Case studies / Monarch Health

Case study · Internal medicine

Every patient walks in with their story already on screen.

Every night, medscrub syncs Monarch Health's eClinicalWorks charts and works through them with four skills the practice wrote itself. The day is read before anyone's in the building.

Monarch HealthMonarch HealtheClinicalWorksSandpoint, Idaho
At a Glance● OnlinePHI protected

RAN LAST NIGHT

Follow Up Visit Summary150
New Patient History2
Wound Care Visit Note10
Sick Visit Prep6

Marv Murchins

68y · MRN-MARV-0001

3 open plans · A1c 8.2% (07/10/2026)

Kee Beverly

2 open plans · A1c 6.2%

Marv Murchins Follow-up

↗ Push to eCW

LAST 3 MD VISITS

07/10/2026 provider, Monarch Health

T2DM — metformin 1000mg BID; A1c 8.2% (07/10), up from 7.8% (04/02)

HTN — lisinopril 20mg daily; BP 148/92 in office

HLD — atorvastatin 40mg qHS; LDL 138 (07/10); eGFR 58 — CKD 3a

OPEN PLANS · 3 OF 5 PLANS UNRESOLVED

Type 2 diabetes — suboptimal control

PLAN Recheck A1c in 3 months; discuss SGLT2i if above goal · 04/02/2026

07/10/2026 A1c 8.2% — above goal, no follow-up scheduled

The morning briefing: skills ran overnight; Marv's follow-up summary and open plans, ready to review. Synthetic patient data.

~2 hrs

saved per clinic day

4 skills

written by the practice, in production

3 plans

open plans caught per patient, avg

5:00 AM

every chart prepped before

medscrub desktop app — Marv's chart with the anatomy body map and chart-aware assistant
Marv's chart in the desktop app: the body map, markers glowing by status, with the chart-aware assistant alongside — de-identified on this device. Synthetic patient data.

Open plans

Did the plan actually happen?

Every visit ends with a plan: recheck the A1c in three months, start a home blood-pressure log, get the shoulder imaged. Then the next patient is roomed, and nobody looks back at that plan until the patient returns. If they return. Monarch's follow-up skill reads each chart the night before and answers four plain questions: what was the last primary complaint, what plan did the clinician and patient agree on, was it followed, and what happened. A plan that never happened shows up as an open plan, flagged before the patient is in the room.

A copy of the record, outside the EMR.

The nightly sync leaves Monarch with their own FHIR copy of the practice's records, on their own machine. Skills, chat and briefings all run against that copy — any AI tool can, not just the ones their EMR vendor decides to sell them. The chart stays in eClinicalWorks; the leverage doesn't.

Custom skills

Four skills, written in the practice's own words.

A skill is a page of plain-English instructions the AI follows for every chart. Monarch wrote theirs the way you'd brief a new hire: what to include, what to skip, how the note should read. They run every night against the synced charts.

Follow Up Visit Summary

Runs for every scheduled patient. Last 3 MD visits, problem-based summary, lab trends with source dates — and the open-plan check on every prior plan.

150 charts / night

New Patient History

Any new patient who hits the clinic gets a full historical build: active conditions grouped with their meds and monitoring data, plus health maintenance gaps.

Runs on arrival

Wound Care Visit Note

Compares the last two wound visits: measurements, tissue composition, procedures performed, treatment plan — documented findings only, never inferred.

10 charts / night

Sick Visit Prep

When a patient books an acute visit, the relevant history is pulled and digested so the MD walks in current — not scrolling.

On schedule change

Follow Up Visit Summary skill page with quality scores and run output
A skill's page in the app: quality scores from nine test fixtures — rule pass, faithfulness, guards — then the run output, ready to copy or write back to eCW. Synthetic patient data.

Writing one takes a paragraph, not a project.

The skill builder takes a plain description — written the way you'd explain it to a colleague — and sets up the data, prompt, output and guardrails itself. Monarch's wound care skill started as a few sentences about what their RN wanted in the note.

Skill Builder✓ Save skill

✦ Just describe what you want

Write it the way you'd explain it to a colleague. No settings required.

Compare the last two wound visits and summarize progression — measurements, tissue, drainage, procedures performed. Documented findings only; if it isn't in the chart, say so.

+ Find overdue mammograms+ Draft a referral letter
✦ Build it for me

✓ Here's what this skill will do

✓ Read the relevant information from each patient's chart, de-identified before any AI sees it
✓ Use the AI to produce a written summary
✓ Show you the result to copy or share by hand

⛨ Nothing is sent or written to a chart until you review and approve it.

The skill builder: a plain-English description in, a scheduled nightly skill out.

WOUND CARE VISIT NOTE · MARV MURCHINS

Clinical Progression

L plantar forefoot DFU: 2.1 x 1.4 x 0.3 cm (07/17) → 1.6 x 1.1 x 0.2 cm (07/31). Granulation increased ~60→80%; slough decreased. Scant serous drainage, no odor. Periwound intact, no maceration. Offloading adherence documented both visits.

Status Assessment

Improving — decreasing depth, increasing epithelialization.

Procedures Performed

Conservative sharp debridement (07/17); callus reduction (07/31).

Wound Treatment

Cleansing, collagen contact layer, foam secondary; offloading boot continued.

Wound care skill output — documented findings only, ready for the RN's review.

NEW PATIENT HISTORY · MARV MURCHINS

Active Conditions and Clinical Management (Historical)

T2DM — metformin 1000mg BID. A1c 8.2% (07/10/2026), 7.8% (04/02/2026). HTN — lisinopril 20mg daily. HLD + CKD 3a — atorvastatin 40mg qHS; LDL 138, eGFR 58 (07/10/2026).

Health Maintenance and Prevention

Diabetic eye exam not documented in records reviewed. Pneumococcal due per age + DM. CRC screening current (2023 colonoscopy, outside record).

New patient skill output — every item carries a source date.

Ad-hoc chat

Ask the chart anything. The chart never leaves.

Some questions don't fit a skill. The doctor can pull up any patient and just ask — how should we plan the next visit, what changed since March, am I missing anything on this med list. The answer comes from the record itself, pulled straight from the EMR, and he can put the question to whichever LLM he prefers. Names, dates and record numbers are swapped for tokens on the office computer first, so the model — Anthropic, OpenAI, Google, or one running locally — never learns who Marv is.

Marv's name never leaves the computer.

Marv's assistant · Chart-aware⛨ De-identified on this device

How should we plan for the patient's next visit?

Next Visit Planning

Suboptimal control across all three cardiometabolic conditions, plus early CKD (Stage 3a):

ParameterCurrentTargetStatus
A1c8.2%<7%✕ Above goal
BP148/92<130/80✕ Stage 2 HTN
eGFR58CKD Stage 3a

1. Medication intensification — add SGLT2i (renal protection given eGFR 58) or GLP-1 RA (BMI 31.4); suboptimal A1c on metformin monotherapy.

Chart-aware chat on Marv's record. Synthetic patient data.

“The open plans are the part we can't give up. It remembers the plans we made and tells us whether they actually happened.”

Monarch Health · Sandpoint, ID

Walk in tomorrow with every chart already prepped.

Set up is self-service: sign up, download the app, follow the connectivity guide for your EMR, connect the LLM you prefer, and run your first sync — the insights start with it. eCW syncs nightly; Epic, Cerner and athenahealth refresh sub-hourly.

Works with your EMR

EpicOracle CernerathenahealtheClinicalWorks