COZEVA
COZEVA is a cloud platform for health plans and provider organizations that manage quality, risk adjustment and cost under value based contracts. It combines claims, clinical records and other data streams into one view of each patient and each contract, calculates HEDIS and other quality measures on NCQA certified, FHIR based digital measure engines and shows care teams which gaps remain open. For risk adjustment, a machine learning model working alongside clinical rules suggests conditions for review, natural language processing pulls supporting evidence from chart notes, and clinicians confirm or dismiss each suggestion. The PayerOne suite gives health plans their own quality and risk tools, and partner products serve life sciences, research and public health.
CozevaConnect brings the same information into the record. In Epic it runs as an embedded SMART on FHIR app that reads encounter data and writes confirmed conditions to the problem list or visit diagnoses. It also works through single sign on, automated chart retrieval before and after visits, and overlays for Elation and Office Ally that close quality gaps from chart data and create review tasks for suspected conditions.
The platform covers more than 25 million lives, 97,000 providers and 84 clients across 30 states, including Blue Shield of California, the Integrated Healthcare Association, Western Health Advantage, Providence, AltaMed and Elevance. It won the 2026 Best in KLAS award for risk adjustment at the point of care and in home assessments with a score of 91.2, and it holds HITRUST certification.
COZEVA is built by Applied Research Works of San Ramon, California, led by chief executive Khanh Nguyen. Frazier backed Vatica Health closed its merger with the company in October 2025, adding clinical services to the combined business. Subscriptions are quoted, with monthly fees per member or access funded by a health plan.
Capability Axes
An AI Health Index grade measures what a buyer can verify from public sources on the date shown. It is not a rating of how good the product is. A vendor can build an excellent system and grade low on an axis because it publishes nothing an outsider can check. How grades read
Learned components carry the risk adjustment work. A machine learning model identifies suspected conditions for review, working alongside configured clinical rules, and natural language processing extracts supporting evidence from unstructured chart notes, so a coder or clinician sees why a condition was suggested. The same analysis feeds the HCC review tasks created from retrieved charts.
Much of the platform runs without models. HEDIS and other quality measures are calculated from published specifications on the digital measure engines, and population reports, incentive programs and data exchange rely on conventional analytics and transaction processing. The learned layer improves capture of documented conditions inside a broader measurement system.
The risk workflow surfaces suspected conditions with their supporting data, and in the Epic app clinicians confirm or dismiss each one, with only confirmed conditions written to the problem list or visit diagnoses. Provider terms assign responsibility for checking each patient match before information is used, and reporting and campaigns run as configured.
Other integrations automate more. The Elation overlay closes quality gaps automatically from retrieved chart data and describes HCC diagnosis auto confirmation alongside HCC review tasks. Which suggestions confirm without a clinician and which route to a person are not set out, so the line between the two depends on the integration in use.
A machine learning model identifies suspected conditions, supplemented by clinical rules, and natural language processing extracts information from records, with claims and clinical data named as inputs and the points where results reach staff shown. PayerOne Risk supports the Medicare Advantage V28 model, which describes a payment method rather than the internal design of the learned model. The quality side runs on native FHIR digital measure engines certified by NCQA.
Model designs by version, training data provenance and validation thresholds for each suspected condition are not given. The provider terms reserve source code, weights and training data from customer access.
Applied Research Works builds the platform and deploys its own risk and chart extraction models. Exchange partners and integration environments are named, including Epic, Elation and Office Ally, and separate terms cover external suppliers. Vatica Health brings clinical services to the combined business.
No provider or version is listed for each learned component, and no outside language model host, training data license or inference environment is identified for any feature. The roles of the risk engine, health plan data inputs and record interfaces are clear, but they do not add up to an inventory of the model dependencies behind a suggestion.
Blue Shield of California brought COZEVA into a Digital Health Record expansion covering more than 1.5 million members, and the Integrated Healthcare Association chose it as sole analytics partner for measurement year 2025. Providence St. Joseph Health reports earning 15 quality awards after starting on the platform in 2016, and the company reports retention above 95% across 84 clients. KLAS scored it 91.2 in its 2026 risk adjustment segment from customer interviews.
An earlier University of California, Berkeley case study linked higher platform use to better quality process measures but predates the current models. No independent estimate covers suggestion accuracy or outcomes attributable to the learned features.
The provider interface terms require protected information to be handled under the applicable business associate agreement, and that agreement sets safeguards, disclosure controls and return or destruction of data with stated exceptions. The privacy policy allows deidentification or aggregation for further uses under its terms, the chart workflow shows who can confirm a suspected condition before it reaches the record, and the platform holds HITRUST certification.
No rule bars training on identifiable or deidentified inputs across every model and partner. Retention is not set out for each AI service, including the charts retrieved automatically before and after visits.
Applied Research Works is the named business associate in a public agreement incorporated into the user license for covered entities, and the interface terms apply it to provider data that is protected health information. The agreement binds subcontractors that create, receive, maintain or transmit that information on the company's behalf to equivalent restrictions.
Provider access may be paid for by a subscribing organization or arranged directly under a signed statement of work, and the agreement follows the covered relationship either way. No separate fee attaches to it. Cozeva remains an Applied Research Works company after the Vatica Health merger.
DirectTrust lists Applied Research Works, doing business as COZEVA, under HITRUST certification HT0138030726, expiring in July 2028, and the Integrated Healthcare Association describes the platform as HITRUST CSF r2 certified after an earlier interim assessment. The company also holds ONC health IT certification, and NCQA validates its data streams and certifies its measure engines.
The registry names the organization and credential without the system or supplier boundaries for the deployed models and infrastructure, and no full certificate package with that scope is open to read. The NCQA work certifies quality data flows rather than information security.
The platform works in quality reporting, population management and risk adjustment. Provider terms assign responsibility for diagnosis and risk adjustment codes to the provider and require verification before acting, and the risk product follows CMS requirements including the V28 model, with risk adjustment methods built to comply with proposed CMS policy changes. NCQA certifies the measure engines and data streams.
Those standards describe reporting and coverage operations. Suggested conditions reach clinicians at the point of care for verification, and no FDA authorization or reasoned exclusion analysis covers every suggestion generated there, though the outputs concern coding and documentation rather than treatment.
Providers see how a suspected condition was identified and the data behind it before they accept or dismiss it, and quality and risk reporting includes equity views and performance across populations. The terms give audit and monitoring rights and assign verification of each patient match to the provider, and NCQA validates the quality data streams and certifies the measure engines.
Those checks cover data feeds and individual results. No calibration of the risk model across patient subgroups, care settings or insurers is reported, and no fairness review of the suggestions themselves appears, which matters as the platform's risk work reaches more than 25 million lives.
Clinicians can dismiss a suggested condition and check its supporting data before accepting it. The provider interface terms anticipate patient identity mismatches and assign verification of the match to the provider, disclaim accuracy warranties and any promise that all errors will be corrected, and allow a remedy for uncured material breach in certain situations.
No correction deadline applies to a wrong suggestion, and nothing pays a provider back when a suggested code fails a risk adjustment audit. Where an integration closes quality gaps automatically from chart data, an error surfaces only in later review, and operational review stays the main way to catch mistakes.
The SMART on FHIR app lists Epic support on FHIR DSTU2 and R4, opens inside the chart with the patient in context, reads encounter data through a backend FHIR interface and writes confirmed conditions to the problem list or visit diagnoses. The integration joined Epic's App Orchard in 2021, and CozevaConnect also runs as an overlay for Elation and Office Ally and as a single sign on extension for most browser based EHRs, retrieving charts automatically before and after visits. Blue Shield of California runs data exchange with providers in production.
No named site shows both directions of the Epic confirmation and write back workflow in production, and depth varies by record system and contract.
COZEVA is a cloud platform drawing on data from health plans, provider systems and outside sources, and the provider agreement licenses use within the United States. That limit governs where a customer may use the interface, not where data is stored or processed. DirectTrust records outsourced functions performed offshore without naming the data or infrastructure involved.
No standard primary or backup data region is stated, there is no choice of region for the learned services, and there is no customer hosted option. A browser based chart connection and automated chart retrieval do not settle where the analytics platform keeps its copy of the record.
Provider interface terms set monthly subscriptions with possible charges per member per month, and they allow a health plan or other subscribing organization to pay for a provider's use. Amounts sit in a statement of work or attachment, outside expenses may be passed through, and unless an agreement says otherwise fees rise each year by the greater of 5% or the consumer price index. Epic integration came at no extra cost for existing clients when it launched in 2021.
That gives a billing unit and contract mechanics. No starting price is given for risk, quality, analytics or CozevaConnect, and no standard minimum volume is set.
Health plans, provider organizations, value based care networks and partners in life sciences, research and public health use the platform across 30 states. Blue Shield of California shares records between plan and providers, the Integrated Healthcare Association measures performance across many organizations, and Providence, AltaMed and Western Health Advantage run quality and risk programs on it. Workflows cover commercial, Medicare Advantage and Medicaid quality reporting and point of care risk review.
Coverage is organized around populations and contracts rather than a single disease or specialty. State and payer rules, the measure set and access to chart data decide which workflows a provider runs.
Alternatives to COZEVA
Other products that do the same kind of work as COZEVA, and what each one does.
Pelica provides AI agents for health plans and provider organizations managing quality, financial risk, pharmacy adherence, and care coordination.
Reveleer is a Glendale, California company founded in 2009 as a medical record retrieval business and since rebuilt as a value based care platform for health plans and risk bearing providers across Medicare Advantage, ACA Marketplace and Medicaid.
Inovalon is a data business that sells analytics and software on top of its data.
Astrata reads clinical charts so quality teams do not have to. The company builds natural language processing for healthcare quality measurement, aimed at the HEDIS reporting that determines health plan star ratings and quality bonus payments.
Arcadia sells the data foundation underneath value based care, and increasingly the models that run on top of it.
Kennar Health, operated by Toothpicker Holdings, Inc. of Hicksville, New York, sells what it calls an AI workforce for value-based care.
Pricing
Vendor-published figures are labeled as such. Figures labeled “Estimated” are derived from third-party sources and have not been confirmed by the vendor.
| Entry Price | Pricing Basis | BAA Tier | Implementation | Source |
|---|---|---|---|---|
|
Custom subscription quote
|
Provider interface terms allow monthly subscriptions and charges per member per month. An applicable statement of work sets the amounts, and another subscribing organization may cover provider use. | The public business associate agreement is incorporated into the license for covered entities and applies to relevant provider data. There is no separate public agreement price. | Interface setup and additional services can be specified in a signed agreement. The terms allow certain outside expenses to be passed through, and no standard setup charge is stated. | Vendor Published |
The agreement describes annual adjustments and allows access funded by a health plan in some arrangements. Initial rates and minimums for the modules are not stated.