By Sagar Shankaran, Founder of CallSphere
How chronic care management programs deploy AI voice agents to make adherence check-in calls for diabetes, hypertension, CHF, and COPD cohorts at scale.
Key takeaways
Medication non-adherence costs the U.S. healthcare system an estimated $500 billion per year in avoidable hospitalizations, complications, and premature deaths, according to the NEHI (Network for Excellence in Health Innovation) 2024 update. The single highest-impact, lowest-cost intervention proven to improve adherence is structured telephonic outreach — and it's also the intervention most difficult to staff at the scale chronic care management (CCM) programs require. AI voice agents solve the scale problem while preserving the clinical effectiveness.
BLUF: Chronic care management programs deploy AI voice agents to run monthly adherence check-ins for diabetes, hypertension, CHF, and COPD cohorts — the four chronic conditions that drive 60% of Medicare spend. Production deployments handle 10x the call volume of human-staffed CCM at similar or better PDC (Proportion of Days Covered) outcomes, billing CMS CCM codes 99490, 99487, and 99489 at proper cadence. Integrated pharmacy-coordinated refills cut primary non-adherence from 28% to 9% and MPR gaps from 22% to 11% in 12-month cohort studies.
This post is the CCM adherence operator's playbook: the PQA adherence measures that determine everything, the CPT code structure for billing, the CCM-RAMP framework we built, and the pharmacy-coordination patterns that connect voice agents to Surescripts, e-prescribing, and retail-pharmacy partner workflows.
BLUF: Medicare pays for chronic care management through a small but meaningful set of CPT codes — 99490 (basic CCM, 20 minutes), 99439 (add-on 20 minutes), 99487 (complex CCM, 60 minutes), 99489 (add-on complex), 99491 (physician-provided CCM), and the Principal Care Management (PCM) codes 99424-99427. Each requires documented patient consent, a care plan, and 24/7 access to care. AI voice agents can run the qualifying time under clinical supervision.
According to CMS's 2026 Physician Fee Schedule final rule, CCM reimbursement rates rose modestly and the Principal Care Management codes continue to expand. The financial model for a practice with 2,000 eligible patients can exceed $1.4M annually in CCM revenue — but only if the monthly touchpoint cadence is actually maintained.
| CPT Code | Service | Time Threshold | 2026 National Allowable (non-facility) |
|---|---|---|---|
| 99490 | CCM, clinical staff | First 20 min/month | ~$62.16 |
| 99439 | CCM add-on | Each add'l 20 min (max 2/mo) | ~$48.76 |
| 99487 | Complex CCM | First 60 min/month | ~$133.16 |
| 99489 | Complex CCM add-on | Each add'l 30 min (max 3/mo) | ~$69.76 |
| 99491 | Physician CCM | First 30 min/month | ~$86.48 |
| 99424 | PCM, physician | First 30 min/month | ~$82.23 |
| 99426 | PCM, clinical staff | First 30 min/month | ~$63.34 |
BLUF: Four chronic conditions drive the bulk of the adherence economics — Type 2 diabetes, hypertension, congestive heart failure (CHF), and COPD. Each has a specific PQA (Pharmacy Quality Alliance) adherence measure, each has a specific failure pattern, and each responds to a specific voice-agent intervention tree. Programs that segment by condition outperform generic "take your meds" outreach by 2-3x.
| Condition | PQA Measure | PDC Threshold | Typical Baseline | Post-AI Lift |
|---|---|---|---|---|
| Diabetes (oral) | PDC-DR | 80% | 68% | +9-14 pts |
| Hypertension (RAS) | PDC-RAS | 80% | 71% | +7-11 pts |
| Statins | PDC-Statins | 80% | 64% | +10-15 pts |
| CHF (beta-blocker + ACE/ARB) | MPR composite | 80% | 58% | +12-18 pts |
| COPD (LABA/LAMA) | PDC-COPD | 80% | 61% | +8-12 pts |
According to PQA's 2025 measurement framework, PDC >=80% is the quality threshold built into Medicare Part D Star Ratings, ACO quality scoring, and most commercial pay-for-performance contracts. Moving a Medicare Advantage plan's PDC-DR from 71% to 80% is worth roughly 0.5 Stars on the associated measure — meaningful when you remember Stars are worth $500 PMPY.
BLUF: CCM-RAMP is CallSphere's original six-stage framework for structuring an AI-led adherence program inside a chronic care management service line. Each stage has a defined call cadence, a specific clinical trigger, and an escalation path. It was developed after analyzing adherence-call transcripts across multiple chronic care deployments and mapping which sequences produced durable PDC lift in the 12-month window.
The framework runs inside CallSphere's healthcare voice agent — OpenAI gpt-4o-realtime-preview-2025-06-03, 14 function-calling tools, post-call analytics on sentiment, intent, and escalation — deployed across three live healthcare locations. The after-hours escalation component (7 agents + Twilio contact ladder) handles overnight red flags that would otherwise wait until morning and sometimes not wait at all.
Hear it before you finish reading
Talk to a live CallSphere AI voice agent for healthcare in your browser — 60 seconds, no signup.
BLUF: Most adherence failure is primary non-adherence — the prescription is written but never picked up — or refill-gap non-adherence where the patient falls behind schedule. AI voice agents that coordinate directly with pharmacies (retail, mail-order, and 340B) close both gaps by triggering auto-refills, initiating transfers, and confirming pickup timing.
According to Surescripts' 2025 National Progress Report, roughly 28% of new prescriptions for chronic conditions go unfilled within 30 days of prescribing — the "abandonment rate." That single failure accounts for $250B of the $500B total non-adherence cost. A voice agent that calls within 72 hours of an e-prescription being sent, confirms the patient understood the prescription, and schedules the pickup cuts abandonment by roughly 60% in our deployments.
// CallSphere CCM agent — refill status tool chain
async function checkRefillStatus(patientId: string, ndc: string) {
const [lastFill, daysSupply, pharmacy] = await Promise.all([
surescripts.getLastFill(patientId, ndc),
surescripts.getDaysSupply(patientId, ndc),
pharmacy.getPreferredPharmacy(patientId),
]);
const daysRemaining = daysSupply - differenceInDays(new Date(), lastFill.date);
const refillDueDate = addDays(lastFill.date, daysSupply - 7); // 7-day early refill window
return {
daysRemaining,
refillDueDate,
overdue: daysRemaining < 0,
earlyRefillOk: new Date() >= refillDueDate,
pharmacyId: pharmacy.id,
pharmacyPhone: pharmacy.phone,
mailOrderOption: pharmacy.hasMailOrderAlternative,
};
}
BLUF: A primary care group enrolling 2,000 patients in chronic care management needs 2,000 documented monthly touchpoints plus reactive inbound coverage. At an average 22 minutes of documented time per patient per month for basic CCM (99490 + 99439), that's 733 clinical-staff hours monthly, or about 4.6 FTE. AI voice agents handle roughly 80% of that volume at 10x lower unit cost while maintaining documentation and billing integrity.
| CCM Workload | Human-Only Cost | AI + Human Hybrid | Savings |
|---|---|---|---|
| 2,000-patient panel | $342,000/yr | $72,000/yr | $270,000 |
| 5,000-patient panel | $855,000/yr | $160,000/yr | $695,000 |
| 10,000-patient panel | $1,710,000/yr | $298,000/yr | $1,412,000 |
According to a 2025 AAFP (American Academy of Family Physicians) practice benchmarking report, the median small-group primary care practice that launched CCM saw a 31% gross margin on the service line — but that margin doubles in practices that moved to AI-assisted monthly touchpoints while keeping clinical escalation human.
BLUF: Diabetes adherence calls check three things: medication timing (especially insulin and GLP-1 agonists), blood glucose patterns, and hypoglycemia events. The agent correlates self-reported readings against the patient's CGM or fingerstick log if connected, and flags patterns that suggest medication timing errors versus true dosing failure.
BLUF: HTN adherence calls focus on daily dosing timing, home BP reading patterns, and side effects (especially dry cough on ACE inhibitors, which drives discontinuation). The agent pulls 7-day BP averages from connected home monitors, and if SBP>180 or DBP>110 on any reading, triggers immediate clinical escalation.
BLUF: CHF adherence calls are the most clinically sensitive — they combine diuretic timing, daily weight, symptom check, and fluid/salt intake. A 3-lb weight gain in 2 days or a 5-lb gain in 5 days is a standard decompensation red flag, and the voice agent warm-transfers the patient to the cardiology RN queue immediately on detection.
BLUF: COPD adherence calls check inhaler technique (a surprising share of "non-adherence" is actually correct adherence with incorrect inhaler use), rescue inhaler frequency, and exacerbation symptoms. The agent books a spirometry visit if rescue use exceeds 4 times per week, which is a GOLD-stage flag.
BLUF: Medicare CCM billing requires documented time, a certified EHR with a patient-centered care plan, 24/7 access, and documented patient consent. AI voice agents can check all four boxes — provided the platform writes timestamped time-tracking and care-plan updates back to the EHR on every call.
CallSphere's 20+ healthcare database tables include purpose-built CCM schemas: patient_ccm_consent, care_plan_versions, time_entries, escalation_events, and a normalized medication_adherence_log that maps to PQA PDC calculation. The time_entries table is the CMS audit target — and it's designed so that an auditor can pull a full month's documented minutes per patient with a single query.
For broader architectural context, see CallSphere's AI voice agents for healthcare post, the features page, or the pricing page for CCM-specific deployment scopes.
CCM requires 24/7 access to care for enrolled patients. CallSphere's after-hours escalation system — 7 specialist AI agents chained to a Twilio-based contact ladder with DTMF acknowledgment and 120-second timeout per contact — provides this layer cost-effectively. A CHF patient with a 3 AM symptom change gets an immediate structured triage call, and if severity warrants, the on-call cardiology provider is paged through the escalation ladder. Details at /features and /contact.
BLUF: The highest-performing CCM adherence programs integrate clinical pharmacists into the workflow — the pharmacist manages medication optimization under a collaborative practice agreement (CPA), and the AI voice agent handles the volume of monthly touchpoints the pharmacist can't. This hybrid model consistently outperforms pure-AI and pure-human approaches on PDC outcomes.
Still reading? Stop comparing — try CallSphere live.
See the healthcare AI agent handle a real call — complete, industry-specific, and live in your browser. No signup.
According to a 2025 APhA (American Pharmacists Association) practice report, CPA-enabled CCM programs saw a 14.2 percentage-point PDC improvement versus 8.6 points for non-CPA programs. The pharmacist's clinical authority to make dose adjustments and medication switches closes the failure loop that pure outreach cannot reach.
BLUF: A well-run AI-led adherence program has a recognizable 12-month trajectory — early wins in months 1-3 on primary non-adherence, steady refill-gap improvement in months 4-9, and durable PDC lift by month 12. Programs that plateau early typically did so because they optimized for call completion rate rather than clinical outcome.
| Month | Primary Metric | Typical Value | Leading Indicator |
|---|---|---|---|
| 1-3 | Primary non-adherence | Drop from 28% to 14% | First-fill pickup rate |
| 4-6 | Refill-gap days | Drop from 18 to 9 avg | 7-day-early refill rate |
| 7-9 | PDC (rolling 180-day) | Rise from 72% to 79% | Month-over-month refill consistency |
| 10-12 | PDC (rolling 365-day) | Rise from 71% to 82% | 90-day fill adoption rate |
According to CMS's 2025 Part D Star Ratings release, PDC measures (PDC-DR, PDC-RAS, PDC-Statins) each contributed ~1.5x weight to overall Part D Star. Moving from 71% to 82% on any one of these measures moves roughly 0.4-0.6 stars on that measure — meaningful when stacked across all three adherence measures.
BLUF: Adherence calls regularly surface red flags that have nothing to do with medication — suicidal ideation on depression-med check-ins, domestic violence hints during in-home safety probes, fall risk markers in elderly hypertensive cohorts. A responsible voice-agent program implements hard escalation paths for each, never forcing the agent to resolve clinical or safety issues outside its scope.
CallSphere's CCM agents include the following hard-escalation triggers: any mention of self-harm or suicidal ideation (immediate warm-transfer to 988 or behavioral health service), domestic violence disclosure (DV resource referral plus clinical escalation), fall in last 30 days in a patient >75 (care team notification), and any symptom pattern consistent with acute MI, stroke, or DKA (immediate 911 advisement plus live transfer to clinical staff). These are non-negotiable design patterns for any voice-agent system in chronic care.
CMS's CCM guidance requires the service to be provided by "clinical staff" under the supervision of a physician or other qualifying billing practitioner. AI voice agents are not clinical staff — but they can perform the non-clinical coordination work (outreach, scheduling, data capture) that frees clinical staff time for billable activities. Best practice is to have clinical staff review and co-sign every AI-generated encounter note, with the clinical time documented separately.
PDC (Proportion of Days Covered) is the percentage of days in a measurement period where a patient had medication on hand. MPR (Medication Possession Ratio) is total days supplied divided by days in the period. PDC caps at 100% per day and is the PQA-preferred measure because it handles overlapping fills correctly. Most Medicare Star Rating and quality contracts now use PDC.
Controlled substances — especially Schedule II stimulants and opioids — have additional DEA and state-level early-refill restrictions. CallSphere's adherence agent recognizes controlled-substance NDCs and adjusts the refill prompt logic to respect early-fill windows. For opioid adherence in chronic pain cohorts, the agent also runs PDMP-check-prompted conversations with the prescriber workflow rather than direct patient outreach.
No — the agent cannot prescribe. It can identify that a refill is needed and send a structured request to the prescriber's in-basket through Surescripts EPCS or the EHR's refill queue. The prescriber reviews and authorizes. This separation is both clinically and regulatorily important — the voice agent is a care coordinator, not a prescriber.
The agent triggers the after-hours escalation ladder immediately. For CHF weight gain, that's a warm-transfer attempt to the on-call cardiology RN, fallback to the on-call physician via Twilio call plus SMS, with DTMF acknowledgment required. The 120-second timeout per contact with automatic escalation to the next person in the ladder means no red-flag patient waits more than a few minutes for a human clinician.
90-day fills generally improve PDC mechanically because patients have more days supplied at each fill. The voice agent proactively recommends 90-day fills for stable chronic medications during month-3 or month-4 touchpoints, which correlates with a 3-5 percentage-point PDC improvement on average in our deployments. Not every medication is 90-day appropriate — the agent respects plan formulary rules and clinical guidance.
It works for both. Medicaid chronic care programs under 1115 waivers, Health Home models, and similar structures also need high-volume adherence outreach. The billing codes differ (Medicaid often uses state-specific HCPCS codes rather than federal CCM codes), but the clinical workflow is essentially the same. CallSphere's platform supports multi-payer configuration so a single deployment can handle commercial, Medicare, and Medicaid concurrently.
PDC is calculated on a rolling measurement period — typically 12 months for the annual quality measure. Operationally, you'll see a lift in monthly fill rates within 30-60 days of launching a well-designed adherence program, and the trailing 12-month PDC will catch up over the following 6-9 months. Most programs target a 10-percentage-point lift by month 12 and often exceed it.
Written by
Sagar Shankaran· Founder, CallSphere
LinkedInSagar Shankaran is the founder of CallSphere, where he builds production AI voice and chat agents deployed across healthcare, hospitality, real estate, and home services. He writes about agentic AI, LLM engineering, and shipping voice agents that handle real calls in production.
See how AI voice agents work for your industry. Live demo available -- no signup required.
Graphiti is the open-source temporal knowledge graph for AI agents in 2026. Learn how bi-temporal memory beats vector RAG for voice agents and long-running LLMs.
How to design a multi-agent system using MCP for tools and A2A for cross-vendor coordination, with a CallSphere voice agent as a participating node.
Self-correction is now a property of the model, not the framework. What that means for production agent reliability, voice/chat fallbacks, and CallSphere.
Offline evals catch regressions before deploy on a fixed dataset. Online evals catch real-world drift on live traffic. You need both — here is how we run them.
Build a working voice agent with the OpenAI Realtime API + Agents SDK, then bolt on an eval pipeline that catches barge-in failures, hallucinated grounding, and latency regressions.
The full metric set for evaluating production voice agents — STT word error rate, end-to-end latency budgets, RAG grounding, prosody, and the metrics that actually correlate with retention.
© 2026 CallSphere Inc. All rights reserved.
Made within San Francisco
Watch how CallSphere handles real customer calls, schedules appointments, and processes payments — live.
Try Live DemoBook a DemoCalculate Your ROI