Signal review without false certainty
Surveillance workflows should separate early warning from confirmed action and expose data delays, bias, and alternative explanations.
AI workflows for the decisions public health teams make every day.
We map the questions public health professionals ask, then build reviewable AI workflows that public health teams can inspect before action.
Human accountable
Source linked
Evaluation first
Source linked
Human accountable
Evaluation ready
The APHI Method
APHI is designed as an operating system for disciplined public health AI work: start with the question, build the workflow, evaluate the output, then decide what is ready for a pilot.
01
Question
What decision or operational task is the team trying to support?
02
Workflow
What inputs, outputs, review roles, and boundaries are required?
03
Draft
What can AI summarize, compare, structure, or flag for review?
04
Review
Which qualified human approves the output before use?
05
Evaluate
How are accuracy, safety, usability, equity, and source quality tested?
Built from real public health AI needs
APHI extends the practical logic of the Public Health AI Handbook into a product system: evaluate before adoption, keep humans accountable, and make field constraints visible in every workflow.
Read the Public Health AI HandbookEvidence briefs with source discipline
AI output needs visible citations, uncertainty, reviewer questions, and a clear boundary between draft assistance and final guidance.
Evaluation that survives real operations
Each workflow should document accuracy, safety, usability, equity, drift, reviewer acceptance, and when not to deploy.
Workflow Library
APHI organizes AI support around the work public health teams already do. Each workflow includes the question, intended user, input requirements, AI-assisted output, human-led boundary, and evaluation method.
Open the library
Category
Example question
Output
Surveillance
Is this pattern unusual?
Signal review memo
Evidence
What source supports action?
Citation checked brief
Communication
What should the public know?
Reviewed advisory draft
Evaluation
Is the program working?
Indicator and rubric pack
Flagship workflow: outbreak signal review
A serious APHI workflow does not just produce a polished paragraph. It creates a reviewable evidence trail: signal summary, uncertainty, source checks, reviewer questions, escalation logic, and explicit limits.
Workflow statusDraft specification
Decision boundaryEpidemiologist approval
Evaluation pathShadow review before pilot
Example output
Outbreak Signal Review Memo
- Signal summary
- Respiratory syndromic visits increased in two districts over the current reporting window.
- Uncertainty
- Recent reporting delays and care-seeking changes could explain part of the pattern.
- Reviewer questions
- Does the pattern persist after late reports? Are laboratory confirmations aligned?
Evaluation that funders and agencies can inspect
APHI makes evaluation visible. Every workflow should move through readiness levels only when its risks, tests, reviewers, and failure modes are documented.
View evaluation framework
Accuracy
Claim and citation checks
Safety
Refusal and escalation rules
Usability
Reviewer acceptance rubric
Equity
Bias and missingness review
Governance
Privacy and approval path
Health department pilot pathway
APHI can be scoped as a practical pilot: choose one workflow, define data and review boundaries, run shadow evaluation, and decide whether the output improves the real public health process.
1
Define the operational question and reviewer role.
2
Build a workflow packet with inputs, outputs, limits, and review rules.
3
Test against synthetic, historical, or partner-approved data.
4
Report accuracy, safety, usability, equity, and implementation findings.
Build public health AI around real work and accountable review.
APHI is for institutions that want useful AI support with evidence standards, review boundaries, and implementation discipline.