Curbside Health

THE GRADE DESCRIBES THE MECHANISM, NOT THE QUALITY, and this record belongs on the index's honest non inflation roster alongside Candid Health, HeartSciences and Schrodinger. Curbside's own AI page states the scope precisely: AI structures and drafts content from guidelines and PDFs, supports updates as evidence changes, and lets a system run more pathways without proportional staffing. It is an AUTHORING AND MAINTENANCE ACCELERANT. THE CLINICAL LOGIC ITSELF IS HUMAN AUTHORED, and the company says so deliberately, stating that AI assists and clinicians decide and that there is no autonomous clinical decision making. Strip the AI and a complete saleable product remains, which is what the company sold before generative tooling existed. WHY THIS IS INDEXED RATHER THAN REJECTED, and the line matters because Orion Health Care Pathways was rejected from the same source list: Orion makes no AI claim anywhere and its pathway designer is an administrator drafting tool. Curbside applies AI to a real and non trivial task, extracting explicit decision logic from unstructured clinical prose, and documents where it is and is not used. In a market that rewards overclaiming, a vendor that scopes its AI down and markets the absence of black boxes as the feature is graded accurately, not penalised.

THE STRONGEST AXIS ON THIS RECORD, AND THE OVERSIGHT IS THE PRODUCT RATHER THAN A CONSTRAINT BOLTED ONTO IT. The AI never reaches a patient facing decision. It drafts content which a named clinical owner reviews, a multidisciplinary workflow approves and a version control system tracks before anything goes live, and the published principles state human in the loop at every stage, transparent logic and provenance, and no autonomous clinical decision making. Because the output is a visible decision tree rather than a score, a clinician can disagree with a specific branch rather than with an opaque recommendation, which is the property the Kinometrix record credited in a different form. HELD AT B ON A SPECIFIC AND SERIOUS GAP: the reviewer is the entire safety layer for the AI drafting step, and no error rate is published, so the burden that layer carries is unquantified. A reviewer approving a fluent AI generated pathway is exposed to precisely the automation bias the Epic In Basket record describes, where a draft that is usually right invites less scrutiny than a blank page. Nothing in the published material addresses reviewer burden or review depth.

AN UNUSUAL SPLIT WORTH NAMING: the SCOPE of the AI is disclosed better than almost anything in this index, and the TECHNOLOGY is disclosed not at all. On the scope side there is a dedicated page naming three specific applications, a stated philosophy, and an explicit negative boundary that the system performs no autonomous clinical decision making, plus a separate Responsible AI page in the site footer. On the technology side there is no model class, no foundation model named, no description of how guideline prose is converted into decision nodes, and no evaluation of any kind. THE GAP THAT MATTERS MOST IS THE EXTRACTION ERROR RATE. If AI drafts a pathway from a PDF and a clinician approves it, the accuracy of that drafting step determines what the reviewer has to catch, and nothing is published about how often it gets a decision node wrong or in which direction it errs. Publishing that figure would be the single highest value disclosure available to this vendor and would separate it decisively from every content vendor that simply asserts AI assistance.

No published study of any kind was located, and the absence is unusually visible because the company advertises the category. THE TOP LEVEL NAVIGATION CARRIES AN EVIDENCE AND INSIGHTS MENU WITH AN EVIDENCE LINK, AND THAT PAGE RETURNS A 404. Recorded as a factual observation rather than an inference, and worth re checking on refresh since it may simply be unbuilt on a recently rebuilt site. What exists is a reach figure, more than one million patient lives impacted, which counts exposure rather than effect, and one named implementation at OSF HealthCare, a sixteen hospital system with 2,131 licensed beds. No compliance delta, no variation reduction figure, no outcome, no denominator and no comparison period is published anywhere. This is a product whose entire thesis is that guidance fails at execution rather than at evidence, so a published adherence figure with a denominator is the natural and expected proof, and it is exactly what is missing.

A privacy policy, terms of service and a separate compliance statement are published and linked, and NONE WAS OPENED THIS PASS, recorded explicitly per the Droxi and Sickbay lesson rather than left implicit. No statement was located anywhere in the material read about whether the platform holds PHI. THE CONTRAST WITH C8 HEALTH IS INSTRUCTIVE AND BELONGS ON BOTH RECORDS: C8 states plainly that it operates without requiring PHI, which is a strong and checkable architectural claim, while Curbside is silent, and its SMART on FHIR posture implies patient context reaching the pathway. THE SHARPEST UNANSWERED QUESTION IS THE COMMUNITY LAYER. Organisations share and copy each other's pathways through the platform, so establish what is shared by default versus by explicit election, whether an institution's own protocols become visible to competitors, and whether customer authored content is used to train or improve the drafting models. A content sharing network makes data use a live question rather than a theoretical one, in the same way CLEW's change control plan did for retraining.

The footer compliance statement page was opened and read in full, so this assessment rests on an evidenced absence rather than a failed search. The page is titled Commitment to Compliance and states that Curbside Health is designed to support healthcare organisations operating in highly regulated environments. It then lists four compliance focus areas as bare labels: healthcare data protection standards, clinical governance and auditability, regulatory defensibility, and enterprise risk management support. HIPAA is not mentioned. Neither is a business associate agreement, a covered entity, a business associate, SOC 2, HITRUST or ISO 27001. The single concrete commitment anywhere on a page about healthcare compliance concerns accessibility standards. The operative sentence is that the company partners with customers to support compliance reviews, assessments and documentation as needed, with the call to action being a demo request. That is a coherent posture rather than an oversight, and it should be read as one: compliance is handled entirely inside the sales process, and nothing is established before a buyer talks to someone. Two observations temper it. The four focus areas render with unreplaced stock placeholder images, which alongside a broken evidence link suggests a site still being built out following a 2026 rebuild, so this is worth re checking. And the absence matters more here than it would elsewhere because the product is electronic health record integrated and names SMART on FHIR, so patient context reaches it. The privacy policy and terms of service were not opened in this review.

A DEDICATED SECURITY PAGE THAT NAMES NO CERTIFICATION AT ALL, which is a specific and reportable posture rather than simple silence. The controls described are substantive and appropriate: defence in depth architecture, role based access controls, encryption in transit and at rest, secure development and deployment practices, ongoing monitoring and risk assessment, access provisioning with least privilege reviews, audit logging and activity monitoring, incident response and breach notification procedures, and support for customer third party security reviews. WHAT IS ABSENT IS ANY INDEPENDENT ATTESTATION: no SOC 2 of any type, no HITRUST, no ISO 27001, no trust centre, and security documentation available only by request through the demo form. DESCRIBING YOUR OWN CONTROLS IS SELF ATTESTATION, WHICH IS PRECISELY WHAT AN INDEPENDENT AUDIT EXISTS TO REPLACE. Contrast C8 Health in the same segment, which specifies SOC 2 Type II and runs a public trust centre. A SOC 2 Type II would move this grade immediately and is the obvious next step for a company selling into enterprise health systems.

No FDA clearance, none claimed, and none apparently required. The grade records the absence; the note records that the regulatory position is among the cleanest in the index. THE CLINICAL DECISION SUPPORT EXCLUSION TURNS ON WHETHER A CLINICIAN CAN INDEPENDENTLY REVIEW THE BASIS FOR A RECOMMENDATION, AND HERE THE BASIS IS LITERALLY DRAWN AS A DIAGRAM. The pathway is a human authored, version controlled decision tree with a named clinical owner, displayed as explicit logic rather than inferred output. That is the strongest possible position on the basis test, sitting alongside C8 Health and in direct opposition to the Epic Sepsis Model, whose basis is disclosed to nobody. The company markets adjacent to this argument, stressing auditability and defensibility and alignment with CMS and regulatory priorities, but never states a regulatory rationale explicitly, so ask for it directly.

A DISTINCTION THIS RECORD MAKES CLEARER THAN ANY OTHER IN THE INDEX AND WHICH SHOULD BE REUSED: CONTENT GOVERNANCE AND MODEL GOVERNANCE ARE DIFFERENT THINGS, AND A VENDOR CAN BE EXCELLENT AT THE FIRST WHILE PUBLISHING NOTHING ON THE SECOND. Content governance here is genuinely strong and specified: named clinical owners, defined review cycles, multidisciplinary oversight, version history and audit trails, and explicit status showing what is live, outdated or under review. That directly addresses the pathway staleness exposure this index identified across the whole segment. Model governance is asserted rather than evidenced: the Responsible AI principles list four commitments with no method, no evaluation, no error rate and no audit behind any of them. A SECOND EXPOSURE MIRRORS THE C8 HEALTH FINDING AND APPLIES TO ANY CONTENT SHARING NETWORK: when organisations copy each other's pathways, the practice patterns and resource assumptions of the institutions that publish most become the default others inherit, and a pathway that assumes staffing or equipment a hospital does not have is not a best practice there. Nothing addresses who curates the shared library or how local applicability is assessed.

The most specific interoperability claim in this segment, and better than its two closest peers on the standard even while thinner on named partners. SMART ON FHIR IS NAMED as the integration model, which is more concrete than AgileMD, which describes deep embedding but names no EHR and no standard, and than C8 Health, which reaches the EMR through IP based magic links. Two operational claims carry real weight if they hold: pathway logic and orders remain under clinician control with no information services involvement after the initial integration, which addresses the bottleneck that keeps most institutional content stuck in queues, and the community layer lets organisations copy pathways ACROSS DIFFERENT EHRS, which implies the content model is genuinely EHR agnostic rather than Epic specific. HELD AT B ON THE HEDGE AND THE GAPS: the phrasing is SMART on FHIR READY rather than in production, no EHR vendor is named anywhere, no marketplace or app listing was located, and no FHIR resources or scopes are described.

One named implementation, and it is a substantial one: OSF HealthCare, a sixteen hospital integrated system with 2,131 licensed beds, nearly 24,000 staff and more than 150 care locations across Illinois and Michigan, announced February 2024. Third party marketplace material adds a concentration in children's hospitals and paediatric providers, which is a coherent segment because paediatric pathways are poorly served by adult oriented content libraries. Reach is given as more than one million patient lives impacted. WHAT IS MISSING IS EVERYTHING A BUYER WOULD USE TO PLAN: no customer count, no implementation timeline, no resourcing expectation, no hosting model, and no data residency statement of any kind. A CURRENCY CAVEAT IS STATED RATHER THAN IMPLIED, the same handling given to AlertWatch:OB and Kinometrix: the only named reference dates from February 2024 and no more recent named deployment was retrieved, against a site that was evidently rebuilt in 2026. Establish current commercial traction directly.

No pricing published at any level. No rate card, no stated unit of pricing, no indicative band and no implementation or content migration fee. Every route on the site leads to a demo request form, including the request for security documentation. Worth noting for contrast that this vendor does NOT exhibit the upside published price gated asymmetry recorded against five other vendors in this run, because it publishes no customer return figures either. That is consistency rather than transparency, but it is a materially different posture from quantifying a buyer's savings while withholding the cost.

Coverage is broad and, unusually, the audiences are named rather than implied: health systems, clinical programmes and service lines, professional societies, physicians and clinical editors, and education and training programmes, the last of which supports the workforce and training cross listing. Content types span clinical pathways and ClinApps, guidelines and protocols, policies and regulatory content, clinical calculators and decision aids, and antimicrobial stewardship with antibiograms. THE ANTIBIOGRAM CASE IS THE STRONGEST FIT ARGUMENT ON THE RECORD AND IS WORTH REUSING: an antibiogram is by definition institution specific, derived from that hospital's own local resistance patterns, so it cannot come from a national content library and must be authored and governed locally. That is the clearest illustration of why local pathway tooling exists alongside published guideline products. The professional societies segment is strategically notable, offering societies a route to translate guidelines into deployable pathways rather than publications, which places Curbside in the same structural position as the C8 Health collaboration with the American Society of Anesthesiologists. Held at B because breadth is claimed across every setting while named evidence sits in essentially one health system plus a paediatric concentration.