IMI Get in touch
Who it's for

Four places where reading the whole patient changes what happens next

Each of these teams already has decision support, monitoring, or an agent in place. What they're missing is the same thing: a single, continuously updated read on the patient that every piece of that system can share.

Hospital-at-home care team 1 · Hospital at home
Remote monitoring at home 2 · RPM
Digital health builders 3 · AI builders
Health system command centre 4 · Health systems

1 · Hospital-at-home programs

When one condition's plan collides with another's

Acute patients at home rarely have one condition. The care team is spread across a command centre, visiting clinicians and a remote attending, and each piece of decision support reasons about one condition at a time.

  1. A patient with heart failure and chronic kidney disease is on day 4 of a heart-failure admission at home, decongested and stable.
  2. The heart-failure pathway sees room to step up sacubitril/valsartan from 49/51 mg to 97/103 mg twice daily before the episode ends.
  3. That morning's bloods show potassium 5.4 mmol/L and eGFR down from 48 to 36 since admission.
  4. The layer holds the titration back and tells the attending why: “Potassium 5.4 and eGFR 36 on 18th Sept.”
HF pathway Suggests titration ↑ Renal labs K 5.4 · eGFR 36 Context layer Reads both, same step Hold titration Reason shown to attending

What the layer does

No one wrote a rule linking heart failure to the kidney. The kidney's limits apply to any plan for that patient, because every condition is read in the same step.

Why it matters

Decision support that suggests a dose increase without reading the potassium is worse than none. A held-back suggestion with a named reason is one the team can trust.

2 · RPM and virtual-care vendors

When the danger sits below every threshold

Monitoring runs on per-reading thresholds. They flood nurses with alerts that mean little, and stay silent on patterns that mean a lot, because no single reading crosses a line.

  1. A heart-failure patient is two weeks out of hospital, on a connected scale. The program alerts on a gain of 2 lb in a day or 5 lb in a week.
  2. Weight rises 1.3, 1.1, then 1.4 lb over three days. Nothing fires.
  3. Pharmacy fill data shows the diuretic prescribed at discharge was never picked up.
  4. The layer proposes a nurse call today: “+3.8 lb in 3 days; discharge diuretic not filled.”
Alert threshold: 2 lb/day +1.3, +1.1, +1.4 lb — none fire Pharmacy fill data Context layer Reads trend + record Call nurse today — +3.8 lb, unfilled Rx

What the layer does

It reads the trend against this patient's own baseline and against everything else known about them. If the medication record is out of date, it says so rather than guessing.

Why it matters

The prompts that reach your nurses carry their reason, including the ones no threshold would have raised. That is the difference between a monitoring feed and a clinical service.

3 · Digital health and AI builders

When your agent reads the record but not the patient

Clinical agents and copilots are fed raw records: long, unordered, and silent on what the care team has already ruled out. The agent has to rediscover clinical judgement on every call, and it will sometimes get it wrong.

  1. A patient messages your care-team assistant: “Can I take ibuprofen for my knee?”
  2. The record lists heart failure, stage 3 chronic kidney disease, a loop diuretic and sacubitril/valsartan, spread across problem lists and medication tables.
  3. The layer hands the agent that picture already joined up, with the clinician-authored limit that applies: avoid NSAIDs for this patient.
  4. The agent drafts a reply advising against ibuprofen and suggesting an alternative. A clinician confirms it before it is sent.
“Can I take ibuprofen?” Problem list · Meds scattered, unordered Context layer Rule: avoid NSAIDs Agent drafts reply ✓ Clinician confirms → sent

What the layer does

It gives your agent structured, current context and the rules that bound it, and records what the agent read and why it answered as it did.

Why it matters

You build the agent. We give it a patient to reason about, with a trace you can show a clinical buyer.

4 · Health systems

When each vendor sees only its own slice

A single patient can sit in a heart-failure monitoring program, a diabetes program and primary care at once. Each vendor judges its own data, and every one is a separate integration for your IT team to run.

  1. The diabetes team starts empagliflozin, which has a mild diuretic effect. The patient is already on a loop diuretic.
  2. Over the next week, the heart-failure program sees weight fall 2.5 lb and systolic pressure drift down about 15 mmHg. Its rules watch for weight gain, so nothing fires.
  3. Neither vendor sees the other's data. The layer sees both.
  4. It proposes to the heart-failure clinician: “Review loop diuretic dose; weight and BP falling since empagliflozin started 10 Sep by the diabetes team.”
Diabetes program HF program Primary care Context layer One record, all programs To HF clinician: review diuretic dose

What the layer does

One context per patient, across every program that touches them, sitting beside your EHR rather than replacing it.

Why it matters

Your clinicians see the patient, not the vendor boundary.

See where this fits in your program

Whichever of these looks like you, the underlying gap is the same — walk through it with us.

Talk to our founders