Medication-adherence Clinical Decision Support
Adherence declines quietly. The signal shouldn’t.
The 30 days after discharge are where medication adherence often falters — and where a facility sees least. RxSmart.ai is facility-grade, non-device clinical decision support that scores each enrolled patient every cycle and surfaces each finding — a suggested clinician, action, and review window — for the care team to decide.
Non-device CDS under Section 520(o)(1)(E) · Proprietary scoring engine · Every recommendation routes through a licensed clinician
The visibility cliff
Inside the hospital, every dose is documented. At discharge, the signal ends.
Hospital-wide — including the ICU — every administered dose is nurse-documented in the eMAR with an exact timestamp. At discharge that signal ends completely, and the system falls back to self-report and retrospective pharmacy-fill data (PDC / MPR). The ICU patient stands at the top of the highest cliff: the most medication churn, the most complex discharge regimen, and the least self-management capacity after critical illness — then zero visibility.
The medication error is born at the step down from the ICU.
Across 58 ICUs in 34 US hospitals (and 2 in the Netherlands), 450 of 985 patients — 45.7% — carried a medication error forward at the transition out of the ICU. The most common: an ICU-only drug continued (28.4%) or a home condition left untreated (19.4%). Three-quarters of those errors reached the patient; few caused harm — the point is the error travels, not that it always harms (Tully AP, Hammond DA, Li C, et al., 2019, Critical Care Medicine 47:543–549).
Adherence at admission predicts readmission
In a 385-patient cohort at Cedars-Sinai, patients with low or intermediate adherence (MMAS-4, measured at admission) were readmitted at 20.0% versus 9.3% for high adherence — an adjusted 2.54-fold higher odds of readmission (Rosen et al., 2017, Patient Preference and Adherence).
Where preventable readmissions come from
Among potentially preventable medication-related readmissions: 35% trace to non-adherence, 35% to prescribing errors, and 30% to transition errors (Uitvlugt et al., 2021, Frontiers in Pharmacology).
The national cost of non-adherence
Medication non-adherence is associated with approximately 125,000 US deaths per year, at least 10% of hospitalizations, and an estimated $100–289 billion in annual health care cost (direct and indirect). These figures reflect associations rather than established causation.
Sources: Annals of Internal Medicine (2012 systematic review); Duke Health (Bosworth); APhA / HHS.
Adherence erodes after discharge
A qualitative longitudinal study at Geneva University Hospitals — 21 patients, 75 interviews over two months post-discharge — found adherence erodes in the transition to autonomous self-management (Solh Dost et al., 2024, BMC Health Services Research 24:620).
How RxSmart.ai bridges the continuum — ICU to home, six steps.
1
Identify — while still admitted
The candidate is flagged inpatient, ICU or ward. Enrollment is FHIR-only: connect the EHR, search, and import the patient with active medications and active conditions. No manual entry.
2
Baseline — the reconciled discharge list
The reconciled discharge medication list becomes the anchor the outpatient signal is measured against.
3
Discharge trigger — the 30-day window
Discharge starts the 30-day Transitional Care Management window the platform detects, with billing advisories staged for your certified coder.
4
Dual-tier signal — device or device-free
A connected smart cap follows the device-supply path: 99454 at 16+ of 30 days transmitting, 99445 covering 2–15 days — the two are mutually exclusive, and the certified coder assigns the family; the platform never chooses. Device-free patients (self-report / pharmacy fills) follow time-based codes only, honestly labeled as the weaker signal.
5
Review — every finding lands on a human
The ratified bands order the clinician’s worklist: ALERT (25–49) surfaces the patient for pharmacist review, with a suggested review window of 72 hours; CRITICAL (0–24) surfaces the patient for coordinator / physician review, with a suggested review window of 24 hours. The clinician decides whether and when to act.
6
Beyond day 30 — no second cliff
The patient steps down to standard cohort-dashboard monitoring — the continuum has no cliff of its own.
The honest boundary
RxSmart.ai extends a prospective, per-patient signal beyond the hospital door; it does not verify ingestion and does not replicate eMAR-grade certainty at home. PDC is retrospective and aggregate; our score is prospective and per-patient. We are building the pilot data to demonstrate it. Tier thresholds are operationally ratified thresholds, under prospective pilot validation.
The engine
Scores that route to action.
The engine reads signals your data already carries and computes a single 0–100 adherence score. The method is a proprietary scoring engine — the score, its tier, and the recommended action are shown; the formula is confidential.
Missed doses — frequency and recency of gaps
Refill latency — delay between due and fill
Polypharmacy — interaction and regimen complexity
Engagement — response and touchpoint patterns
Cost barriers — affordability signals that predict abandonment
| Tier | Score | Action & window |
|---|
| STABLE | 70–100 | Routine monitoring |
| MONITOR | 50–69 | Watch at next touchpoint |
| ALERT | 25–49 | Pharmacist outreach < 72h |
| CRITICAL | 0–24 | Coordinator / physician < 24h |
| NO DATA | — | Verify enrollment / source |
Thresholds are operationally ratified and under prospective pilot validation — not represented as clinically validated cut-points.
See it work
Move the signals, watch the tier and action change.
An illustrative demonstration of how signals map to a tier and a clinical action. This demo uses simplified illustrative math — not the proprietary scoring engine.
The five agents
Purpose-built minds on your live cohort.
Each agent does one job and reports back for clinician review. Agents inform — they never act on a patient without you. They operate only inside the platform, on your enrolled cohort, after sign-in.
Triage
Ranks the cohort by escalation urgency and drafts the day's outreach worklist.
Billing
Reviews device-supply eligibility (99445 / 99454) and time-based capture (99470 / 99457 / 99458), flags missing interactive-communication documentation, and surfaces the RPM / RTM distinction for your certified coder.
Trends
Surfaces adherence trajectories — who is declining before they cross a tier boundary.
Sentinel
Watches for co-firing risk patterns that historically precede readmission, and for data-integrity gaps.
Copilot
Answers staff questions about the live cohort in plain language for nurses and care coordinators.
Economics
A flat fee, designed to be offset by reimbursement.
RxSmart.ai is offered at flat per-patient facility pricing, structured so that CMS remote-monitoring reimbursement can offset the cost. Contact us for facility pricing. All figures below are illustrative — subject to the current CMS Physician Fee Schedule and geographic adjustment, and not a guarantee of payment.
Billing alongside Transitional Care Management
RPM or RTM monitoring may be billed alongside a TCM service period — only one of RPM or RTM, never both — provided no unit of time is counted toward more than one code. Whether and how to bill is determined by your certified coder against current CMS rules and AMA CPT time-counting requirements. RxSmart does not select or assign billing codes.
Where it fits
Different from a reminder app, a pill bottle, or an EHR module.
| Capability | RxSmart.ai | Reminder app | Smart bottle | EHR module |
|---|
| Computes clinical risk (0–100 score) | Yes | No | Partial | Partial |
| Time-bound action per tier | Yes | No | No | Partial |
| Tamper-evident audit chain | Yes | No | No | No |
| Reimbursement-aligned billing support | Yes | No | Partial | Partial |
| Hardware required | No | No | Yes | No |
General category comparison for orientation only; individual products vary. RxSmart.ai is non-device CDS under Section 520(o)(1)(E) and is not FDA approved, cleared, or authorized.
Deployment
From signature to first scored cohort in about a week.
1
Sign the BAA (days 1–2)
A standard, attorney-reviewed Business Associate Agreement. No data is connected before it is executed.
2
Connect your data via FHIR (days 3–5)
Authorize RxSmart.ai against your FHIR R4 endpoint with read-only scopes; your tenant is provisioned on encrypted infrastructure.
3
First scoring pass (day 6)
Every enrolled patient is scored and tiered; the worklist is triaged on first login.
4
Operate & bill (week one)
Clinicians work the tiers; billing-eligible encounters are staged with documentation and audit provenance.
Request a 30-day pilot
Training
Your team is onboarded inside the platform.
An interactive clinical training library — modules, knowledge-check questions, and saved progress — maps each concept to a workflow: the tiers and escalation windows, the routing rules, the billing distinction, FHIR enrollment, MemoryAnchor framing, and audit-chain review.
Tiers & windows
How the ratified bands map to a 72-hour or 24-hour clinical action, and the override workflow.
Billing workflow
The CY2026 code set, mutual exclusivity, the interactive-communication rule, and the RPM / RTM distinction.
FHIR & audit
Enroll via the EHR with no hand-typed PHI, and verify the SHA-256 chain before compliance review.
Inside the platform
A worklist ordered by clinical urgency.
After sign-in, the cohort surfaces highest-need patients first — NO DATA, then CRITICAL, ALERT, MONITOR, STABLE.
RxSmart.ai · Cohort worklist (illustrative)
Patient — CHF
missed 11 / 30 · 12 days lateCRITICAL 18
Patient — T2DM
missed 6 / 30 · 5 days lateALERT 41
Patient — COPD
missed 3 / 30 · on deviceMONITOR 61
Patient — HTN
missed 1 / 30 · on deviceSTABLE 88
MemoryAnchor
A per-patient memory buffer — for every patient.
MemoryAnchor holds each patient's profile, people, medications, and daily routine, and returns them on demand through a patient link and a caregiver link. It works for any patient who benefits from having their own information held and handed back — someone managing several medications, a patient newly discharged and adjusting to a changed regimen, anyone who wants a caregiver kept in the loop, and, among these, patients living with memory difficulty. Adherence signals from the buffer are surfaced for the care team's review. For patients living with cognitive impairment, MemoryAnchor is cognitive support, not treatment: it does not treat, restore, or cure dementia or Alzheimer's disease, and it is not a substitute for clinical care.
FAQ
Questions hospitals ask.
How is the score different from MPR or PDC?
MPR and PDC are retrospective, aggregate ratios. The engine combines multiple live signals into a prospective, per-patient 0–100 score that routes to a specific clinical action each cycle. We are building pilot data to demonstrate comparative performance; we do not claim the score outperforms PDC without that study.
Is it HIPAA-aligned?
A Business Associate Agreement is executed with each facility before any patient data is connected, and every action seals to a tamper-evident SHA-256 audit chain.
Does it replace our EHR?
No. RxSmart.ai reads from your EHR via FHIR R4 and supports clinicians; it is not a system of record.
What does the Section 520(o)(1)(E) exemption apply to?
It applies to the software's status as non-device Clinical Decision Support: the clinician remains the decision-maker, and RxSmart.ai is never described as FDA approved, cleared, or authorized.
What happens to our data if we cancel?
Your tenant data is handled per the Business Associate Agreement and the data-handling terms agreed at signing.
Who built RxSmart.ai?
Bionectech, Inc. (Texas).
Honest status
What is live in production today.
Live in production
- Cohort dashboard, patient roster, and admin console
- Scoring engine with the ratified four-band tiering and NO-DATA handling
- Five-agent console and MemoryAnchor per-patient memory buffer
- JWT authentication and role-based access control
- SHA-256 append-only audit chain
- CPT staging for the CY2026 code set (99445 / 99454 device tiers; 99470 / 99457 time tiers (mutually exclusive); 99458 add-on only after 99457 is met) with interactive-communication capture — RPM vs RTM set by your certified coder
- Discharge medication-reconciliation view (FHIR-computed)
- FHIR R4 patient import (Epic, Cerner, athenahealth)
- Interactive clinical training library
Get started
Start with a 30-day evaluation — on synthetic or de-identified data first.
The evaluation runs on a synthetic or de-identified cohort, so your team sees the engine score, tier, and stage a full worklist within a week — before any PHI is involved. Patient data is connected only after a Business Associate Agreement is executed.