Most avoidable hospitalizations do not begin in the hospital. They begin with small signals that appear days or weeks earlier.
A missed follow-up appointment. A skipped medication refill. A canceled home care visit. A caregiver calling more often. A fall that does not lead to admission but changes confidence. A person with complex needs becoming less stable at home. Each signal may seem manageable in isolation. Together, they may indicate that someone is moving toward crisis.
This is where artificial intelligence may become one of the most important future tools in community-based care. Not as a replacement for care coordinators, nurses, case managers, direct support professionals, or family caregivers, but as a system-level assistant that helps them identify risk earlier. This sits directly across the Data, Insight & Performance Intelligence Knowledge Hub, the Innovation, Pilots & Emerging Models Knowledge Hub, and the Leadership, Governance & Organizational Capability Knowledge Hub.
The real opportunity is not AI writing care plans or producing polished reports. The deeper opportunity is AI helping providers, health plans, managed care organizations, and HCBS systems detect deterioration before emergency department use, hospitalization, or placement breakdown occurs.
Why Care Coordination Is Becoming Too Complex for Manual Systems Alone
Care coordination in community-based care has become increasingly complex. A person receiving HCBS may be supported by primary care, specialists, home health, behavioral health, LTSS providers, transportation services, pharmacy providers, family caregivers, housing partners, and managed care teams. Each organization may hold part of the picture, but rarely does one person see the whole risk pattern in real time.
This creates a major challenge for primary care and care coordination, health and social care coordination, and closed-loop care coordination and data exchange.
A care coordinator may know that a participant missed a visit. A nurse may know there has been medication instability. A caregiver may report that the person is more confused. A claims system may show increased emergency department use. A home care provider may see more canceled shifts. Individually, these are operational details. Together, they may point to avoidable hospitalization risk.
AI could help connect those signals faster than manual review allows.
The AI Care Coordinator: What the Role Could Actually Mean
The phrase “AI care coordinator” should be used carefully. AI should not make clinical decisions, ration care, replace human relationships, or determine eligibility without oversight. But AI could support the care coordination function by scanning large volumes of data and highlighting people whose risk profile is changing.
In practice, AI could monitor:
- missed or shortened visits;
- recent hospital discharge;
- medication refill delays;
- falls, wounds, infections, or functional decline;
- behavioral health escalation;
- transportation failure;
- caregiver stress signals;
- housing instability;
- repeat emergency department use;
- unresolved referrals or incomplete follow-up.
This links closely to AI and automation in care, technology-enabled care, interoperability and data exchange workflows, and data collection and data quality.
The goal is not prediction for its own sake. The goal is earlier human action.
Operational Example: Preventing a Hospital Admission Before It Happens
Consider an 82-year-old Medicaid HCBS participant living at home with congestive heart failure, diabetes, mild cognitive impairment, and limited caregiver support. She has recently been discharged from the hospital. Her care plan includes home care visits, medication monitoring, primary care follow-up, and caregiver support.
Over a ten-day period, several small changes occur. A home care visit is canceled. Her follow-up appointment is missed because transportation fails. Pharmacy data shows a delay in medication refill. A caregiver calls the care coordinator twice in one week, saying she “does not seem herself.” No single event triggers an emergency response.
An AI-supported system detects that the pattern resembles previous cases that led to hospitalization. It increases the participant’s risk score and alerts the care coordination team.
The care coordinator reviews the alert. A nurse contacts the participant. Transportation is rearranged. Medication reconciliation is completed. The caregiver receives additional support. A primary care appointment is rescheduled within 48 hours. The participant remains safely at home.
Required fields must include: participant identifier, risk trigger, data source, alert date, assigned reviewer, intervention action, follow-up timescale, outcome status, and escalation decision.
Cannot proceed without: human review of the AI alert, confirmation that data sources are current, documented contact attempt, clinical or care coordination judgement, and clear accountability for the next action.
Auditable validation must confirm: the alert was reviewed, the decision was recorded, the intervention was completed, and the outcome was evaluated against hospitalization, emergency department use, participant stability, and caregiver strain.
Why Managed Care Organizations Will Be Interested
This type of AI-supported care coordination is highly relevant to managed care organizations, state Medicaid agencies, health plans, and value-based provider networks. Avoidable hospitalization is not only a quality issue. It is also a cost, access, capacity, and system performance issue.
For MCOs and funders, AI-supported prediction could support value-based payment design, funding and payment models, avoidable utilization governance, and contract management and provider performance.
If used well, AI could help answer questions such as:
- Which participants are most at risk of avoidable hospital use?
- Which provider networks are identifying risk early?
- Where are care coordination actions delayed?
- Which interventions reduce emergency department use?
- Which populations are underserved by current outreach models?
- Where does caregiver strain predict crisis?
This does not mean using AI to reduce access to care. The ethical purpose should be the opposite: identifying need earlier so people receive the right support before crisis occurs.
The Next Generation Risk Dashboard
AI-enabled care coordination would need to be visible through practical dashboards, not hidden inside technical systems. A useful dashboard would show risk clearly enough for operational teams, supervisors, quality leaders, and executives to act.
A next-generation dashboard might include:
- hospitalization risk by population group;
- recent risk score increases;
- caregiver strain indicators;
- missed service patterns;
- medication-related risk;
- housing instability flags;
- unresolved referrals;
- recent discharge follow-up status;
- repeat-crisis utilizer patterns;
- intervention completion rates.
This connects to assurance dashboards and metrics, dashboard operating rhythm and performance cadence, outcomes frameworks and indicators, and evidence packs for funders and regulators.
The dashboard should not simply show who is “high risk.” It should show whether action is being taken, whether interventions are completed, and whether outcomes improve.
Operational Example: Using AI to Prevent Repeat Crisis Use
A behavioral health provider identifies that a small group of people repeatedly cycle between crisis stabilization, emergency departments, short inpatient stays, and unsupported return to the community. Staff know the pattern exists, but the organization struggles to predict who is most likely to bounce back into crisis.
An AI model reviews crisis history, appointment attendance, housing instability, medication disruption, peer support engagement, emergency contacts, and prior stabilization outcomes. It flags people whose current pattern resembles previous repeat-crisis users.
A multidisciplinary review is triggered. The provider adds peer support, confirms medication access, checks housing risk, reviews safety planning, and schedules proactive follow-up after discharge.
Required fields must include: crisis episode date, prior utilization history, current risk factors, assigned reviewer, stabilization plan, follow-up contact schedule, and escalation threshold.
Cannot proceed without: clinical review, person-centered planning, consent-aware information sharing, documented crisis pathway, and confirmation of community follow-up.
Auditable validation must confirm: whether the person avoided repeat crisis use, whether supports were delivered as planned, and whether learning was added to the organization’s crisis prevention model.
This supports repeat-crisis utilizer prevention, crisis diversion governance, and mental health crisis response and continuity.
Ethical Risks: Prediction Must Not Become Exclusion
AI risk prediction can create serious ethical problems if poorly governed. A person being labeled as “high risk” may experience more monitoring, more restrictive service planning, or reduced autonomy if systems are not carefully designed.
Providers, MCOs, and system leaders must ensure that predictive care coordination supports better care, not automated control.
Strong governance should address:
- algorithmic bias;
- transparency for participants and families;
- human review of AI recommendations;
- privacy and confidentiality;
- appeal and correction routes;
- data accuracy;
- equitable access to interventions;
- clear accountability for decisions.
This is why AI-enabled care coordination must be linked to trust, transparency and ethical data use, privacy-by-design and risk mitigation practices, data governance and information accountability, and ethics, integrity and public trust.
AI should recommend. Humans must decide.
Operational Example: Reducing Hospital Readmission After Discharge
A community-based provider works with adults leaving the hospital who need short-term support, home-based services, medication follow-up, and family coordination. Historically, some people return to hospital within 30 days because discharge risks are not resolved quickly enough.
An AI-supported discharge model reviews hospital discharge data, prior admissions, medication complexity, home care capacity, caregiver availability, mobility concerns, transportation barriers, and follow-up completion. It flags people whose discharge plan is incomplete or unstable.
The provider uses the alert to prioritize the first 72 hours after discharge. A nurse reviews medication reconciliation. A care coordinator confirms follow-up appointments. A home care supervisor checks visit capacity. A family navigator confirms caregiver concerns.
Required fields must include: discharge date, primary risk factors, medication reconciliation status, home visit confirmation, follow-up appointment status, caregiver contact, and readmission risk review.
Cannot proceed without: confirmed handover information, assigned care coordinator, documented escalation route, and review of immediate safety risks.
Auditable validation must confirm: whether the person remained at home, whether follow-up occurred within target timescales, whether medication risks were resolved, and whether the discharge pathway reduced avoidable readmission.
This aligns with hospital discharge and transitional care, post-acute care interfaces, and referral management and closed-loop follow-up.
What Providers Should Do Before AI Is Added
AI will not fix poor operating systems. If records are incomplete, referrals are not closed, risk categories are inconsistent, and outcomes are poorly defined, AI will simply process weak data faster.
Before implementing AI-supported care coordination, organizations should strengthen data quality standards, referral workflows, risk stratification definitions, outcome measurement frameworks, care coordination accountability, and governance oversight.
The most successful organizations will not be those that purchase the most sophisticated AI platform. They will be those that create the strongest operational discipline around how AI insights are reviewed, challenged, escalated, and acted upon.
What Community-Based Care Could Look Like by 2035
By 2035, predictive care coordination could become a standard capability across high-performing HCBS, LTSS, behavioral health, and community care systems.
Organizations may no longer wait for:
- hospital admissions;
- caregiver collapse;
- housing breakdown;
- behavioral health crisis;
- emergency department utilization;
- service failure;
- avoidable institutional placement.
Instead, systems may identify emerging risk while there is still time to intervene.
This future aligns closely with preventative value and early intervention, avoided costs and demand reduction, value-based care innovation, and long-term system impact.
Conclusion: AI Will Not Replace Care Coordinators, But It May Transform Care Coordination
Artificial intelligence is unlikely to replace care coordinators, nurses, case managers, clinicians, direct support professionals, or family caregivers. The human elements of trust, judgment, advocacy, empathy, and relationship-building remain fundamental to effective care.
However, AI could fundamentally change the information available to those professionals. It could help organizations identify risk earlier, prioritize resources more effectively, reduce avoidable hospitalizations, strengthen value-based care strategies, and improve outcomes across complex community-based systems.
The future of AI in HCBS may not be about replacing people. It may be about helping people see patterns, risks, and opportunities that are currently hidden inside fragmented systems.
If that happens, the AI Care Coordinator will not be a machine replacing human care. It will be a new layer of intelligence helping community-based care systems intervene before crisis becomes inevitable.