AI-Powered TB Detection

Smarter Decisions.
Better Outcomes.

AI-powered risk prediction for TB treatment outcomes helps clinicians identify high-risk patients early and improve care across Kenya.

Get Started Sign in

What a clinician sees

One number — and what to do about it.

AI Analysis Example patient

Risk of Unfavourable Outcome

32%

Moderate risk

Early intervention recommended

Secure & encryptedRole-based, audited access
Evidence-basedKenyan TB register data
Built for impactFinds failures early
Clinician friendlySimple, fast, clear

Why TB Predict?

Most patients finish TB treatment well. The ones who do not — who die, fail treatment or are lost to follow-up — are usually easier to reach early than late. TB Predict helps find them early.

Know who needs you most

Every assessed patient gets a probability of an unfavourable treatment outcome, so follow-up time goes to the patients most likely to need it.

The right model for each patient

Several trained models are kept ready. Each patient is routed to the one that performs best for patients like them, and the reason is recorded with the prediction.

Checked against real outcomes

Predictions are compared with the outcomes recorded in the TB register, and with the existing TIBU risk score, before anyone is asked to rely on them.

Counties covered: 47
Clinical inputs per prediction: 19
Fields required to score: 3
Interoperability standard: FHIR R4

Features

Everything the programme needs to act on a prediction

Findings

Where the model works and where it does not: a threshold explorer, subgroups, extreme cases, and the TIBU score side by side.

Monitoring

One filter bar over every assessment: the Kenya map, the risk mix, trends over time and the facilities carrying the load.

Patient history

Every assessment recorded for a patient, their risk over time, and the TIBU score beside the platform's own.

Follow-up worklist

Patients flagged at high risk become a worklist for their facility: pending, contacted, resolved and overdue.

Guided assessment

Five short steps and a review before scoring. Only age, sex and county are required; anything not yet known can be left blank.

HL7 FHIR R4 API

Every clinical read and write goes through a standards-based API, so other health systems can use exactly what this site uses.

How It Works

From registration to follow-up

Watch one patient move through the platform.

Step 1 of 4 · Assess

1
Assess

A clinician enters the patient's registration details — or updates them when the month-2 smear result comes in.
2
Score

0.78

The patient is routed to the best-suited model, which returns a probability, a risk band and why that model was chosen.
3
Act

Follow-up raised

High-risk patients are flagged, and a follow-up is raised on their facility's worklist.
4
Learn

Outcome recorded

As outcomes are recorded, predictions are compared with what actually happened, model version by model version.

Interoperability

Built on HL7 FHIR R4

Clinical systems read patients, observations and predictions as standard FHIR resources, and request a prediction with a single operation. Access is limited to each caller's facility, county or national scope, and every read is audited.

Read the API docs Capability statement

Request a prediction

POST /fhir/Patient/$predict-outcome
Authorization: Bearer <token>
Content-Type: application/fhir+json

HTTP/1.1 200 OK
{
  "resourceType": "Parameters",
  "parameter": [
    { "name": "riskAssessment",
      "resource": { "resourceType": "RiskAssessment", … } },
    { "name": "task", … }
  ]
}

Need an account?

Accounts are for TB programme staff at facility, sub-county, county and national level. Every request is reviewed by a super-administrator, and you are emailed the decision.

Request access

Integrating a system?

The FHIR API documentation lists every endpoint, with example requests and responses. Definitional endpoints can be explored without an account.

Open the API docs

Smarter Decisions.Better Outcomes.