Step 3 of 12Evidence
03 · Facts, claims and gaps

Examine the evidence pack

Sign in to save your work

Read every document. Separate what is confirmed from what is claimed, and note what is missing.

Business request

Pilot extension paper

From: Emergency Department Operations Director

The pilot has reduced average time to first triage from 11 to 4 minutes. We propose extending it to the paediatric emergency department and using it at night without a second triage nurse.

  • Extension planned before the safety case is signed off
  • Paediatric use is outside the validated population
Supplier validation summary

TriageLogic validation (extract)

MeasureSupplier result
Sensitivity for categories 1–2 (overall)92%
Patients in validation set48,000 adults, one urban hospital
Results by sex, age band or ethnicityNot reported
Patients under 16Excluded
Free-text in languages other than EnglishNot tested
  • 8% of the most urgent patients were under-triaged in validation
  • The hospital has not replicated the results locally
Pilot data

Nine-week pilot figures

From: ED performance team

MeasureResult
Patients triaged with CareAssist2,400
Nurse changed the suggested category96 (4%)
Nurse review recorded within 15 minutes71% of audited records
Patients later upgraded to category 1–2 after a low suggestion19
Average time to first triage4 minutes (from 11)
Incident report

Serious incident SI-0417 (summary)

A 58-year-old woman reported jaw pain, nausea and tiredness at the kiosk. CareAssist suggested category 4 with medium confidence. The nurse confirmation was recorded 52 minutes later, during a busy period. She waited 2 hours 40 minutes and was then diagnosed with a heart attack.

  • Atypical heart-attack symptoms are more common in women
  • No alert fired for the late nurse confirmation
  • Investigation not yet concluded
Staff survey

Triage nurse survey (22 responses)

  • 61%: 'When it is busy I usually accept the suggestion'
  • 45%: 'I am not sure what the confidence level means'
  • 32%: 'I have never been trained on what CareAssist gets wrong'
Clinical safety status

Clinical safety case and hazard log

  • Hazard log in draft, 14 hazards identified, 5 without controls
  • Clinical Safety Officer has not signed the safety case
  • No agreed process for reporting suspected AI errors

Facts, claims and gaps

Keep these three apart. A decision-maker needs to know which is which.

What does the evidence pack establish?

What is asserted by the supplier or the business but not yet evidenced?

Show any calculation. Say what the figures do and do not tell you.

Name the specific documents, data or tests you would ask for

Answers save automatically and sync to your account when you are signed in.