Healthcare OS · Risk Stratification

Risk scores you can trust, defend, and act on.

Most risk stratification returns a static prediction in a batch report, with no way to see how it was reached. IASORA scores every patient across 34 factors, shows the weight and the source behind every contribution, proves its accuracy on your own history before go-live, and turns each risk level into governed clinical action.

Book a demo See how the score is built →

0.75 to 0.80 AUROC on 30-day readmission · validated on your population before go-live · every factor traceable

A score nobody can explain is a score no clinician will act on and no board will sign off.

Platform sequence

From a continuous score to coordinated care.

Four steps, running on one record. Nothing here is a batch job, and nothing stops at a number.

Step 1

Ingest the record

Conditions, medications, labs, vitals, encounters, care gaps and utilisation consolidated into one longitudinal FHIR record per patient.

Step 2

Score the patient

A normalised 0 to 100 composite across four weighted risk areas, combining an always-on deterministic baseline, established clinical indices and a governed learned layer.

Step 3

Explain the number

Every point traces back to the factor, the weight and the data source, with tier movement over time and the clinical triggers stated in plain language.

Step 4

Trigger the care

Risk levels map to action queues. Outreach starts, response windows are enforced, and anything unanswered escalates to a care manager.

Population view

Your whole panel, stratified and current.

Risk stratification overview showing how many patients hold a current score, the share at rising risk or above, the distribution across five risk levels from preventive to complex, and the population average for each of the four weighted risk areas.
Every patient with a current score, distributed across five levels, above the population average for each weighted risk area. The share sitting at rising risk or above is the number that governs your capacity plan.

Glass-box architecture

Transparent weights. Continuous updates. One-way clinical escalation.

The score is a composite of four weighted risk areas across 34 factors. On top of it sit clinical alert rules that can raise a patient's action priority. By design, they cannot lower it, so an acute event is never averaged away by a settled history.

How an IASORA risk score is composed and how a risk level can change The composite score is built from four weighted risk areas: clinical risk and documentation integrity at 30 per cent over 11 factors, utilisation and cost risk at 30 per cent over 9 factors, continuity and access risk at 20 per cent over 7 factors, and quality and outcome risk at 20 per cent over 7 factors. Separately, action priority runs across five levels, from level 0 preventive through stable, rising risk and high risk to level 4 complex or escalating. A clinical alert rule can raise a patient's level, and no rule can automatically lower one. COMPOSITE SCORE · 0 TO 100 30% 30% 20% 20% Clinical risk anddocumentation integrity Utilisation andcost risk Continuity andaccess risk Quality andoutcome risk 11 factors 9 factors 7 factors 7 factors ACTION PRIORITY · LEVEL 0 TO LEVEL 4 A clinical alert rule can raise a patient's level No rule can automatically lower one Preventive Stable Rising risk High risk Complex Level 0 Level 1 Level 2 Level 3 Level 4

The score reflects baseline severity. The level is the action priority. Keeping them separate is what lets a readmission or a missed post-discharge follow-up raise a patient today, without quietly rewriting the score that the readmission has not yet earned.

In the product

The rule, and the patient it moved.

Clinical alert rules listing the level each rule raises a patient to, the condition that triggers it, and how many patients it has fired for, above a banner stating that rules never lower a level.
Three of the clinical alert rules, each stating the level it raises a patient to and how many times it has fired. Rules are inspectable, versioned, and constrained so that none of them can lower a level.
A patient score breakdown showing a baseline risk score of 33 out of 100, a risk level of 3 for high risk, the three clinical alerts that caused the escalation, and a timeline of every risk level the patient has held over the preceding month.
A baseline score of 33 puts this patient in rising risk. Three clinical alerts raise the action priority to high risk, the reason is stated on the record, and every tier change is kept alongside it.

Operational impact

The score did not sit in a report. It triggered the care.

The score said rising risk
A baseline of 33 / 100. On score alone she sits at Level 2, in a cohort of several thousand, and no care manager would reach her this month.
The rules said high risk
A critical domain alert, heart-failure decompensation signs, and an uncontrolled diabetes trend. Action priority raised to Level 3, with the trigger recorded.
What IASORA did
Opened an urgent care manager task, started outreach in her language against a countdown the queue enforces, notified her PCP, and wrote the tier change to her chart.

Platform capabilities

Everything the score needs to survive clinical, actuarial and board review.

Transparent 34-factor engine

Four weighted risk areas across 34 factors, resolved into one normalised 0 to 100 composite.

Established clinical indices

NEWS2, heart failure risk, CHA₂DS₂-VASc, eGFR on CKD-EPI 2021, Wells DVT and Morse Fall, alongside Charlson, Elixhauser, HHS-HCC, LACE and NHS eFrailty.

External model adapters

Connect a third-party or proprietary scoring engine, including ACG-compatible endpoints, without custom integration work.

Governed learned layer

An optional learned adjustment, promoted to production only after accuracy, calibration, time-split and fairness checks pass. Reversible in one step.

Continuous re-scoring

Scores recalculate as ADT feeds, results and encounters land on the FHIR pipeline, so nothing waits for a nightly batch.

Versioned model governance

Read-only standard models with tenant-tuned copies. Every weight, threshold and alert rule change is versioned with an audit diff.

Architectural distinction

Most risk stratification stops at prediction.

Traditional risk stratificationIASORA Healthcare OS
ExplainabilityA black-box score with weights you cannot inspectEvery factor, weight and alert rule traceable in plain language
Data freshnessPeriodic batch runs, already stale when they landContinuous re-scoring off the FHIR event pipeline
Fit to youA generic national model applied unchangedValidated and tuned on your own population before go-live
What happens nextA score in a report or a spreadsheetRisk levels trigger tasks, outreach response windows and clinical escalation
GovernanceTrust required, with no version historyClone-and-edit lifecycle, versioned audit diffs, reversible learned layer
Model architectureOne fixed algorithmWeighted baseline, clinical indices, external adapters and a governed learned layer

Proven accuracy

Validated on your historical data before deployment.

Calibration and predictive accuracy are proven against your own retrospective data before any production workflow is switched on. Three outcomes are measured.

30-day readmissionPost-discharge readmission likelihood, used to target transition-of-care workflows.
30-day ED visitRising-risk patients with a high probability of rapid emergency department utilisation.
90-day inpatient admissionMedium-term deterioration, used to trigger proactive care management.

AUROC measures how well a model separates higher-risk from lower-risk patients, where 0.5 is a coin toss and 1.0 is perfect. IASORA scores between 0.75 and 0.80 on 30-day readmission in retrospective evaluation across a 323,000-patient history. Your own number is produced the same way, on your population, before go-live.

Implementation

A risk score you can defend at the board and in the clinic.

Bring a cohort and a year of history. We will validate the engine on your own data before go-live, then show you how every risk level turns into governed clinical action.

Book a demo See it running in Care Hub →