Global Health AI Data Governance

Data sovereignty, large language models, and One Health governance for global public health AI. The material is maintained separately so each operational question has a stable, focused reference.

Learning Objectives
  • Identify the evidence and controls relevant to this decision area
  • Distinguish technical performance from operational and population impact
  • Apply the included framework without extending claims beyond the cited evidence

Use explicit targets, populations, thresholds, and decision consequences. Require external evidence and local monitoring where deployment can affect people or programs. Preserve uncertainty and document limits.

Introduction

This focused reference is part of the broader Global Health AI Data Governance overview. It preserves the detailed methods, examples, and exercises while reducing page size and improving direct navigation.

Data Governance in Global Health

The Governance Challenge

Issue: International health collaborations involve data sharing across borders, raising questions about: - Sovereignty: Who owns health data? Who controls it? - Privacy: How to protect individual privacy across jurisdictions? - Benefit sharing: How to ensure LMICs benefit from research using their data? - Capacity: How to ensure fair partnerships when capacity is unequal?

Data Colonialism Risk

Data Colonialism

Data colonialism refers to the extraction of data from LMICs by high-income countries/companies for their benefit, with limited benefit to the source populations.

Examples: - Genomic data collected from African populations, stored in Western biobanks, used for drug development that benefits high-income populations - Health data from LMIC electronic health records used to train commercial AI models sold back to LMICs at high prices - Research collaborations where LMIC partners collect data but have no say in how it’s used

Result: Value extraction without benefit sharing - repeating colonial patterns in the digital age.

Principles for Equitable Data Governance

class EquitableDataGovernance:
 """
 Framework for equitable data governance in global health AI

 Based on:
 - CARE Principles (Collective benefit, Authority to control, Responsibility, Ethics)
 - FAIR Principles (Findable, Accessible, Interoperable, Reusable)
 - Data sovereignty principles
 """

 def __init__(self, data_source_country, data_user):
  self.data_source = data_source_country
  self.data_user = data_user
  self.governance_framework = self.create_framework()

 def create_framework(self):
  """
  Create data governance framework

  Covers:
  - Consent and authorization
  - Data access and use
  - Benefit sharing
  - Capacity building
  - Intellectual property
  """
  framework = {
   'consent_authorization': {
    'individual_consent': 'Required for identifiable data',
    'community_consent': 'Required for community-level data',
    'institutional_authorization': 'Required from data source institution',
    'government_authorization': 'May be required for export'
   },
   'data_access_use': {
    'data_location': 'Preference for local storage and processing',
    'data_transfer': 'Minimize cross-border transfer when possible',
    'access_control': 'Source institution retains access control',
    'use_restrictions': 'Data use limited to agreed purposes',
    'secondary_use': 'Requires additional authorization',
    'commercial_use': 'Requires separate agreement with benefit sharing'
   },
   'benefit_sharing': {
    'authorship': 'Local partners as co-authors on all publications',
    'ip_sharing': 'Joint IP ownership for innovations',
    'revenue_sharing': 'Royalties from commercial applications',
    'capacity_building': 'Commitment to train local team',
    'data_return': 'Analysis results returned to source community'
   },
   'capacity_building': {
    'training': 'Local team trained in AI methods',
    'infrastructure': 'Investment in local computing infrastructure',
    'sustainability': 'Plan for local capacity to continue work',
    'knowledge_transfer': 'Code, models, documentation shared with local team'
   },
   'intellectual_property': {
    'ownership': 'Joint ownership of AI models and algorithms',
    'licensing': 'Preferential licensing to source country',
    'patents': 'Joint patents with benefit sharing',
    'open_source': 'Preference for open-source when possible'
   }
  }

  return framework

 def create_data_sharing_agreement(self):
  """
  Template for equitable data sharing agreement

  Based on actual agreements from equitable global health partnerships
  """
  agreement = f"""
  DATA SHARING AGREEMENT

  Between: {self.data_source} (Data Source)
  And: {self.data_user} (Data User)

  1. DATA DESCRIPTION
  [Describe data: type, volume, collection methods, etc.]

  2. PURPOSE
  [Specific research questions or applications]

  3. DATA ACCESS AND USE
  - Data will be stored in {self.data_source} or mutually agreed secure location
  - {self.data_user} granted access for specified purposes only
  - Any additional use requires written approval from {self.data_source}
  - Data cannot be shared with third parties without written approval

  4. GOVERNANCE
  - Joint steering committee with equal representation
  - {self.data_source} retains ultimate authority over data use
  - Disputes resolved through [mediation process]

  5. CAPACITY BUILDING
  - {self.data_user} commits to train [X] local data scientists
  - Investment of $[Y] in local computing infrastructure
  - All code, models, and documentation shared with {self.data_source}
  - Plan for sustainable local capacity within [Z] years

  6. INTELLECTUAL PROPERTY
  - Joint ownership of all AI models and algorithms
  - Joint authorship on all publications
  - Patents filed jointly with 50/50 ownership
  - Preferential licensing to {self.data_source} for local use

  7. BENEFIT SHARING
  - {self.data_source} researchers co-authors on all publications
  - 50% of any commercial revenue returned to {self.data_source}
  - AI tools made available to {self.data_source} at no cost
  - Analysis results returned to source communities

  8. PRIVACY AND SECURITY
  - Data de-identified per {self.data_source} regulations
  - Security measures: [encryption, access controls, audit logs]
  - Compliance with {self.data_source} data protection laws

  9. DURATION AND TERMINATION
  - Agreement valid for [X] years
  - {self.data_source} can terminate with 30 days notice
  - Upon termination, {self.data_user} deletes all data

  10. ACCOUNTABILITY
  - Annual progress reports to {self.data_source}
  - External evaluation at [Y] years
  - Community advisory board provides oversight
  """

  return agreement

 def assess_partnership_equity(self):
  """
  Assess whether partnership is equitable

  Red flags:
  - Data source has no say in data use
  - No local capacity building
  - No benefit sharing
  - One-way knowledge transfer
  - Exploitative authorship practices
  """
  assessment = {
   'data_control': {
    'question': 'Does data source retain control over data use?',
    'red_flag': 'Data user has unilateral control',
    'green_flag': 'Data source retains ultimate authority'
   },
   'capacity_building': {
    'question': 'Is there meaningful capacity building?',
    'red_flag': 'No training or infrastructure investment',
    'green_flag': 'Substantial investment in local capacity'
   },
   'benefit_sharing': {
    'question': 'Are benefits shared equitably?',
    'red_flag': 'All benefits go to data user',
    'green_flag': 'Benefits shared through authorship, IP, revenue'
   },
   'knowledge_transfer': {
    'question': 'Is knowledge transferred to source?',
    'red_flag': 'One-way transfer (data out, nothing back)',
    'green_flag': 'Code, models, methods shared with source'
   },
   'sustainability': {
    'question': 'Will local capacity be sustainable?',
    'red_flag': 'Dependent on external expertise indefinitely',
    'green_flag': 'Clear path to local ownership and sustainability'
   }
  }

  return assessment

# Example: Assess proposed data partnership

governance = EquitableDataGovernance(
 data_source_country='Kenya',
 data_user='University of Example'
)

# Check framework
framework = governance.create_framework()
print("Data Governance Framework:")
print(json.dumps(framework, indent=2))

# Generate agreement template
agreement = governance.create_data_sharing_agreement()
print("\nData Sharing Agreement Template:")
print(agreement)

# Assess partnership equity
equity_assessment = governance.assess_partnership_equity()
print("\nPartnership Equity Assessment:")
for dimension, criteria in equity_assessment.items():
 print(f"\n{dimension.upper()}:")
 print(f" Question: {criteria['question']}")
 print(f" Red flag: {criteria['red_flag']}")
 print(f" Green flag: {criteria['green_flag']}")

Large Language Models in Global Public Health

Large language models (LLMs) represent a distinct paradigm from the narrow AI systems covered earlier in this chapter. Unlike task-specific diagnostic tools (chest X-ray AI, malaria detection), LLMs are general-purpose language systems that can support surveillance reporting, outbreak communications, health education, community health worker triage, and population health decision support through natural language interaction.

African Evidence: Trials and Benchmarks Answer Different Questions

Two 2026 examples show why local evidence must be matched to the claim. A pragmatic cluster-randomized trial in Kenyan primary care evaluated generative AI decision support in routine facilities. Several expert-rated process measures improved, but the prespecified 14-day treatment-failure outcome did not differ significantly, so the study does not establish improved patient outcomes (Agweyu et al., 2026).

IyawoBench evaluates triage decisions for synthetic febrile-illness cases derived from Nigerian primary-care distributions. Its local disease context is useful, but synthetic cases, limited reference labeling, and a fixed evaluation design do not establish prospective safety. The v2 update adds measures for escalation, downgrade, middle-tier instability, and expected deployment cost, which should be interpreted with the original benchmark rather than as an independent validation (Gabriel et al., 2026, preprint; Gabriel and Olawuyi, 2026, preprint).

Together, these studies support a staged program: locally grounded benchmarks for failure discovery, independent clinician labeling and transportability testing, then prospective comparison of workflow, safety, resource use, equity, and patient outcomes. The evaluation framework is detailed in Evaluating AI Systems.

This versatility makes LLMs particularly relevant for public health in resource-limited settings: they can potentially address multiple gaps simultaneously (surveillance, communication, triage), augment overstretched public health workforces, and scale to diverse populations without task-specific retraining. However, this same flexibility creates unique risks including hallucinations (confident but false information), privacy concerns with population health data, and deployment challenges that differ from narrow AI systems.

Open-Weight Models: Opportunity for Public Health Systems

Computational efficiency breakthrough for resource-constrained health departments:

Open-weight LLMs such as DeepSeek, Llama 3, and Mistral offer promise for public health agencies with limited IT budgets by dramatically reducing computational requirements and costs while achieving performance comparable to proprietary models (Ong et al., Nature Health, 2026).

DeepSeek deployment in health systems:

DeepSeek, developed under hardware access limitations, has been deployed in 90 Chinese tertiary hospitals since January 2025 for clinical decision support, patient communication, and administrative functions. For public health applications, the key advantage is cost: estimated at only 6.71% of proprietary model costs (OpenAI o1), making LLM capabilities accessible to health departments that cannot afford enterprise AI contracts (Chen et al., Journal of Medical Systems, 2025).

The “too fast, too soon” warning applies to public health:

Chinese medical researchers have raised substantial concerns about rapid deployment without adequate validation. These concerns extend to public health applications: using LLMs for surveillance triage, outbreak communications, or policy recommendations without prospective validation could lead to systematic errors at population scale (Wong et al., JAMA, 2025).

Public health deployment considerations:

Advantage for Public Health Risk for Public Health
Cost efficiency: Accessible to under-resourced health departments Safety monitoring gaps: Population-scale errors if hallucinations go undetected
Local deployment: Can run on health department servers, protecting sensitive surveillance data Integration complexity: Requires technical capacity often limited in smaller health departments
No vendor lock-in: Avoids dependency on commercial platforms that may discontinue public health pricing Version control challenges: No centralized updates across decentralized public health systems
Customization potential: Can be fine-tuned on local disease patterns, languages, cultural contexts Hallucination risks in communications: False health information at scale during outbreaks

Public health implication:

Open-weight LLMs offer genuine opportunity to extend public health capacity in resource-limited settings, but deployment must include rigorous validation on public health tasks (surveillance coding, risk communication, triage accuracy) with human oversight. The rapid scale-up in clinical settings without prospective trials offers a cautionary example for public health.

LLM-Enhanced Public Health Applications

MomConnect Enhanced with LLMs: SMS-Based Maternal Health at Scale

The MomConnect SMS chatbot in South Africa (covered in Case Study earlier) has evolved to incorporate LLM capabilities for more sophisticated triage while maintaining the low-tech SMS infrastructure that enabled 4 million+ user reach (Ong et al., Nature Health, 2026).

Public health implementation:

  • Base system remains SMS: Preserves accessibility for users with basic feature phones (no smartphone required, works on 2G networks)
  • LLM layer for triage: Natural language processing identifies urgency signals in patient messages (“severe headache + swelling” → preeclampsia risk)
  • Human escalation pathway: Urgent cases flagged for immediate nurse review, not autonomous LLM response
  • Population-scale impact: Maintains simplicity for 900,000 active users while backend LLM complexity improves triage accuracy

Why this hybrid approach succeeds:

The system uses LLM capabilities where they add value (understanding natural language in 11 South African languages, detecting urgency patterns) while avoiding LLM weaknesses (autonomous clinical decisions, internet dependency). The human-in-the-loop design catches LLM errors before they reach patients, critical for maternal-fetal health outcomes at population scale.

Public health metrics:

From the original MomConnect evaluation (2018 RCT, n=8,486):

  • Prenatal care utilization: +8 percentage points (76% vs. 68%, p<0.001)
  • HIV testing uptake: +8 percentage points (93% vs. 85%)
  • Postnatal visit within 6 weeks: +18 percentage points (82% vs. 64%, p<0.001)
  • Cost-effectiveness: $1.20 per woman per pregnancy, with 15:1 ROI from avoided emergency costs

Lessons for public health LLM deployment:

SMS + LLM backend hybrid architecture allows sophisticated triage at population scale without requiring smartphone adoption. This model is directly applicable to outbreak communications, syndromic surveillance reporting, and community health worker support in LMICs.

DeepDR-LLM: Hybrid AI for Population Diabetes Management

A multimodal system combining image-based deep learning with language models demonstrates how LLMs can augment primary care capacity for chronic disease management in resource-limited settings, directly relevant to public health prevention programs.

System design for population health:

DeepDR-LLM integrates diabetic retinopathy screening (image AI) with personalized diabetes management recommendations (LLM) adapted to Chinese primary care settings. The system was trained on 371,763 real-world management recommendations from 267,730 participants (Li et al., Nature Medicine, 2024).

Prospective validation results (public health outcomes):

In a prospective study comparing patients under unassisted primary care physicians (n=397) versus PCP + DeepDR-LLM support (n=372):

  • Medication adherence: Patients with newly diagnosed diabetes showed significantly better self-management behaviors throughout follow-up (p<0.05)
  • Diabetic retinopathy referrals: Patients with referable DR were more likely to adhere to referrals (p<0.01)
  • Diagnostic accuracy: PCP accuracy for identifying referable DR increased from 81.0% unassisted to 92.3% with DeepDR-Transformer assistance

Public health significance:

Hybrid systems combining task-specific AI (image analysis) with general-purpose LLMs (clinical guidance) may offer more reliable population health support than LLMs alone, reducing hallucination risks while maintaining personalization capabilities for chronic disease management programs.

AfriMed-QA: Validating LLMs for Diverse Populations

The evaluation gap in public health AI:

Most medical AI benchmarks (USMLE, MedQA) are developed from Western medical education contexts, using disease patterns, treatment options, and healthcare infrastructures common in high-income countries. Public health systems deploying LLMs for surveillance, communications, or decision support need benchmarks reflecting local disease burdens, resource constraints, and cultural contexts.

AfriMed-QA: First pan-African medical AI benchmark:

AfriMed-QA is the first large-scale pan-African, multi-specialty medical question-answer dataset designed to evaluate LLM performance in contexts relevant to African healthcare (Olatunji et al., ACL 2025).

Dataset composition:

  • ~15,000 questions spanning 32 clinical specialties
  • Contributors: 621 medical professionals from over 60 medical schools across 15 African countries
  • Question types: Expert multiple-choice questions (4,000+), short-answer questions (1,200+), consumer health queries (10,000)
  • Recognition: Awarded Best Social Impact Paper at ACL 2025

Why this matters for public health:

When 30 different LLMs were evaluated using AfriMed-QA, performance patterns differed substantially from Western benchmarks. Models that excelled on USMLE showed weaker performance on Africa-specific questions, revealing gaps in knowledge about:

  • Endemic infectious diseases: Malaria, schistosomiasis, trypanosomiasis (neglected tropical diseases critical for public health surveillance)
  • Resource-adapted protocols: Treatment options available in LMIC settings, not just what’s optimal in HICs
  • Traditional medicine interactions: Community health practices that affect public health program design
  • Local drug formularies: What’s actually available for public health programs to recommend
  • Cultural and linguistic considerations: Critical for outbreak communications and health education

Public health application:

Before deploying any LLM in public health systems serving African or LMIC populations, validation against AfriMed-QA or similar context-specific benchmarks is essential. Performance on Western medical exams does not guarantee performance in LMIC public health contexts.

Access for public health practitioners:

The dataset is publicly available at afrimedqa.com and through the GitHub repository, enabling health departments and researchers to evaluate and fine-tune LLMs for their specific population contexts before deployment in surveillance systems, health communication platforms, or decision support tools.

Environmental Justice and LLM Sustainability in Public Health

The hidden cost of computational intensity:

Training and deploying LLMs consume vast computational resources, with environmental impacts that disproportionately affect LMICs even when the technology is developed and deployed primarily in high-income countries.

Energy and emissions profile:

  • Training emissions: A single large model training run can emit hundreds of tons of CO₂ equivalent
  • Inference costs: While individual queries consume relatively little energy, cumulative daily use across public health applications (surveillance coding, outbreak communications, health education chatbots) scales dramatically
  • Water consumption: Data centers require substantial water for cooling; water scarcity is more acute in many LMIC regions
  • Hardware lifecycle: Rare-earth mining for GPUs and electronic waste disposal create environmental burdens concentrated in resource-extraction regions

The public health equity dimension:

LMICs contribute minimally to AI development emissions but bear disproportionate climate impacts from environmental degradation and water scarcity. As public health systems in high-income countries adopt LLM-based surveillance tools, communications platforms, and decision support systems, the cumulative carbon footprint grows while benefits accrue primarily to well-resourced settings.

Climate-health nexus:

Public health practitioners understand the health impacts of climate change (heat-related illness, vector-borne disease expansion, food insecurity). LLM deployment in public health should be evaluated through climate-health lens: Do the population health benefits justify the environmental costs, particularly for applications in climate-vulnerable regions?

Sustainable deployment strategies for public health:

  1. Edge deployment over cloud: Running LLMs locally on health department servers reduces data transfer energy costs and supports data sovereignty for sensitive surveillance data
  2. Model efficiency: Smaller, task-optimized models rather than general-purpose large models where appropriate (syndromic surveillance may not need GPT-4 scale)
  3. Renewable energy: Data centers powered by solar, wind, or hydroelectric rather than fossil fuels
  4. Shared infrastructure: Regional public health data centers serving multiple countries, reducing redundant computational capacity

Policy implication:

As LLMs are evaluated for public health applications, environmental sustainability should be explicit criterion, alongside population health effectiveness and cost-effectiveness. The most sustainable public health AI solution may be the simplest one that achieves public health objectives, not the most sophisticated.

Global Governance for Public Health AI

WHO Global Initiative on AI for Health (GI-AI4H):

Launched in July 2023 by WHO, the International Telecommunication Union (ITU), and the World Intellectual Property Organization (WIPO), GI-AI4H provides an institutional framework for coordinating responsible AI development and deployment globally (WHO/ITU/WIPO, 2023).

Strategic focus areas for public health:

  1. Standards development: International standards for AI evaluation, ethics, clinical validation, and benchmarking applicable to public health surveillance and decision support systems
  2. Knowledge transfer: Facilitating data sharing, collaboration, and best practice dissemination among public health stakeholders worldwide
  3. Health system strengthening: Prioritizing low- and middle-income countries through a scaling program initially targeting 12-18 countries with relevant AI use cases

Public health applications:

GI-AI4H standards can guide public health departments deploying LLMs for surveillance coding (ICD-10, SNOMED CT automation), outbreak risk communication (multilingual health messaging), and health education (chatbot accuracy standards).

Lancet Global Health Commission on AI and HIV:

This Commission synthesizes evidence on AI’s economic and health impacts across different settings, with explicit focus on guiding responsible AI model development and creating actionable guidance for stakeholders in regulation and adoption. HIV surveillance, prevention, and treatment programs are early adopters of AI in public health, providing lessons for broader LLM deployment (Lancet Global Health, ongoing).

Gates Foundation AI equity initiatives:

Philanthropic funding targeting AI equity, language inclusivity, and ensuring equitable access to AI benefits in all settings. Projects focus on developing language technologies for under-resourced languages (critical for multilingual public health communications) and supporting local capacity building (Gates Foundation, 2024-2026).

WHO Global Strategy on Digital Health (2020-2027):

The Seventy-eighth World Health Assembly extended the Global Strategy on Digital Health from 2025 through 2027 in May 2025, with a follow-up framework for 2028-2033 under development (WHO, May 2025). Since its adoption, 129 countries have established national digital health strategies. The strategy covers national digital health infrastructure, interoperability standards, and health data governance, providing the overarching framework within which AI-specific initiatives (GI-AI4H, WHO ethics guidance) operate.

Implication for public health practitioners:

These frameworks provide actionable guidance for health departments deploying LLMs in global public health contexts. Rather than navigating regulatory ambiguity alone, using WHO standards, Lancet Commission recommendations, and philanthropic partnership opportunities can accelerate responsible implementation while ensuring equity.

Practical Guidance for LLM Deployment in Public Health Systems

Pre-deployment checklist for health departments:

Before deploying any LLM system in public health applications:

  1. Validate on public health tasks (surveillance coding accuracy, risk communication clarity, triage appropriateness), not just general medical knowledge
  2. Validate on context-specific populations (AfriMed-QA for Africa; similar datasets for Asia, Latin America where available)
  3. Assess infrastructure requirements (internet connectivity for cloud models, computational capacity for edge deployment, electricity reliability)
  4. Evaluate language support (does LLM support local languages and dialects, or only English? Critical for health communications)
  5. Verify safety mechanisms (how does system handle uncertainty, avoid hallucinations, escalate to public health professionals?)
  6. Establish oversight (who reviews LLM outputs before population-level communications? Who monitors surveillance coding accuracy?)
  7. Plan for sustainability (who maintains system when external funding ends? What is long-term cost structure for health department budget?)
  8. Address data sovereignty (where is surveillance data processed and stored? Does this comply with local regulations and public health confidentiality requirements?)
  9. Measure environmental impact (what are energy/water requirements? Can renewable energy support deployment? Do public health benefits justify environmental costs?)

Red flags warranting rejection:

  • LLM vendor cannot demonstrate performance on public health tasks (surveillance, risk communication, triage)
  • System requires always-on internet connectivity in settings with unreliable access (fails during outbreaks when most needed)
  • No mechanism for public health professional oversight of LLM outputs before population-level dissemination
  • Deployment plan lacks sustainable financing beyond external grant period (common failure mode in public health pilots)
  • Surveillance or population health data will be processed on foreign servers without clear data protection agreements
  • Vendor dismisses environmental impact questions or lacks sustainability data

Green flags supporting adoption:

  • Prospective validation in deployment setting on public health tasks (not just clinical medicine benchmarks)
  • Offline/edge deployment capability for low-connectivity environments (SMS + backend LLM architecture like MomConnect)
  • Human-in-the-loop design with clear escalation pathways to public health professionals
  • Open-weight model with local customization and fine-tuning potential for local disease patterns and languages
  • Integration with existing public health information systems (DHIS2, OpenMRS, national surveillance platforms)
  • Health department or ministry of health ownership with budget commitment (not donor-dependent pilot)
  • Explicit environmental sustainability assessment appropriate for climate-health context

The Path Forward: Co-Development for Public Health

The extractive model (what to avoid):

  1. HIC institution develops LLM on Western data and disease patterns
  2. Pilots system in LMIC public health department with external funding
  3. Publishes papers on “global health AI” without local authorship
  4. Funding ends, system abandoned, no local capacity remains
  5. Public health department returns to manual processes, wasted investment

The co-development model (what to pursue):

  1. Joint problem definition: LMIC and HIC public health partners identify priority use cases together (surveillance gaps, communication barriers, workforce shortages)
  2. Shared data governance: Training data includes LMIC disease patterns and public health contexts, with local ownership and benefit-sharing
  3. Capacity building from inception: Local public health researchers and informaticians co-lead development, not just deployment
  4. Prospective validation: Local public health validation before scale-up (surveillance accuracy, communication effectiveness, triage appropriateness)
  5. Sustainable financing: Health ministry integration and budget commitment, not donor dependency
  6. Knowledge transfer: Local teams can maintain, improve, and adapt systems independently

Public health practitioner role:

Public health professionals in high-income countries evaluating LLM vendors or research partnerships should demand evidence of co-development, equitable partnerships, and sustainable implementation rather than extractive “deploy and abandon” models. Public health practitioners in LMICs should advocate for local ownership, capacity building, and benefit-sharing rather than accepting passive recipient roles in technology transfer.

One Health: AI Across the Human-Animal-Environment Interface

The National One Health Framework to Address Zoonotic Diseases (NOHF-Zoonoses), released in January 2025 by CDC, USDA, and DOI, represents the first coordinated U.S. approach to diseases that spread between animals and humans. AI applications in One Health surveillance address threats that single-sector approaches miss.

Why One Health Matters for Public Health AI

75% of emerging infectious diseases are zoonotic, originating in animals before spilling over to humans. Recent examples:

Pathogen Animal Reservoir Human Impact
SARS-CoV-2 Bats (likely via intermediate host) Pandemic, 7+ million deaths
H5N1 Avian Influenza Poultry, wild birds, dairy cattle Ongoing outbreak, 2024-2025
Mpox Rodents Global outbreak, 2022-present
Ebola Bats Recurring outbreaks in West/Central Africa

Traditional surveillance operates in silos: human health agencies track human cases, veterinary agencies track animal disease, environmental agencies track wildlife. AI can integrate these data streams to detect threats earlier.

The NOHF-Zoonoses Framework (2025-2029)

The framework establishes seven goals, with AI implications across multiple domains:

Goal 5: Surveillance includes: - Enhanced monitoring systems for zoonotic diseases - Data interoperability across human, animal, and environmental sectors - Advanced pathogen detection technologies

Goal 6: Laboratory emphasizes: - Strengthened diagnostic capabilities - Genomic sequencing for pathogen characterization - Data sharing protocols across agencies

AI Applications in One Health

1. Spillover Risk Prediction

AI models can integrate: - Wildlife population data (migration patterns, population density) - Livestock management practices (factory farming density) - Land use changes (deforestation, agricultural expansion) - Climate data (temperature, precipitation affecting vector habitats) - Human encroachment patterns

Demonstrated: Retrospective models have identified risk factors for known spillover events. Theoretical: Prospective prediction of novel spillover events remains unvalidated.

2. Genomic Surveillance Integration

Combining genomic data from human and animal isolates improves outbreak understanding:

  • SARS-CoV-2 surveillance detected deer-to-human transmission through genomic analysis
  • H5N1 sequencing revealed adaptation patterns in dairy cattle
  • Nextstrain and GISAID platforms enable cross-species phylogenetic analysis

See Genomic Surveillance and Pathogen Analysis for technical details on pathogen genomics.

3. Antimicrobial Resistance Tracking

Antimicrobial resistance surveillance must connect human health, animal production, food systems, and environmental sampling. Detecting the same resistance marker in more than one sector does not by itself establish cross-species transmission. Interpretation requires organism attribution, genomic context, exposure data, and appropriate epidemiologic comparison.

The Genomic Surveillance and Pathogen Analysis chapter provides the canonical treatment of resistome analysis, wastewater AMR surveillance, genomic phenotype prediction, forecasting baselines, and AMRnet. One Health programs should use those methods to coordinate population-level surveillance rather than infer patient-level susceptibility or treatment.

Case Study: H5N1 in U.S. Dairy Cattle (2024-2025)

The detection of H5N1 avian influenza in U.S. dairy cattle in 2024 illustrated One Health gaps:

What happened: - Virus detected in dairy herds across multiple states - Genomic evidence showed sustained cow-to-cow transmission - Limited human cases among dairy workers - Questions arose about spillover timing and extent

AI-relevant lessons: 1. Surveillance data across species was not integrated in real-time 2. Genomic data sharing between veterinary and public health agencies was delayed 3. Predictive models for mammalian adaptation were not deployed operationally 4. Worker health surveillance relied on voluntary reporting

For detailed analysis of H5N1 AI applications, see the Biosecurity Handbook.

Implementation Barriers

Data silos: Human health (HHS), animal health (USDA), wildlife (DOI), and environmental (EPA) data systems do not interoperate. The U.S. One Health Coordination Unit (launched January 2024) aims to address this, but technical integration remains incomplete.

Jurisdictional complexity: Federal agencies, state agencies, tribes, and international partners all hold relevant data. No single entity has authority to mandate integration.

Privacy asymmetry: Human health data faces HIPAA and 42 CFR Part 2 restrictions. Animal and environmental data generally do not. AI systems must navigate different data governance regimes.

Resource gaps: One Health surveillance requires expertise across disciplines (epidemiology, veterinary medicine, ecology, data science). Few jurisdictions have this capacity.

Key Resources