Credo Health

The grade describes the mechanism, not the quality. Genuine model work sits at the centre: inferring that a record should exist from signals in the records you already hold is a real inference task, and agent assisted workflows drive the recovery.

But the company describes its offering as combining digital and manual retrieval with AI driven analysis and human expertise, and the value a customer buys includes an operational capability, chasing charts by telephone, portal and fax, that is not reducible to a model. That operational reach is the differentiator as much as the inference is, which is why this is B rather than A.

The product's default is unusually honest and that is the substance of this grade. Rather than assembling whatever records arrive and presenting the result as a patient history, Care Map explicitly tells the user what is still missing, flagging unresolved gaps rather than silently delivering an incomplete picture. Combined with source citation on every element, a clinician can see both where each fact came from and where the holes are, which is a materially better epistemic position than any product that hands over a summary with no completeness signal.

Held at B because the inference itself is unmeasured in both directions. No false positive rate is published for records the system believes should exist but do not, which sends staff chasing phantoms, and no false negative rate for gaps it fails to notice, which leaves the clinician confident in a record that is still incomplete. Ask for both.

Output is source cited, which matters more here than in most records because the history is stitched from many organisations and a clinician needs to know which one each element came from. Beyond that nothing is disclosed: no model or model family named, no accuracy figure for encounter footprint inference or gap detection, no evaluation methodology.

The figure the company does publish, that 30 to 50 percent of high value encounters are not returned digitally, is a claim about the national networks rather than about its own performance, and it is discussed on the interoperability axis.

Two features of this record are unusual and the second is disclosed in the company's own terms, which is why this sits above the floor. The first is the artefact. The product's purpose is to gather a patient's history from every holder and present it as one summarised record before a visit, and that consolidated history exists nowhere else, not even in the ordering practice's own system, because the company creates it.

Retention of the assembled artefact is therefore a distinct question from retention of the source documents, and neither is addressed. The second is the onward use, which the company's own terms disclose: information it supplies feeds risk adjustment coding. So the record is not only read by a clinician before a visit, it is mined for diagnoses that determine payment.

A summarisation model operating where surfacing more codeable conditions has direct financial value has an incentive gradient a purely clinical summariser does not, and nothing published describes how the two uses are separated or whether the same output serves both. Two further parties sit in the chain: a partnership with a release of information business, and telephone and fax retrieval channels, which put records through media with no access control at all at the receiving end. Ask what is retained and for how long, whether retrieved records inform model development, and whether clinical summarisation and coding support are the same output.

More research activity than most at this stage, none of it outcome data. The company reports a body of work combining survey findings from more than 500 physicians and care team members, insights from a point of care beta with more than 150 clinicians, and real world pilot evidence on what drives action in clinical workflows, particularly in Medicare populations. That is a substantial evidence gathering effort and it is unusual for a seed stage company.

It remains C because it is vendor run and vendor reported, no customer organisation is named, and nothing measures whether more complete records changed a clinical decision or an outcome. Record increase figures are described as estimates based on partner reported data.

Refresh trigger: if the beta and pilot findings are published with methodology, revisit this grade.

Converted from Not Rated after a second search. No retention period, training use statement or de identification posture was located. The prior note's framing of the manual retrieval path stands, and two things now sharpen it.

The first is what the assembled artefact is. The product's purpose is to gather a patient's history from every holder and present it as one summarised record before a visit. That consolidated history exists nowhere else, not even in the ordering practice's own system, and the company creates it. Retention of that artefact is therefore a distinct question from retention of the source documents, and nothing addresses either.

The second is the onward use. The company's own terms disclose that information it supplies feeds risk adjustment coding. That means the record is not only read by a clinician before a visit; it is mined for diagnoses that determine payment. A summarisation model tuned in an environment where surfacing more codeable conditions has direct financial value has an incentive gradient that a purely clinical summariser does not, and nothing published describes how the two uses are separated or whether the same output serves both.

The partnership with a release of information business adds a third party to the handling chain, and the telephone and fax channels put records through media with no access control at all at the receiving end.

Ask what is retained and for how long, whether retrieved records inform model development, and whether clinical summarisation and coding support are the same output.

Upgraded from Not Rated after a second search, which reached the company's published master terms and found a control the first pass did not.

The company operates a customer verification process, and its terms require the customer to represent and warrant that it is a regulated entity under the health privacy rule, with the company entitled to verify that representation through third party information sources. In record retrieval that is the load bearing control. The lawfulness of obtaining someone's records on another party's behalf depends entirely on whether that party is entitled to receive them, so a supplier that verifies its customer's regulatory status before retrieving anything is addressing the right question rather than the easy one.

A second disclosure in the same terms deserves attention because it allocates a significant risk. The company states plainly that it is not practising medicine and does not warrant that any diagnosis code assigned for risk adjustment purposes on the basis of information it supplies will comply with applicable law, and that such assignment is the customer's sole responsibility. Risk adjustment coding is among the most actively enforced areas in healthcare, and a supplier feeding records into that process is right to be explicit about where responsibility sits. A buyer should read it as a clear allocation rather than as boilerplate.

The prior note's central question remains open. In the manual retrieval path, who is the requester of record, and on what basis, is not stated.

Ask who requests, under which provision, and what agreement covers the release of information partner.

Converted from Not Rated after a second search. No SOC 2, HITRUST, ISO 27001, trust centre or report request path was located.

One comparative observation makes this more informative than a bare absence. In the record retrieval and release of information segment, an independently attested security posture is the norm rather than the exception: competing retrieval services and portals routinely advertise a SOC 2 Type 2 alongside their compliance claims, because law firms, insurers and health systems ordering records have made it a procurement condition. Measured against its own segment, the absence here is conspicuous rather than merely unstated, and that is the right frame for a buyer.

The exposure is also unusual in shape. This business exists to assemble records from many holders into one place, so its systems accumulate complete clinical histories for patients who have no relationship with the company and never chose it. The concentration is the product. Add a partnership with a release of information business handling inbound requests, and a manual path running over telephone, portal and fax, and the surface spans systems the company operates and channels it does not.

Ask whether an attestation exists and what it covers, how retrieved records are protected in transit and at rest, what controls apply to the fax and telephone paths, how staff access to assembled histories is limited and logged, and what security terms bind the release of information partner.

No FDA clearance or device authorisation was located and none is expected, since the product retrieves, assembles and summarises records rather than diagnosing. The regulatory exposure sits in information exchange and release of information rules rather than in device regulation, and in the risk adjustment audit regime discussed on the governance axis.

The grade describes incentive structure and disclosure, and this record introduces a mechanism the index has not previously separated out. The commercial pitch is that value based care organisations can risk adjust accurately and drive superior reimbursement outcomes, so more complete retrieval raises risk scores. But the mechanism is different from the coding gradient tracked elsewhere: nothing is coded more aggressively, and no condition is inferred.

Conditions that genuinely exist and are genuinely documented, at another organisation, are brought into view. That is the most defensible version of the gradient available, and it also improves care independently of payment, which is the counterweight and belongs stated.

Two things keep it at C. Applying the distinction drawn on the Kennar record, retrieving external documentation is retrospective in character rather than prospective, so it lives in the part of risk adjustment where audit scrutiny concentrates.

And no fairness, subgroup or demographic performance disclosure was located, which matters here in a specific way: patients whose care is most fragmented across organisations, and whose records are most likely to sit in systems that do not participate well in exchange, are disproportionately those with unstable housing, frequent moves or coverage churn. If retrieval completeness varies by population, so does everything downstream of it.

Output is source cited, and that matters more here than in most records because of what the product is. A history stitched together from many organisations arrives as one document, and a clinician reading it needs to know which organisation each element came from, because provenance carries reliability: a medication list from the prescribing practice and one transcribed from a fax carry different weight, and a citation lets the reader apply that judgement.

Per element attribution across a multi source assembly is the right mechanism and few products of this shape offer it. Held at C because nothing measures the inference that decides what appears. No accuracy figure for encounter footprint inference or gap detection was located, and those steps determine whether the assembled history is complete.

The failure is the one this index keeps finding: a record presented as a full history that quietly omits an encounter is more dangerous than an obviously partial one, because the clinician stops looking. The figure the company does publish, that a large share of high value encounters are not returned digitally, is a claim about the national networks rather than about its own performance and does not fill the gap. No warranty, indemnity or remediation commitment was located. Ask for recall on encounter detection against a known complete history, and what the clinician is shown about what the assembly could not find.

The deepest retrieval footprint in this category, and it earns the grade by not stopping where everyone else stops. THREE national networks are named, Carequality, CommonWell and eHealth Exchange, alongside regional health information exchanges, which is broader national coverage than any other record here.

Then comes the part that is genuinely unique: the company states that 30 to 50 percent of high value encounters are NOT returned digitally by those networks, and it pursues them through agent assisted telephone, secure portal and fax retrieval. Every other retrieve and summarise product in this index stops at what the exchanges hand back.

Records are then deduplicated and stitched with source citation, and point of care delivery into the EHR is achieved through partnership rather than natively, which is the one soft spot in an otherwise exceptional record on this axis.

Converted from Not Rated after a second search. No hosting model, cloud provider, region or residency commitment was located.

The architecture is nonetheless partly visible from what the company describes, and it is more distributed than a software product would be. Retrieval runs through several channels at once: electronic connections where they exist, a partnership with a release of information business handling inbound requests through its own platform, and a manual path using telephone, secure portal and fax where a holder supports nothing better. That means records enter the company's environment from many directions, some of them through systems it does not operate.

The fax path deserves a specific mention because it is easy to dismiss as legacy plumbing. A received fax is an image arriving over a channel with no authentication, no encryption in the ordinary case, and no delivery confirmation to the sender's satisfaction, and it has to land somewhere before anything digitises it. Where that is, and what protects it, is a deployment question rather than an operational detail.

The accumulated histories are the other half. A retrieval business necessarily builds a store of assembled records, and where that store sits, how long it persists, and whether one customer's retrievals are separated from another's are the questions this axis exists to ask.

Ask where the platform and the record store run, what the retention schedule is, how the fax and portal intake paths are secured, and which subprocessors including the release of information partner touch records.

No price, tier or pricing mechanism was located.

The commercial model is worth probing specifically, because the manual retrieval path is a variable operational cost the vendor carries per record chased, which pricing must absorb somewhere, and because record volumes vary enormously between patients. Establish whether you are charged per request, per record retrieved or on a flat subscription, since those distribute that variability very differently between the parties.

Coverage is defined by buyer and by population rather than by specialty, and the population argument is a sharp one. Three buyer types are addressed, value based care providers, management services organisations and health plans, and the company singles out Medicare populations as the setting where care is most distributed across organisations and where retrieval therefore matters most.

That is a coherent targeting argument rather than a generic claim, since the value of finding records elsewhere scales directly with how fragmented a patient's care has been. Graded B because no clinical specialty depth or instrument level behaviour was located, and the workflow addressed is pre encounter preparation rather than a range of care settings.