Appendix O: Teaching with the Public Health AI Handbook
MPH programs, preventive medicine residencies, epidemiology courses, and health department training teams can teach public health AI from the Public Health AI Handbook without asking permission or paying a fee. The materials below are built for a course director or workforce lead with limited calendar time: short session tracks, maps to the competency frameworks programs already report against, journal club questions for every chapter, case discussions, and assessment items with answer points. The full 12-week seminar, with a grading rubric, a learner AI-use policy, and an assignment-by-competency table, is the Course Syllabus Template.
These materials support local teaching. They are not an accredited course and do not replace a program’s own curriculum review or competency documentation.
Who the Materials Serve
No programming background is needed for the four-session track. The practical chapters add Python work for learners who want it.
| Setting | Suggested starting point | Typical slot |
|---|---|---|
| MPH elective or module in an informatics, epidemiology, or policy course | Eight-session track, or the 12-week syllabus | Half-semester elective or course unit |
| MPH core courses that need one AI session | One session from the four-session track | Single class |
| Preventive medicine and public health residencies | Four-session track | Didactic series |
| Health department staff training | Four-session track, mapped to the Core Competencies | Brown-bag series or training day |
| Epidemiology or surveillance journal club | Chapter questions in the journal club guide | Monthly |
| Doctoral seminars and DrPH courses | Eight-session track with the journal club guide | Seminar |
Session Tracks
Each session runs 60 to 90 minutes: learners read the assigned chapter sections beforehand, the instructor frames the question for ten minutes, the group works through one case or one primary study cited in the chapter, and the session closes with one or two assessment items. The competency columns name where a session gives learners practice; the program decides whether an activity counts as an assessment for its own competency matrix.
Four-Session Track
| Session | Question for the session | Read before | Case for discussion | CEPH competency | Core Competencies domain |
|---|---|---|---|---|---|
| 1 | What would show that an AI tool works in public health practice? | The Data Problem in Public Health AI; Evaluating AI in Public Health Practice | Google Flu Trends | 4, 11 | Data Analytics and Assessment |
| 2 | When does an automated signal deserve an epidemiologist’s time? | AI in Disease Surveillance and Outbreak Detection; Epidemic Forecasting: Methods, Evaluation, and AI | BlueDot and early COVID-19 detection | 1, 3 | Public Health Sciences |
| 3 | Who bears the errors, and who decides whether the tool is used? | AI Ethics, Bias, and Equity in Public Health; Health AI Privacy, Security, and Data Governance; AI Governance in Public Health: Regulation and Oversight | Allegheny Family Screening Tool | 6, 12, 15 | Health Equity; Policy Development and Program Planning |
| 4 | Why do working prototypes fail in health departments? | AI Deployment in Healthcare: Why Most Prototypes Fail; AI in Public Health Emergency Operations | COVID-19 prediction models | 22 (with a system map) | Leadership and Systems Thinking; Management and Finance |
Eight-Session Track
Twelve Sessions or a Semester
The Course Syllabus Template sets out a 12-week seminar with weekly readings, discussion questions, assignments, a grading rubric, a learner AI-use policy, a table of assignments that can assess CEPH competencies, and a 6-week intensive version. Programs that start with the four- or eight-session track can move to it without changing readings.
Competency Maps
CEPH MPH Foundational Competencies
CEPH’s 2024 accreditation criteria require an accredited school or program to assess every MPH student at least once on each of 22 foundational competencies, through a specific assessment activity it documents in a competency matrix; assessment in group work needs a method for judging each student individually, should take place outside the internship, and must take place outside the integrative learning experience (CEPH, 2024). The table shows where the handbook can supply the material for such an activity.
| CEPH foundational competency | Chapters that supply the material | Activity that could assess it |
|---|---|---|
| 1. Apply epidemiological methods to settings and situations in public health practice | Surveillance; forecasting; genomics | Critique of an automated outbreak signal against case definitions and baseline counts |
| 2. Select quantitative and qualitative data collection methods appropriate for a given public health context | The Data Problem in Public Health AI | Data plan for a proposed AI tool, including what each source can and cannot measure |
| 3. Analyze quantitative and qualitative data using biostatistics, informatics, computer-based programming, and software | Building Your First Public Health AI Project; AI-Assisted Coding for Public Health Analysis | Reproducible baseline model with documented validation |
| 4. Interpret results of data analysis for public health research, policy, or practice | Evaluating AI in Public Health Practice | Written interpretation of a model’s external validation for a health officer |
| 6. Discuss the means by which structural bias, social inequities, and racism undermine health and create challenges to achieving health equity | AI Ethics, Bias, and Equity in Public Health | Bias audit of a published risk tool, including the choice of label |
| 11. Select methods to evaluate public health programs | Evaluating AI in Public Health Practice | Evaluation plan with operational and population outcomes |
| 12. Discuss the policy-making process, including the roles of ethics and evidence | AI Governance in Public Health: Regulation and Oversight | Brief comparing regulation with agency governance for one tool |
| 15. Evaluate policies for their impact on public health and health equity | AI Governance in Public Health: Regulation and Oversight; AI in Global Health and Equity | Equity impact assessment of an AI procurement policy |
| 18 and 19. Communication strategies for different audiences; audience-appropriate content for a non-academic audience | AI, Health Misinformation, and the Infodemic | Plain-language explainer or prebunking message, presented orally |
| 22. Apply a systems thinking tool to visually represent a public health issue in a format other than standard narrative | AI Deployment in Healthcare: Why Most Prototypes Fail | Causal loop diagram or system map of how a tool would change surveillance or response |
Two CEPH notes affect design. Competency 22 is not met by a logic model or evidence table, and competency 21 requires direct engagement with someone from another profession or sector; role-play does not count (CEPH, 2024). An AI module that pairs MPH students with clinicians, data engineers, or agency IT staff can address competency 21 when the program documents that engagement.
Core Competencies for Public Health Professionals
For health department training, the Council on Linkages Between Academia and Public Health Practice adopted the current Core Competencies on October 21, 2021: 56 competency statements in eight domains, with three tiers that describe front line and program support, program management and supervisory, and senior management and executive leadership responsibilities (Council on Linkages, 2021).
The four-session track suits front line and program staff. For managers and executives, sessions 3 and 4 carry the decisions those tiers own: procurement, governance, and when to stop a tool.
Journal Club Guide
The chapter is the briefing; the paper is the evidence. Choose one primary study cited in the chapter, assign it with the chapter, and give the group 60 minutes in three parts: what the study claims (10 minutes), whether its design supports that claim (30 minutes), and whether the result would hold in the group’s own jurisdiction and population (20 minutes). The chapter questions further down are written for that last part and can be used with any study the chapter cites.
Starter Papers
Six studies the handbook discusses in detail make good first sessions. Each is read with its chapter, and the question gives the middle 30 minutes a focus.
| Study | Read with | Design question |
|---|---|---|
| Lazer et al., 2014, Science: the analysis of Google Flu Trends | AI in Disease Surveillance and Outbreak Detection | The authors warned against treating big data as a substitute for traditional data collection. Which current surveillance signals carry the same risk, and what ongoing validation would catch drift? |
| Cramer et al., 2022, PNAS: evaluation of the US COVID-19 Forecast Hub | Epidemic Forecasting: Methods, Evaluation, and AI | Forecasts were scored with the weighted interval score across horizons. Why does strong average ensemble performance not guarantee accuracy for one jurisdiction in one week? |
| Obermeyer et al., 2019, Science: racial bias in a widely used risk algorithm | AI Ethics, Bias, and Equity in Public Health | The algorithm used spending as a proxy for need. Which proxies in your agency’s data stand in for need, and how would you test them for the same failure? |
| Wynants et al., 2020, BMJ: systematic review and critical appraisal of COVID-19 prediction models | Public Health AI Case Studies | Which appraisal criteria from the review would you apply to a vendor’s model before using it in response planning? |
| Roozenbeek et al., 2022, Science Advances: inoculation videos across seven preregistered studies | AI, Health Misinformation, and the Infodemic | The studies measured discernment between manipulative and neutral content. What outcome and follow-up period would show the effect carries over to a public health campaign? |
| Abràmoff et al., 2018, npj Digital Medicine: pivotal study of autonomous diabetic retinopathy detection | AI in Diagnostic and Clinical Decision Support | What would a screening program need to match (cameras, operators, population, referral pathway) for the study’s results to apply at population scale? |
Questions by Chapter
| Chapter | Discussion questions |
|---|---|
| History of AI in Public Health | Which barriers that limited earlier systems still apply to current tools in health departments? What evidence would show that a new approach has endured rather than repeated an earlier wave? |
| Machine Learning Fundamentals | Was the study’s problem supervised, unsupervised, or something else, and was that the right choice? What simpler baseline should the model have been compared with? |
| The Data Problem in Public Health AI | Which delays, gaps, and reporting changes in the data could the model mistake for a real trend? Which populations are missing from the source, and how would that change the output? |
| AI in Disease Surveillance and Outbreak Detection | Who verified the signal, and how long did verification take? What response capacity did the system assume, and does your agency have it? |
| Epidemic Forecasting: Methods, Evaluation, and AI | How was the forecast scored against what happened, and over what horizon? How did behavior change or pathogen evolution affect accuracy? |
| AI in Genomic Surveillance and Pathogen Analysis | Which epidemiologic evidence was combined with the genomic data, and could the genomic result stand without it? How does sequencing coverage in the study compare with yours? |
| Geospatial AI and Spatial Epidemiology | Were the holdout areas spatially separate from the training areas? Which decision would the map change, and what is the cost of a wrongly flagged neighborhood? |
| AI in Diagnostic and Clinical Decision Support | Did performance hold on external data from a different population? What would a public health program need to add for a clinical tool to work at population scale? |
| AI for Substance Use and Overdose Prevention | What action follows when the model flags a person or a place, and is treatment capacity available? How were stigma and privacy addressed in the design? |
| Evaluating AI in Public Health Practice | Which claim does the study support: technical performance, transportability, operational utility, or population impact? What evaluation would your agency require before adopting the tool? |
| AI Ethics, Bias, and Equity in Public Health | Which label or proxy could encode existing inequity? Which subgroups were analyzed separately, and which were not? |
| Health AI Privacy, Security, and Data Governance | Which combination of released fields could re-identify a person in your jurisdiction? What agreement and safeguards govern the data flow described? |
| AI Safety in Healthcare: Protecting Patients and Populations | How could the model, the people using it, and the workflow combine to cause harm? Which controls were built before deployment, and which were added after a failure? |
| AI Deployment in Healthcare: Why Most Prototypes Fail | Who owned the system after the pilot ended? How was drift detected, and what was the predefined response? |
| AI in Public Health Emergency Operations | Where in the incident structure did the AI output enter, and who was accountable for acting on it? How were sources, timestamps, and uncertainty recorded for the decision log? |
| Your AI Toolkit for Public Health | Could another analyst reproduce the study’s results from the code and environment described? Which tool choice would be hardest to maintain in a health department? |
| Building Your First Public Health AI Project | Was the project scoped around a decision someone would make? Did the model beat a simple baseline on data it had not seen? |
| AI-Assisted Coding for Public Health Analysis | How were errors in AI-generated code found, and which ones would have changed the results? What review step should an agency require before such code runs on surveillance data? |
| Emerging AI Technologies for Public Health | What decision does the technology serve, and what is the non-AI baseline? At what stage of evidence maturity is the claim? |
| AI in Global Health and Equity | What did the service need beyond the model (inputs, staff, connectivity, maintenance)? Who governs the system and its data after external funding ends? |
| AI Governance in Public Health: Regulation and Oversight | Which duties in the study come from regulation and which from organizational governance? Who in your agency can stop the tool? |
| AI, Health Misinformation, and the Infodemic | Was the intervention prebunking, debunking, or detection, and what outcome was measured? Who delivered the message, and would that messenger be trusted in your community? |
| Large Language Models in Public Health: Theory and Practice | Which outputs in the study were checked against primary sources, and how often were they wrong? What data could the model see, and under what agreement? |
| AI-Driven Behavioral Interventions in Public Health | Where does the intervention move from persuasion toward manipulation? Was the effect measured against an active comparator or only against no intervention? |
Case Discussions
The Public Health AI Case Studies appendix holds full cases, and the Case Study Library - Overview summarizes them. A role-based format works well: assign learners the roles of health officer, epidemiologist, data engineer, community representative, and agency counsel, and have each argue from that role before the group reaches a decision.
| Case | Prompts |
|---|---|
| BlueDot and early COVID-19 detection | What did the early alert make possible, and what still depended on people and agencies? How should an agency weigh a commercial alert against its own surveillance? |
| Google Flu Trends | What caused the estimates to drift, and how could the drift have been caught sooner? Which current tools rely on similar signals? |
| ProMED-mail and HealthMap | Which tasks did automation take on, and which stayed with human moderators? What would be lost if the human layer were removed? |
| COVID-19 prediction models | Why did so many models have limited impact on decisions? What standards should apply to models built during an emergency? |
| Allegheny Family Screening Tool | Which families bear the consequences of false positives and false negatives? What governance and transparency would the group require before a similar tool is used? |
| MomConnect in South Africa | What made national scale possible? Which parts of the service would a model change, and which would it leave untouched? |
Assessment Questions
The items below suit short-answer quizzes, oral examination prompts, or the closing minutes of a session. Answer points are collapsed so the page can be projected.
- Explain the difference between surveillance and forecasting, and why epidemic forecasting is harder than weather forecasting.
- What does the history of Google Flu Trends teach about models built on behavioral data?
- A vendor reports a high validation score for an outbreak-detection tool. What does that score show, and what does it leave unanswered for your agency?
- A risk tool used health care spending as a stand-in for need. How did that choice produce biased results?
- Why is removing names from a dataset insufficient to protect privacy?
- A colleague wants to paste line-list data into a consumer chatbot to summarize it. What is the correct response?
- During an emergency, an AI system produces a summary for the incident briefing. What role may it play, and what controls are required?
- A deployed model’s performance declines over a year. Name the types of drift that could explain it and the monitoring that should already exist.
- Distinguish regulation from organizational governance for an AI tool used by a health department.
- Compare prebunking with debunking, and explain why the messenger matters.
- Surveillance detects what is happening now; forecasting predicts what will happen. Forecasting is harder because people change behavior in response to predictions, pathogens evolve, and small errors grow over time (Epidemic Forecasting: Methods, Evaluation, and AI).
- Signals that depend on human behavior and platform design change over time, so a model that fit past data can drift without warning; validation against conventional surveillance has to continue after launch (AI in Disease Surveillance and Outbreak Detection).
- Technical performance in one setting does not establish transportability, operational utility, population impact, equity, or lifecycle safety (Evaluating AI in Public Health Practice).
- Groups that receive less care generate less spending at the same level of need, so a model trained on spending rates them as lower risk (AI Ethics, Bias, and Equity in Public Health).
- Combinations of other fields, such as ZIP code, date of birth, and sex, can single out individuals, and risk depends on the released fields and on other data available to an attacker (Health AI Privacy, Security, and Data Governance).
- Protected health information does not go into consumer language models. Enterprise use still needs a documented data flow, applicable agreements and safeguards, and organizational approval (Large Language Models in Public Health: Theory and Practice).
- AI may support the common operating picture through triage, retrieval, summarization, and mapping. It does not own the incident, set objectives, or replace accountable command. Controls include approved sources, source links, timestamps, stated uncertainty, human review, and decision logs (AI in Public Health Emergency Operations).
- Data drift, concept drift, and label drift. Monitoring with predefined thresholds, an operational owner, versioning, and a predefined response should be in place before deployment (AI Deployment in Healthcare: Why Most Prototypes Fail).
- Regulation sets legally enforceable duties. Organizational governance sets decision rights, evidence thresholds, and operating controls, including for tools that no regulator has reviewed (AI Governance in Public Health: Regulation and Oversight).
- Prebunking builds resistance before exposure; debunking corrects after exposure and can amplify a claim if poorly targeted, although true backfire is rare. Trusted messengers change how a correction is received (AI, Health Misinformation, and the Infodemic).
Citing the Version Used in a Course
The handbook is revised continuously, so a syllabus should record the date on which learners read it. The Changelog lists what changed after that date.
Tegomoh B. The Public Health AI Handbook: Evaluating AI Tools for Public Health Practice. 2025. Accessed [date]. doi:10.5281/zenodo.18263442
Reuse and Adaptation
The handbook text is licensed under the Creative Commons Attribution 4.0 International License (Creative Commons, CC BY 4.0). Programs and agencies may copy, excerpt, translate, and adapt it into slides, course packs, learning management systems, training modules, and assessments, including for courses that charge tuition, provided they give credit, link to the license, and note any changes. An attribution line can read:
Adapted from The Public Health AI Handbook by Bryan Tegomoh, MD, MPH (publichealthaihandbook.com), licensed under CC BY 4.0. Changes: [describe].
Studies, figures, code, and quotations from other sources keep their own licenses; cite them from the original.
Contact
Using this handbook in a course, program, or agency training? Tell us. A line on the program, the sessions used, and what did not work helps decide what to revise next.