AI Risk Stratification Software Compared: 11 Tools for ACOs and Health Plans (2026)

By Gautamdev Chowdary, co-founder and CTO, Zynix AI · Updated November 7, 2026

Disclosure: Zynix AI makes one of the platforms in this guide. Every description of another vendor comes from that vendor’s own website as of November 7, 2026, linked in the source notes. This is not a ranking: products change, so confirm current capabilities with each vendor.

Key takeaways

Risk stratification software sorts a population by who is most likely to need care soon: an admission, a readmission, an ED visit, rising cost or a declining chronic condition. The tools differ in two ways that matter more than any accuracy claim: how the score is built, and whether anything happens after the score is produced.

• Scoring engines such as the Johns Hopkins ACG System, Milliman MARA, Cotiviti’s DxCG Intelligence and Merative’s Truven analytics produce risk scores that other tools and teams act on.
• Healthy (formerly ClosedLoop) is a machine-learning platform for building and explaining your own predictive models.
• Lightbeam, Health Catalyst, Innovaccer, Arcadia and Elligint Health put risk scores into population health or care management workflows.
• Zynix AI predicts readmission, mortality, disease progression and utilization, shows the drivers behind each prediction, and turns each one into a next step that its care management platform and voice and text agents carry out.

What does risk stratification software do?

Risk stratification is now part of Medicare billing as well as good practice. To bill Advanced Primary Care Management services, CMS requires practices to “use defined diagnoses, claims, or other electronic data to group your patient population by risk to identify which patients need specific services.” An earlier AHRQ brief on care management made the same point: “it is critical that providers understand which patients are likely to benefit from more intensive CM.”

Most risk stratification tools do some of four things:

• Score: assign each patient a concurrent or prospective risk score from claims, clinical and sometimes social data.
• Predict specific events: estimate the chance of an admission, readmission, ED visit, rising cost or disease progression.
• Explain: show the factors behind each score so a clinician can check it.
• Act: turn high and rising risk into care management enrollment, outreach or a scheduled visit.

Scores are not outcomes

A list of high-risk patients changes nothing until someone reaches those patients. Before comparing models, decide who will act on the list and how fast. If your care team already has a full caseload, a better score adds names to a list nobody works. In that case, weigh how each tool connects to care management and outreach as heavily as how it builds the score. For the workflow side, see using AI to identify rising-risk patients early.

Risk stratification software at a glance

Each row summarizes the vendor’s own description as of November 7, 2026. “Method” uses the vendor’s own terms; where a vendor does not say how its models are built, the table does not guess. The last column is our reading of who each may suit, not a ranking.

The vendors, one by one

Johns Hopkins ACG System

The ACG System was developed at the Johns Hopkins Bloomberg School of Public Health and is built on the premise that “clustering of morbidity is a better predictor of health care services resource use than the presence of specific diseases.” Using “diagnosis, pharmacy and lab data,” it can “predict patients at risk for future hospitalization, readmission, high cost and high pharmacy use” and stratify a population by patterns of disease and resource use. It is licensed software, including an integrator license for vendors that build it into their own applications; Lightbeam, for example, lists it in its risk engine.

Milliman MARA

Milliman, founded in 1947, sells Milliman Advanced Risk Adjusters (MARA), claims-based models that offer “concurrent and prospective risk scores” and predict annual spending by service category. Its Rising Risk models group members “by whether their cost will stay the same, decrease, or rise,” and its models can “identify risk of emergency department visit or hospital admission.” MARA v5 adds social categories from coded Z codes. Results load into your own reporting or Power BI templates, with Milliman consulting as an option. It serves health plans, payers, TPAs, providers and ACOs, with models for Commercial, Medicaid, Medicare Advantage and MSSP populations.

Cotiviti DxCG Intelligence

DxCG Intelligence is part of Cotiviti’s risk adjustment business and “has utilized Cotiviti’s proprietary predictive models to transform healthcare data into risk scores for individual members.” Its models are grouped into budgeting and underwriting, medical management and performance assessment bundles for commercial, Medicare and Medicaid populations, and the latest version adds social determinants of health factors. Listed uses include “stratify clinical groups to define outreach strategies” and “prioritize individuals for targeted interventions.” It is built for health plans. Cotiviti’s HCC coding services are covered in our risk adjustment software comparison.

Merative (Truven Flexible Analytics)

Merative’s Truven Flexible Analytics is claims analytics delivered as a service, hosted on Microsoft Azure. Its predictive model list includes risk of hospitalization, risk of rising cost, risk of emergency department utilization and HCC Medicare models, alongside episode groupers, disease staging and population classification. It supports care management, financial analysis and provider performance work for health plans, employers, providers and government buyers.

Healthy (formerly ClosedLoop)

Healthy, headquartered in Austin, Texas, offers “accurate, explainable, and actionable predictions of individual-level health risks.” It is a machine-learning platform: customers choose algorithms such as XGBoost, random forest and elastic net, get bias checks and drift monitoring, and see “auditable individual-level predictions and contributing factors.” Its ACO-Predict models cover unplanned admissions, readmissions, unplanned dialysis initiation, end of life and chronic disease progression, and can use CMS BCDA claims feeds. Outreach is done by the customer in its own systems.

Lightbeam Health Solutions

Lightbeam’s risk engine combines the Johns Hopkins ACG System, NYU Avoidable ED algorithm, Charlson Comorbidity Index and CMS/HHS HCC models with its own predictive and prescriptive models and an engageability score. Its stated aim is to “predict risk, prioritize rising-risk, and automate prescriptive workflows.” It has described models for avoidable admissions, mortality within a given timeframe for end-of-life support, and likely dual eligibility. Scores feed its TCM, CCM and rising-risk management workflows and patient engagement, for ACOs, MSOs, providers and payers.

Health Catalyst

Health Catalyst offers “out-of-the-box models for readmits, ED use, risk, and more,” with an explicit focus on explainability: “Prioritize interpretability with clear explanations and inputs,” plus bias detection and model fairness audits. Customers can configure prebuilt models or build their own, and “integrate scores into EHRs, dashboards, and workflows.” Its VBC Intelligence product segments attributed populations “to identify high-, rising-, and impactable-risk populations.”

Innovaccer

Innovaccer’s population health analytics use predictive models to “identify patients likely to incur higher future healthcare costs” and say they combine “clinical, claims, and SDoH data” for risk stratification, refined with machine learning on the customer’s data. Its readmission management offering identifies high-risk patients for readmission prevention and automates post-discharge follow-up. It fits best where Innovaccer already holds the data. See Zynix AI vs Innovaccer.

Arcadia

Arcadia places AI stratification inside its Ascent platform, where its stated approach is to “predict patient needs and prioritize proactive outreach.” Its Ascent Care module adds a care navigation agent for prioritizing patients and automating enrollment, tasks and follow-up, and its Insight module lets teams deploy predictive models into workflows. Arcadia’s governance principles include explainability, so users can understand why an action was recommended. The pages we read do not name the specific outcomes its models predict.

Elligint Health (formerly HealthEC)

HealthEC and VirtualHealth combined to form Elligint Health, and healthec.com now redirects to it. Its Advanced Analytics provides “Population Risk Stratification to identify rising-risk, high-cost, and high-opportunity member populations,” and its Helios care and utilization management platform turns the findings into care team work. It also says “Helios integrates with the risk modeling solutions you already use,” so it can act on another vendor’s scores. It serves health plans, ACOs, clinically integrated networks and MSOs.

Zynix AI

Zynix AI’s Predictive Risk makes four predictions: readmission, mortality, disease progression and utilization. Each is built from the patient’s record across claims, EHR, ADT, lab and pharmacy data, and every prediction lists the drivers behind it so a clinician can check it against what they know. Mortality predictions only flag patients for a clinician-led serious-illness conversation, advance care planning and palliative-care review; nothing happens for the patient until a clinician has reviewed the flag.

Each prediction comes with a next step. A readmission risk becomes a post-discharge care plan and outreach, disease progression becomes chronic care management enrollment, and rising utilization becomes outreach to rising-risk patients before an avoidable ED visit. The embedded care management platform assigns each step an owner and a deadline, and voice and text agents reach the patient and book the visit. Analytics tells you who. Zynix makes sure something happens.

How to choose risk stratification software

• You need actuarial scores for contracts, budgets or benchmarks: compare scoring engines such as the ACG System, MARA, DxCG and Merative, and ask which populations each model was built for.
• You have a data science team that wants its own models: look at Healthy and Health Catalyst, and ask how explanations, bias checks and drift monitoring work.
• You already run a population health platform: check the stratification in Lightbeam, Innovaccer or Arcadia before buying a separate model.
• Your problem is that nobody works the list: choose a tool that turns each prediction into a task and outreach, such as Lightbeam, Elligint Health or Zynix AI.

For any vendor, ask what each model predicts and over what time window, which data it needs, how drivers are shown to clinicians, how the model is checked on your own population, and what happens the day a patient crosses a threshold. For background on the method, see AI in value-based care: risk stratification and care gaps.

Frequently asked questions

What is the best risk stratification software?

There is no single best tool; the right one depends on what you will do with the score. Health plans that need actuarial scores often compare the Johns Hopkins ACG System, Milliman MARA and Cotiviti DxCG. Teams that build their own models look at Healthy and Health Catalyst. ACOs and MSOs that need scores turned into care management and outreach compare Lightbeam, Innovaccer, Arcadia, Elligint Health and Zynix AI.

What is the difference between risk stratification and risk adjustment?

Risk adjustment uses documented diagnoses to set payments and benchmarks, as in the CMS-HCC model. Risk stratification ranks patients by the chance of a future event, such as an admission or rising cost, so a care team knows whom to reach first. The two share data but answer different questions.

Is AI risk stratification better than traditional models?

Vendors on both sides publish their own comparisons, and results depend on your population and data. Grouper and actuarial models are long established and easy to compare across organizations; machine-learning models can use more data types and be trained on your population. Ask any vendor to validate its model on your own data and to show the drivers behind each prediction.

What data do risk stratification tools need?

Most start with claims: diagnoses, procedures and pharmacy fills. Many add EHR data such as labs and vitals, ADT feeds for admissions and discharges, and social determinants of health. Claims-only models are available sooner for health plans; clinical and ADT data help provider organizations act earlier.

Do Medicare practices have to risk-stratify patients?

To bill Advanced Primary Care Management services, CMS requires practices to use diagnoses, claims or other electronic data to group their patient population by risk and identify which patients need specific services. ACOs also stratify to decide where care management effort goes.

Related reading

• Using AI To Identify Rising-Risk Patients Before They Deteriorate
• How AI Transforms Value-Based Care: Population Health, Risk Stratification, and Care Gap Closure
• Population Health Management Software Compared: 8 Platforms for Value-Based Care (2026)

Source notes

Centers for Medicare & Medicaid Services: Advanced Primary Care Management services. https://www.cms.gov/medicare/payment/fee-schedules/physician-fee-schedule/advanced-primary-care-management-services

Agency for Healthcare Research and Quality: Care Management: Implications for Medical Practice, Health Policy, and Health Services Research (2015). https://www.ahrq.gov/ncepcr/care/coordination/mgmt.html

Johns Hopkins ACG System: Applications. https://www.hopkinsacg.org/applications/

Johns Hopkins ACG System: About the ACG System. https://www.hopkinsacg.org/about-the-acg-system/

Johns Hopkins ACG System: License types. https://www.hopkinsacg.org/license-types/

Milliman: Milliman Advanced Risk Adjusters (MARA). https://www.milliman.com/en/products/mara

Milliman: MARA v5. https://www.milliman.com/en/insight/unlock-superior-population-health-insights-mara-v5

Cotiviti: DxCG Intelligence. https://www.cotiviti.com/solutions/risk-adjustment/dxcg-intelligence

Merative: Truven Flexible Analytics. https://www.merative.com/healthcare-analytics/flexible-analytics

Healthy: Home page. https://www.gethealthy.com/

Healthy: ACO-Predict. https://www.gethealthy.com/aco-predict/

Healthy: Healthy Platform. https://www.gethealthy.com/healthy-platform/

Lightbeam Health Solutions: AI-driven risk stratification. https://lightbeamhealth.com/ai-risk-stratification/

Lightbeam Health Solutions: Avoidable admissions model announcement (March 2025). https://lightbeamhealth.com/news/lightbeam-ai-model-enables-average-41-relative-reduction-in-avoidable-admissions/

Health Catalyst: Machine learning and AI. https://www.healthcatalyst.com/healthcare-analytics/data-analytics/machine-learning-ai

Health Catalyst: VBC Intelligence. https://www.healthcatalyst.com/products/vbc-intelligence

Innovaccer: Population health analytics. https://innovaccer.com/products/population-health-analytics

Innovaccer: Readmission management. https://innovaccer.com/solutions/readmission-management

Arcadia: Ascent, arcadia.io/ascent/, and AI philosophy and strategy, arcadia.io/ai-philosophy-strategy/ (page text retrieved November 7, 2026; the site blocks automated link checks).

Elligint Health: Value-based care. https://elliginthealth.com/value-based-care/

Elligint Health: Advanced Analytics. https://elliginthealth.com/analytics/

Elligint Health: HealthEC and VirtualHealth unite to form Elligint Health. https://elliginthealth.com/resources/news/virtualhealth-and-healthec-unite-to-form-elligint-health/

About the author

Gautamdev Chowdary is co-founder and CTO of Zynix AI, where he leads engineering for the Zynix platform, its AI agents and ZynixLLM.