Using AI To Identify Rising-Risk Patients Before They Deteriorate

December 12, 2025

By Jay Chowdappa, MD, co-founder and CEO, Zynix AI · Updated November 18, 2026

Key takeaways

Rising-risk patients are those whose labs, visits, prescriptions or emergency use are trending the wrong way before they become high cost. AI finds them by tracking those changes across clinical, claims and pharmacy data and estimating readmission, disease progression, utilization and, for clinician review, mortality risk. The value comes only when each flag ends in outreach and a booked visit.

  • Rising-risk patients are not yet the sickest, which is why earlier outreach can still change their course.
  • Static risk scores built on last year's claims miss trends that only show up in recent labs, prescriptions and visit history.
  • Dynamic risk models combine clinical, claims, pharmacy and social needs data and refresh as new data arrives.
  • Detection has to trigger a specific next step with an owner, such as an outreach call, a booked visit or a care manager review.
  • Agents handle outreach, scheduling and reminders, while clinicians handle symptoms, medications and care plan decisions.

The Rising-Risk Population

In value-based care, the highest-cost patients often aren’t the ones who are already critically ill — they’re the ones who are about to become critically ill. These rising-risk patients represent the greatest opportunity for intervention, but traditional care models lack the tools to identify them reliably before deterioration occurs.

Beyond Static Risk Scores

Traditional risk stratification assigns patients a score based on historical data — diagnoses, utilization patterns, demographics. But these static scores miss the dynamic signals that indicate a patient is trending toward a crisis. A diabetic patient whose HbA1c is slowly climbing, who has missed two recent appointments, and who just filled a prescription for a corticosteroid is flashing warning signs that a static risk score won’t capture.

AI-Powered Dynamic Risk Detection

AI models can continuously analyze streams of clinical, claims, and behavioral data to detect these emerging risk patterns. By processing lab trends, medication changes, appointment adherence, ED utilization, and social determinants of health, AI creates a dynamic risk picture that updates as new data arrives.

From Detection to Action

Identifying rising-risk patients is only valuable if it triggers timely intervention. Zynix AI’s rising-risk outreach agent automatically initiates outreach to identified patients, scheduling follow-up appointments and routing new symptoms or barriers to the care team under the care plan’s escalation rules, so already-overburdened care managers spend their time on the patients who need them.

The Value-Based Care Impact

Organizations that implement AI-powered rising-risk detection aim for reductions in avoidable hospitalizations, emergency department visits, and total cost of care. More importantly, they aim for improvements in patient outcomes that define success in value-based care.

Signals that mark a patient as rising risk

Rising risk shows up as a change in trajectory rather than a single high score. The most useful signals are the ones that move between visits and can be acted on this month.

  • Clinical trends: lab values drifting the wrong way, such as a climbing HbA1c or falling kidney function, and new diagnoses that change a care plan.
  • Utilization: a recent emergency department visit, an observation stay or a hospital discharge reported through ADT feeds.
  • Medications: new high-risk prescriptions, gaps in refills, or several new prescribers in a short period.
  • Engagement: missed appointments, an overdue annual wellness visit, or no primary care contact for an extended period.
  • Social needs: transportation, food, housing or cost barriers captured during outreach or screening.

What a rising-risk model should predict

A rising-risk model is useful when it predicts events a care team can act on this month. For most ACOs those are four: readmission, disease progression, utilization such as emergency visits, and mortality.

The first three can drive outreach, a booked visit or care-management enrollment. Mortality risk is different: it goes to the patient’s clinician, who decides whether a serious-illness conversation, advance care planning or a palliative-care review is appropriate. No agent contacts a patient about prognosis.

Example: one rising-risk flag, from signal to visit

Example (fictional): a 70-year-old ACO patient with type 2 diabetes and heart failure. Over one month her latest HbA1c is higher than the one before, she misses a primary care visit, an ADT feed reports an emergency department visit for shortness of breath, and a new prescriber adds a medication. No single signal puts her in the high-risk program. Together they change her trajectory.

  1. Her disease-progression and utilization risk rises, and she lands on this week’s rising-risk worklist with a named care manager as owner.
  2. An agent calls within the hours the team has set, confirms she is home, asks about barriers such as transportation and books a primary care visit.
  3. She mentions new swelling in her legs. The agent gives no advice and routes the call to the care manager by rule.
  4. The care manager calls the same day, asks a pharmacist to review the new medication against her list, and moves the visit sooner.
  5. At the visit, her physician reviews the trend and updates the care plan. The worklist item closes with the visit documented.
  6. The agent runs the follow-up check-ins on the care plan’s schedule, and the team watches whether her signals improve.

Every step has an owner and a time, and every clinical judgment sits with a clinician. That is what turns a score into care.

Who acts on a rising-risk flag

A flag is only useful when someone owns the next step. The work splits cleanly between what agents can do at scale and what needs a clinician.

Agents can reach out, run a structured check-in, book a visit, send reminders and capture barriers such as transportation. New symptoms, medication concerns and anything else that needs clinical judgment go to a care manager or the on-call clinician by rule, and agents never diagnose, triage clinically or recommend treatment. Care plans for chronic and preventive care keep each patient's tasks in sequence, so a flag ends in a completed visit or a documented reason it did not happen.

Frequently asked questions

What is a rising-risk patient?

A rising-risk patient is someone whose health and costs are trending upward but who is not yet among the highest-risk, highest-cost patients. Typical signs include worsening lab values, new chronic diagnoses, missed appointments, recent emergency visits and social barriers. Because they have not deteriorated yet, timely outreach and follow-up can still change their trajectory, which makes them a priority population in value-based care.

How does AI identify rising-risk patients?

It combines data a single report rarely brings together: claims history, EHR diagnoses and lab results, ADT notifications of emergency visits and admissions, pharmacy fills, appointment history and social needs screening. Models look for changes over time rather than a fixed score, and rank patients so care teams see who needs attention first. Risk and care gap analytics that feed worklists are one way to put this into practice.

Which risks should a rising-risk model predict?

Focus on events the care team can still change: hospital readmission, disease progression, utilization such as emergency visits and admissions, and mortality. Each prediction needs a defined next step. Readmission and utilization risk usually lead to outreach and a booked visit, progression risk to a clinician’s review, and mortality risk to a clinician-led serious-illness conversation. Test any model on your own population before acting on it.

What should happen after a patient is flagged as rising risk?

The flag should create a task with an owner and a deadline. Usually that means outreach to check in and book a primary care or care management visit, a clinician’s review of medications and recent utilization, and help with barriers such as transportation. Track whether the visit happened and whether the patient’s signals improved, not just whether a call was made.

How is rising-risk work different from high-risk care management?

High-risk care management focuses on a small group of patients who already have frequent admissions and complex needs, usually with intensive nurse involvement. Rising-risk work covers a larger group earlier in their trajectory, so it relies on lighter-touch, scalable steps: regular check-ins, timely visits, medication follow-up and help with barriers, with escalation to a care manager when a patient’s signals worsen.

Related reading

About the author

Jay Chowdappa, MD is co-founder and CEO of Zynix AI. He is a physician, and ACOs led by him generated $300M+ in shared savings before he started Zynix AI to help care teams follow through on what their data already shows.

Explore More Insights

Visit Our Blog