Skip to content
CBS Health
Request a demo
  1. Home
  2. How it works
How it works

Capture, organise, understand, review, act.

Step four is a person. That is the whole of the difference between decision support and an automated clinician, and it is enforced in the architecture rather than in a policy document.

  1. Capture

    Reports, letters, labs and vitals arrive from the systems that already hold them, or are uploaded directly.

  2. Organise

    Each item is filed against the right patient and the right episode, and placed in time.

  3. Understand

    The engine summarises, compares against previous values, and flags what has changed.

  4. Review

    A clinician reads the synthesis with its sources beside it, and accepts, changes or rejects each part.

  5. Act

    What the clinician signed goes to the record, and the follow-up it implies goes to a queue with an owner and a date.

In detail

What each step actually involves.

1. Capture
Reports, letters, labs and vitals arrive from the systems that already hold them, or are uploaded directly.
HL7 v2 and FHIR R4 for hospital feeds; direct upload for documents that only exist as paper or PDF. Nothing is retyped.
2. Organise
Each item is filed against the right patient and the right episode, and placed in time.
Identity resolution runs before anything is filed, because the failure mode of getting it wrong is a document on the wrong patient. Ambiguous matches are queued for a human, not guessed.
3. Understand
The engine summarises, compares against previous values, and flags what has changed.
Every statement it produces keeps a pointer to its source. A synthesis that cannot cite itself is not shown.
4. Review
A clinician reads the synthesis with its sources beside it, and accepts, changes or rejects each part.
This step cannot be configured away. It is the reason the tool is decision support and not a decision.
5. Act
What the clinician signed goes to the record, and the follow-up it implies goes to a queue with an owner and a date.
The audit trail records what was shown, what was signed, by whom, and when -- which is what a clinical governance review needs and rarely has.
The shape of it

Data in, judgement out, care continued.

Four stages. The third is drawn heaviest on purpose: the software converges on a person, it does not route around one.

Patient data

Reports · History · Labs · Vitals · Medicines

Clinical Buddy engine

Analyse · Summarise · Compare · Flag · Structure

Clinician workspace

Review · Validate · Decide · Plan

Continuous care

Targeted follow-up · Patient guidance · Alerts

The same five steps, on one patient

Ramesh Kumar, 58, type 2 diabetes of fourteen years.

Synthetic, and followed through a single review.

  1. Before the clinic

    Fourteen years of record, read in ninety seconds.

    Ramesh has been seen in three departments. CBS assembles the chronology -- diagnosis in 2012, retinopathy screening in 2019, a nephrology referral last year -- and puts the summary on the clinician's dashboard with each claim linked to its document.

  2. On opening the case

    A trend nobody had seen as a trend.

    Each HbA1c had been reported as a single number and each looked tolerable. Plotted as a series, they have climbed from 7.1 to 9.4 over eighteen months. The engine flags the direction, not a diagnosis.

  3. Alongside the labs

    A dose the kidneys no longer support.

    Creatinine has risen; eGFR now sits at 38. CBS notes that the current metformin dose falls outside the recommended range at that eGFR, cites the value and the guideline, and stops there. It proposes no alternative and no new dose.

  4. In the consultation

    The clinician decides. All of it.

    The clinician reads the flag, disagrees with part of the summary, edits it, accepts the rest, and sets the plan. The modification is logged against their name -- not as a correction to be hidden, but as the record of who exercised the judgement.

  5. Afterwards

    The follow-up exists before anyone leaves the room.

    A six-week review is created, dated, and given an owner. Ramesh gets the plan and his medicine changes in Telugu. If the review is not done by week seven, a named clinician hears about it.

See the five steps run on a case you know.

A demonstration on synthetic data proves the interface. One on a case you already know the answer to proves whether the synthesis is any good.

Request a demonstration See how it works