syncs into a clinical data repository the customer owns rather than into the vendor's cloud. Any one of those would be notable; together they let an organisation eliminate the exposure rather than manage it, which is the property this index credits as engineered privacy rather than policy privacy. Two things a buyer must still establish, and they are the reason this is not a perfect record: the unpublished recall figure discussed on the transparency axis, and the hard case for any de identification pipeline, which is an identifier spoken or written conversationally rather than in a structured field.

HIPAA framing runs through the entire product, with references to HIPAA grade accuracy, stripping all eighteen identifiers and reaching a HIPAA compliant build quickly, but no compliance statement and no business associate agreement terms were located. Graded C rather than Not Rated because claims exist and none is an assurance. One point deserves separate attention because it is the product's core commercial argument rather than a detail: if identifiers never reach the model provider, a customer may not need a business associate agreement with that provider at all. That is a genuinely significant proposition and it is exactly the kind of conclusion that should be validated with counsel against your own risk posture rather than accepted from a vendor page.

No SOC 2, HITRUST, ISO 27001 or other attestation was located, and no trust centre or security page was found. Not Rated reflects absent retrieval. The self hosted deployment option shifts much of the security burden onto the customer's own environment, so establish which controls the vendor attests to in each deployment mode rather than treating the answer as uniform.

No FDA clearance, device authorisation or regulatory pathway statement was located. Not Rated reflects absent retrieval. The product summarises and answers questions from the record rather than recommending treatment, so the scoping question is milder than for several records in this category, though the coding assistance and care plan generation use cases named for developers sit closer to the boundary than the summarisation does.

The grade describes disclosure. No fairness, subgroup or demographic performance disclosure of any kind was located, and there is a specific reason that matters more here than in most records rather than less. De identification engines are known to perform unevenly across naming conventions: a model trained predominantly on Western given and family names detects them more reliably than names from other traditions, and the same applies to address formats and identifier patterns. If recall varies by population, then the privacy protection this product sells is itself unevenly distributed, and the people least protected are precisely those whose names the training data underrepresents. That is the same structural shape this index recorded for patient matching accuracy and for speech recognition across accents. Published recall broken down by name origin would be a genuinely differentiating disclosure and no vendor in this space offers it.

Real integration is described and no system is named. The product pulls conditions, laboratory results, vitals, medications and imaging from the EHR and syncs them into a customer owned clinical data repository, and FHIR accuracy is referenced, which implies standards based exchange. But no EHR vendor, integration mechanism, certification or marketplace listing was located anywhere, so an organisation cannot establish whether its own system is supported without asking. The developer facing surfaces, an HTTP API and an MCP server, are genuine integration points and unusually current, but they serve builders rather than answering the EHR question.

The second A on this axis in this category and it is earned differently from the first. Three independent choices are offered rather than one: the de identification proxy runs self hosted or in the cloud, the models can be frontier services or local models under Ollama, and the patient data repository is owned by the customer rather than held in the vendor's cloud. Taken together an organisation can construct a path in which no patient content and no de identified content ever leaves infrastructure it controls, which is the strongest data control posture found in this lane. Held short of a complete answer only because no residency regions, hosting detail or portability commitment is published for the cloud option, and because nothing describes what the vendor can see in a self hosted deployment.

No price, tier or pricing mechanism was located. Not Rated is the house convention for absent pricing rather than a low grade. The 78 percent token cost saving claim is a cost argument rather than a price, and note that it describes savings on the customer's own model spend rather than on what MedScrub charges, so the two should be assessed separately.

Coverage is described by use case rather than by care setting or specialty, and no clinical setting is named anywhere. The clinician facing functions are population health screening, pre visit planning and chart question answering, which point toward ambulatory primary care without saying so, and the developer facing use cases span intake and symptom checking chatbots, coding assistance, referral routing, care plan generation and staging environment data. That is breadth of application rather than depth in any setting, and no specialty specific behaviour, instrument or terminology handling was located.